{"guidance_version":"1.20.26","early_access":{"commitments":["LOCK-IN: an account CREATED before early access ends keeps the PER-CALL prices published at its creation for twelve months after general availability.","INCREASE: prices will increase when early access ends — no amount, no date."],"general_availability":"the announced end of early access — announced by a consumer note and a guidance bump; no date is published before then","ga_announced":null,"lock_in_months":12,"scope":"the per-call price table (pricing.json `prices`) ONLY — the free-tier floor, the signup grant, the case-law search caps and the per-call limits are excluded and may change; the lock-in belongs to the ACCOUNT (its creation date), never to a key; 'keep these prices' means the table published at the account's creation, never a lower table published later","mechanism":"since 1.20.26 the pricing config carries versioned per-call tables (pricing.price_tables, each with an effective date). An account created before general availability (every account while ga_announced is null) is charged by the latest table effective on or before its creation date while its lock-in holds — open-ended while ga_announced is null, then until locked_until = general availability + twelve months (inclusive); an account created on or after general availability, or whose lock-in has passed, is charged by the table in force that day. A reprice is a NEW table with a future effective date plus a consumer note; an early-access account within its lock-in keeps its table, so while general availability is not announced a reprice changes the prices of accounts created on or after its effective date only (an account without a lock-in pays the table in force each day). The table is resolved once per request from the account's creation date — every admitted principal has an account row, created at first sight, so every principal (ecosystem and service principals included) has a creation date and a table. An account's creation date is the UTC day the service FIRST SAW the principal (its first API call, or the console login that reads GET /v1/account) — not the day of signup with the identity provider; an account deleted and recreated starts a new creation date and loses an earlier lock-in; a locked account keeps its creation table even where a later table is lower (founder decisions 2026-11-05). Every date is a UTC calendar day: an account's creation date is the UTC date of its creation, a table starts at 00:00 UTC on its effective date, and locked_until is the last UTC day of the lock-in","locked_until":"null until general availability is announced, and null for an account that is not early-access (created on or after general availability); otherwise the last UTC day the account's table holds","where_to_read":"GET /v1/account (verdict_account) data.pricing {price_table, early_access_account, locked_until, current_table, note}; meter.price_table on every charged envelope (the table that charge was priced by — a replay states the original charge's); the manifest's pricing.price_tables / pricing.current_table and pricing.json price_tables + early_access"},"pricing":{"price_tables":["2026-07-28"],"current_table":"2026-07-28","rule":"the manifest's pricing.prices is TODAY's table — what a standard account created today is charged; an early-access account is charged by the table in force at its creation while its lock-in holds (early_access.mechanism); each charge names its table in meter.price_table; the class multiplier (zero-rated ecosystem principals) applies after the table; effective dates are UTC calendar days — a table starts at 00:00 UTC on its effective date"},"rate_limits":{"model":"token bucket per principal, shared across all replicas, refilled continuously","funded_rpm":30,"free_rpm":5,"funded_threshold_credits":501,"batch_weight":{"rule":"POST /v1/cases/batch consumes ~1 token per 50 requested ids (minimum 1); a full 500-id batch costs 10 tokens","note":"a batch larger than your per-minute capacity is granted when your bucket is FULL (burst-through) — it never becomes a permanent 429"},"on_429":"honor the Retry-After header (also in the envelope as error.retry_after) and you will never see a second 429; weighted refusals make waits of 20-60s normal, so fixed 1s-backoff loops will burn their retry budget","anti_pattern":"a serial tight loop of GETs trips the limiter even at concurrency 1; use the batch endpoint for list workloads"},"absence_semantics":{"rule":"a NON-200 RESPONSE IS NOT ABSENCE. Only an explicit found:false entry (batch) or a case-not-found error envelope is a coverage statement. A 429/5xx recorded as 'not found' under-reported real coverage by ~60 points in a measured consumer incident","coverage_statement":"absence means 'not in the Verdict corpus' (post-release cluster or outside coverage) — NEVER a claim that the case does not exist","beta_blocks":"cited_for / authorship / disposition / treatment signals: an empty block means unrecorded-or-unrecognized, never 'none exists'; absence of a treatment signal is never clearance","liveness_control":"to prove the accusation lane is LIVE, cite a REAL volume with an impossible page — '410 U.S. 9187' returns not_found_in_covered_volume. An out-of-range volume ('923 U.S. 4471') is the wrong control: U.S. Reports ends at 572, so it reads as a coverage gap and gaps never accuse, which is correct and looks identical to a dead gate. A partner spent ten minutes believing our gate was off on exactly that control (1.19.4). 'Clean' and 'not checked' are indistinguishable without a control that must accuse","statute_renumbered":"since 1.20.10 (2026-09-28): a U.S.C. section number the current code lacks but that a current section's OLRC source credit names as its FORMER number ('§ 30508; renumbered § 30526 … Pub. L. 117-263, 2022' — title 46 chapter 305) is verified_statute_exists with status 'transferred' and successor[] naming the new section; before 1.20.10 it read statute_section_not_found (the disposition tables carry whole-title recodifications, not in-title renumberings). THIRD SOURCE (same deploy): the OLRC's own PRIOR PROVISIONS notes ('Prior sections 3076 to 3080 were repealed by Pub. L. 103-337 …' under 10 U.S.C. § 3072; 'Prior sections 583 to 590 were renumbered sections 573 to 580 of this title, respectively'; a chapter-level 'Sections 4701-4776 were repealed' under IRC chapter 39) name thousands of former numbers no table or credit carried, which read statute_section_not_found before 1.20.10 — a one-signal accusation of real historical citations (the count is a property of the build, read /healthz derived_tables). Now verified_statute_exists with status 'repealed' / 'omitted' / 'transferred' (successors named when the note pairs them) / 'former' (the note names the number but states no disposition the reader parses — the sentence is quoted) and the evidence quoting the note under its section, chapter or title. A number no table, credit or note names stays statute_section_not_found; the two ENDS of a LETTERED prior range are carried, its interior only when the note details a member by its own sentence; an interior member of a printed numeric range the note never prints is served with the group-membership wording (UNCONFIRMED) unless its paired successor is printed by the current code, and is not found when the title's disposition table records that renumbering for its neighbours and omits it; a number the note names without a disposition the reader parses serves status 'former' with the sentence quoted; a successor is advised only while live, not repealed after its transfer, and not enacted after the note's own dates (a reused number is said as such). A title whose OLRC sources failed to build on a machine serves statute_title_not_covered on the absent path (a gap; /healthz usc_recodified_failed names it). NAMED WITHOUT A DISPOSITION (same deploy): a number the OLRC names as a section of this title without stating what became of it ('was classified to section 5432 of this title') and that the current code lacks is manual_review_required (status 'unverified' — a former section or a stale cross-reference; review, never the fabrication label), counted in the new summary field statutes_review, in the document status like an accusation, and as a verdict_check_brief flag (a statute title gap is a coverage.gaps entry and a brief gap and counts in coverage.completeness); a U.S.C. RANGE whose endpoint is such a review row is manual_review_required, never 'endpoint does not exist'. PLACEHOLDERS: a multi-section placeholder the Code prints ('§§ 6913, 6914. Repealed.') is indexed under every number it names on a machine whose section index has been re-ingested (7 U.S.C. § 6913 is verified_statute_exists / repealed; before, 593 such numbers were accused). CODIFICATION NOTES: the OLRC's codification and removal notes are read for their stated dispositions ('Prior sections 3081 to 3086 were repealed …' — 20 U.S.C. § 3086; before, ~210 such numbers were accused). DISPOSITION TABLES (same deploy): the OLRC's tables are read wherever the OLRC placed them — title 46's 1,347-row table sits in an amendments note (46 U.S.C. § 688, the Jones Act, is recodified as § 30104; before 1.20.10 it and old titles 10 and 34, the 1978 bankruptcy Table I, titles 17 and 51 read statute_section_not_found). WORDED U.S.C. RANGES (same deploy): '§§ 30505 through 30508', '§§ 1 through 1400Z-2', a list member's '§§ 1981, 1983 through 1999' are read as the range under a plural mark with the end after the start (a single '§' + 'to' is prose: 'Compare § 1983 to 42 U.S.C. § 1985'), and raw is served AS WRITTEN ('42 U.S.C. §§ 1983 through 1999' — before, the end was dropped); a dash inside a section number is read only before a digit ('§ 1983—the statute' is § 1983).","statute_snapshot":"since 1.20.18 (2026-10-17): the U.S. Code tables are a SNAPSHOT of one OLRC release point, and every statute_section_not_found evidence names it ('22 U.S.C. § 1234: title 22 at OLRC release 119-102 (current through Pub. L. 119-102) is fully indexed and has no section 1234') — the edition an absence rests on. A section absent from a covered title whose number lies BEYOND the title's last numbered section at that release (the leading number compared: '10802', '10802a' and '10802-1' all read 10802) is the NEW GAP label statute_section_beyond_snapshot (label_class gap, status 'unverified', no hallucination category, resolution.release names the stamp): a section enacted after the release would not appear in the snapshot, so it is not examined — a coverage gap, never a fabrication ('22 U.S.C. § 10802', enacted 2026-09-18, read statute_section_not_found against title 22 ending at § 10710 before 1.20.18). We cannot tell a later enactment from an invented number past the end; the gap is honest either way — check such a section against the current Code yourself. A RANGE with an end past the snapshot and no end absent inside it is the same gap; an absent START inside the range still accuses, the far end said. A section absent INSIDE the snapshot's range stays statute_section_not_found (after every OLRC table, credit and note reader); a NEW section inserted inside an existing range after the release ('§ 10709a' beside § 10709 and § 10710) is still accused until the tables are refreshed — read the release named in the evidence. The label counts in data.summary.statutes_uncovered and coverage.gaps like statute_title_not_covered, and a document whose only law reading is this gap reads 'unverified', never 'failed' and never an accusation","us_nominative":"since 1.20.10 (2026-09-29): the nominative reporters of the United States Reports — Dallas 1–4 (= U.S. 1–4), Cranch 1–9 (U.S. 5–13), Wheaton 1–12 (U.S. 14–25), Peters 1–16 (U.S. 26–41), Howard 1–24 (U.S. 42–65), Black 1–2 (U.S. 66–67), Wallace 1–23 (U.S. 68–90) — resolve through the U.S. Reports reading by a fixed table ('4 Wheat. 316' = 17 U.S. 316, '1 Cranch 137' = Marbury v. Madison), the Bluebook form '5 U.S. (1 Cranch) 137' resolves as 5 U.S. 137, and a founding-era cite whose official page lies inside the primary corpus's span ('4 Dall. 37', Bas v. Tingy) is answered by the secondary court-records corpus through the U.S. form; the evidence names the nominative form. Another book at the same numbers (the D.C. Circuit's Cranch reports; a circuit case of Dallas reprinted in Federal Cases) is named in the evidence, and the WRITER'S CAPTION AND YEAR decide: when they disagree with the U.S. reading and match the other book's case, that case is served ('Dawes v. Corcoran, 1 Cranch 137 (C.C.D.C. 1803)'); when they match neither, the citation is manual_review_required naming both (a procedural caption — 'Ex parte Jones', 'In re Estate of Smith' — is read as a caption too) — a verification resting on the nominative convention never produces case_name_mismatch or citation_year_mismatch. Where the U.S. page starts no case the other book's reading is review ('1 Cranch 448'). U.S. volumes 1–90 carry the years of their nominative reporter: a secondary-corpus record outside them ('5 U.S. 180' keyed to an 1866 case) is two readings, never the cited case; a secondary record over a differently named containing case verifies only when the writer's caption matches it. verdict_get_case rides the same guarded lookup. Before 1.20.10 a nominative cite read verified, review or a coverage gap by the corpus's parallel rows alone.","cfr_written_year":"since 1.20.10 (2026-09-29): a regulation cited with its own edition year before eCFR's coverage ('24 C.F.R. § 282.4 (1949)') that is not in the current edition is manual_review_required — an edition before 2017-01-03 cannot be checked against eCFR; without a written year the same miss keeps its label. A run of letters glued to a C.F.R. token that is not a word of the sentence (‘§ 240.11Acl-l’, ‘§§ 1604.11-1604.12Acl-l’, ‘Part 1604Acl-l’) makes the unit manual_review_required — on a head, a range end, a list member or a part alike; glued to a paragraph pin (‘§ 1604.11(a)Acl-l’) it is unread and said in the evidence, never dropped; a glued word (‘§ 240.10b-5and’, ‘§ 1.1et seq.’) is the sentence’s own and the unit keeps its reading. A worded part member after a unit (‘§ 1910.1200 and part 1926, subpart Z’) or a subpart written after a part-level member (‘part 1926, subpart ZZ’) that no unit read is named in law_notes — cite it on its own to have it checked; a supplement or official comment written after a part or section (‘pt. 1026, supp. I, cmt. 18(d)-1’, ‘§ 1026.18 cmt. 1’) is named the same way.","statute_single_mark_dash":"since 1.20.10 (2026-09-29): a U.S.C. section written with a SINGLE '§' and a one-digit dash end that this title lacks as a hyphenated section, where the compressed reading verifies ('28 U.S.C. § 2201-2' — §§ 2201 and 2202; '35 U.S.C. § 112-6' — the patent bar's § 112 ¶ 6), is manual_review_required naming both readings (the range under a '§§' mark; paragraph M of § N) — never a verified range the writer did not mark and never statute_section_not_found. Where NO compressed range verifies ('35 U.S.C. § 112-2', '42 U.S.C. § 1985-3', '5 U.S.C. § 706-2', '28 U.S.C. § 2201-9') the string is still manual_review_required when the current code holds the section before the dash and the title numbers no hyphenated section under it — two readings named: paragraph or subsection M of § N (the dash form of '§ 112 ¶ 2', '§ 1985(3)'), or a hyphenated section the code does not hold (founder decision 2026-10-01: one signal never accuses). It stays statute_section_not_found when the section before the dash is not in the current code ('42 U.S.C. § 2000-5'), when the title numbers hyphenated sections under it and lacks this one ('16 U.S.C. § 47-7' beside §§ 47-1 to 47-6), and when the end has two digits, a letter or a zero. Under a PLURAL mark the same string is the range ('§§ 2201-2' = §§ 2201–2202).","statute_written_letters":"since 1.20.10 (2026-09-29): a letter written after a U.S.C. section number is never dropped silently. A letter after a dash ('§ 1983-b', '§ 405-g provides') names the lettered section: when it exists the citation is manual_review_required naming both readings, when it does not the citation is manual_review_required ('names no section') — only a lowercase English word of a closed class ('a', 'an', 'in' …) before a lowercase word is read as the sentence's own word, and the evidence says so ('§ 1983–a statute that gives a remedy' is § 1983; when a second occurrence folds into the first row, verification.notes says so). A dashed pair whose halves are one section written twice or two sections of one family ('§§ 7661-7661', '§§ 7661a-7661'), where those sections exist, and a number followed directly by a letter outside the ASCII alphabet ('§ 7661Í', a scan's reading of a suffix) are manual_review_required with the letter named, never statute_section_not_found and never a verified head; a pair no reading of which names a section ('§§ 9999-9999') keeps the accusation. A join word glued to the number ('§ 1981and 42 U.S.C. § 2000e', '§ 1983and the Fourteenth Amendment', '§ 2000e-2and') is the join word, and when the number alone names no section ('§ 78or 78p') the citation is manual_review_required; the end after a glued range word ('§§ 1983to 1985') and a range end that goes on past its token ('§§ 1396 through 1396-v') are said in verification.notes, never read silently; a run of dashes is one dash ('§§ 1981--1983' is the range); an ordinal ('26 U.S.C.A. 1st') is no section; a short list member after a longer one ('§§ 151, 52, 58') is read by Bluebook compression (§§ 152, 158) only when the literal section is absent or repealed / omitted / transferred — a live literal ('18 U.S.C. §§ 1341, 2' names § 2) is served with the compressed reading noted; a section number no title holds whose scan-sibling exists ('§ 8251' for § 825l, '§ 19731' for § 1973l, '§ 260S' for § 2605) is a two-readings review, never an accusation; a short number carrying a pin after a pinned member ('§§ 30101(8)(B)(ix), 9(B)(viii)') is that section's next paragraph (said, never § 9), a section-shaped member after a single '§' mark ('§ 78j(b), 78t(a)') is said in verification.notes, a U.S. Code paragraph-pin note names the U.S. Code text (never eCFR) as its source, a C.F.R. member written as a bare short number ('§§ 4043.21, 22, 23.', '§§ 404.1520, 27', '§§ 404.1520, 1527') is the previous member's section with its last digits replaced, at a sentence end too — never a number a capitalised word follows (the next citation's volume: '44 Fed. Reg.', '59 S. Ct.'), and a bare number longer than the member's tail ('§§ 4043.21, 2001') is said with its own reason; a bare member carries its pin ('§§ 404.1520(a), 27(d)'); every listed paragraph or paragraph range written after a pin ('§ 1630.2(j), (k), (zz)', '(l)-(zz)') is said as not range-checked; a C.F.R. member after a single '§' ('§ 404.1520, 404.1527') is said, never silent, quoting the token as written beside its compressed reading, and a ';' bare member reads as a ',' one; a listed paragraph after a range pin ('§ 1630.2(a)-(c), (e)') is said; a glyph before a section-shaped continuation ('§ 78»-2(c)(3)', '§ 78» (a)', '§ 78~(b)') is unreadable (review), never the digits before it accused, and a stray mark with nothing after it is named, while a mark after a number whose head does not exist ('§ 198^') is two readings (review); a code-edition year ('§ 309© (1994)') never continues a number; a paragraph range's end ('§ 2000e-2(a)-(z)', '(a)(1)-(2)(A)' = (a)(2)(A), '(e)(2)(B)(i)-(iii)' = (e)(2)(B)(iii)) is a pin of its own aligned by the subdivision ladder — an end that cannot be placed is said as unread, never checked at another level; a slash between section numbers ('§§ 1983/1985', '§§ 1983/ 1985') joins the list — a slash with no digit after it is a mark; a scanned 'l' or 'I' takes its ladder level's reading for every reader — the path as written is checked first, then the scanned reading ('(a)(l)' = (a)(1), '(1)(l)' = (1)(I), the C.F.R. reader alike), the scan named; a section letter followed by an unmatched ')' ('§ 1401Q)') is a damaged paragraph pin, two readings; a C.F.R. section the edition lacks whose scan-sibling reading exists ('§ 1910.120O', '§ 240.10b-S') is two readings, and a token the reader could not take whole ('§ 240.11Acl-l') never accuses its prefix (review when no occurrence of the unit is written cleanly; when another occurrence is, the unit keeps its reading and its evidence names the damaged form as the text writes it — the same row whatever the order of the occurrences; every list member written after letters glued to a unit — with or without '§', pinned, or a range — is said as unread); and '§ 240.10(b)-(5)' reads the parenthesised dashed sub-number like '(b)-5'; a list member written with its stem elided ('§§ 2000e-2(a), -3(a)', under a single '§' too) takes the previous member's stem (2000e-3) and is checked, with its elided or full range end ('-3(a) through -5' / '-3(a) through 2000e-5' are §§ 2000e-3 through 2000e-5; a reversed end, elided or dash-led after a hyphenated start, is said as reversed, never as 'no dashed family'); after a member with no hyphenated stem ('§§ 1983, -5') it is said in verification.notes, never dropped. The same join-word rule holds for C.F.R. sections ('29 C.F.R. § 1910.1200and the standard' is § 1910.1200; no segment of either token ever ends in a join word — '§ 1320a-7band', '26 C.F.R. § 1.401(k)-1and', a range's end '§§ 1981 through 1983and', a list member, 'Parts 1910and 1926'; a glued range word reads the range: '§§ 1983to 1985' is §§ 1983–1985). A dash run is one dash wherever a dash is read ('§§ 2000e–2–5' and '§§ 2000e-2--5' carry the plural mark). A spaced letter before a pin or punctuation is the section's letter ('§ 78 j(b)' reads § 78j as the second reading; '17 C.F.R. 240.10 b-5' is § 240.10b-5). The 1939-Code rule reads the leading integer ('26 U.S.C. § 1152a' is review, never accused). A section number is a number, one letter repeated ('300aa', '1395ww') and dashed sub-numbers of the same shape — the Code's own alphabet: a string outside it that the title lacks ('§ 1983of', '§ 78ec', '§ 2000bb4', '§ 460ec', '§ 9999xyz') is manual_review_required naming the reading (a scan's letter, a glued word, a lost dash — '§ 2000bb-4 exists') and never statute_section_not_found; a run that starts with a join word is the sentence's ('§ 1983andthe rules' is § 1983), a repeated letter before a join word is the letter ('§ 1320a-7band'). Every dash character — hyphen, en and em dash, Word's non-breaking hyphen, the figure dash, the minus sign — is a dash in BOTH lanes ('§ 2000e‑99' is accused, '§§ 1981‑1983' is the range, '48 C.F.R. § 52.212‒4' verifies; before 1.20.10 a non-breaking hyphen ended the token silently). A symbol, a mark or a superscript after the number ('§ 630©', '§ 1983¹') is named in the evidence and the section stands; a LETTER outside ASCII ('§ 7661Í') is review. '&' joins a list ('§§ 922(g) & 924(a)(2)'). A spaced letter run before punctuation ('§ 441 j.') is two readings only when the lettered section exists; before a pin ('§ 78 j(b)', '§ 1983 a(b)') it is the section's letter (review when it names a section, review 'names no section' when it does not); a citation abbreviation or an English word there ('§ 1601 et seq.', '§ 101 ff.', '§ 1983 or,') is the sentence's. In the C.F.R. a letter run written after a space ('§ 404.1520 a(4)') is manual_review_required when the joined section exists, the section before the space with the letters said when it does not, and review when neither exists ('§ 1.91 la(c)', a scan); a small letter directly after a dash ('§ 2560.503-l(f)') is a scan's digit (review); a letter in parentheses whose joined form exists ('§ 240.10(b)(5)' for § 240.10b-5) is two readings; a letter run glued to the token whose head exists ('§ 1910.1200of', '§ 1910.1200OSHA', a range's end '§§ 1910.1000 through 1910.1200of') is two readings; a capital before two small letters ('§ 1910.1200The') is a glued word, said, and the section stands; a spaced capital run before a pin or punctuation ('§ 3.12 Mr.', '§ 1607.4 D,') is prose. The soft hyphen (U+00AD) is a dash in both lanes, and a RUN of dash characters is one dash ('§ 240.10b­-5', '§ 102­-74.265', 'Part 102--74' read as written; a hyphenated part, a subpart id and a part-list member take any dash character too). A relator caption ('United States ex rel. Burnett v. Teller') matches a record naming the sovereign or the relator. A citation whose reporter AS WRITTEN is shared with another sovereign ('Den.', 'Doug.', 'Cox', 'Add.', 'Rob.', 'Johns.', 'Litt.', 'Hard.', 'Allen', 'Phil.', 'Freem. Ch.', 'Mont.', 'Grant', 'A.L.R.', 'Ind. App.' — a closed table read against the corpus), written in its OWN group (after its caption, before the next caption or its year parenthetical) with a reporter of another sovereign, or closed by a court parenthetical of another sovereign ('1 Den. 329, 331, 169 Eng. Rep. 267', '2 Doug. 689 (K.B. 1773)'), is manual_review_required on every label — an abbreviation collision (Denison's Crown Cases, Douglas's King's Bench, not Denio or Michigan's Douglass), never a name or year accusation and never a fabrication claim; a neighbouring American citation, a '(citing 169 Eng. Rep. 267)' parenthetical, '(Ch. Div. 1952)' and '(Ch. 1944)' (New Jersey's and Delaware's Chancery) and '61 A.C. 94' (the Advance California Reports) are untouched — a reporter only an American court prints is never demoted. In the attribution lane: an acronym whose letters walk the record's words ('ISKCON' = I·S·K·CON, 'UAW') and a written word that extends the record's word ('Waterworks' / 'Water', 'Johns' / 'Johnson') are case_name_variant ADVISORIES, never a clean pass and never a mismatch ('FDA' for the Federal Trade Commission still accuses); a written name that ends inside the record's word ('Robert' / 'Roberts', 'Williams' / 'Williamson') is a variant too, never a clean pass — only a possessive mark (\"Symes's\" / 'Symes') or the record's whole words written without their spaces ('DuPont' / 'Du Pont de Nemours', 'Airlines' / 'Air Lines', 'School Comm.' / 'School Committee' — anchored on an equal word, never by an abbreviation alone: 'Stevens' is not \"St. Mary's\") is the same name; a caption the extractor cut at an abbreviation ('Carter/Mondale Pres. Comm. v. FEC' read as 'Comm. v. FEC') carries the caption run before it as a SECOND reading that can only turn a failure into a pass by spelling the record's first words (a sentence ending in a capitalised word never changes the class); a caption cut at a comma is a fragment, never evidence; an unmarked abbreviation opening the record's word ('Imp Co.' / 'Improvement Co.') is a variant; a one-sided record ('In re Rosenfield') is named by either written side; a period-marked token opening a stopword ('R.R. Comm.') is that stopword; every Unicode space glyph reads as a space inside a citation ('347 U.S. 483' is checked like '347 U.S. 483'), and a year-shaped page that every source refused, beyond the volume's last page, with no caption, pin, year or parallel leg around it AND the artefact's own signature (a no-break or other space glyph inside the string, or a date word before it — a month or 'dated', never a bare preposition: 'Jan. 12 Cal. 2020') is a date, said in the notes and never served (a short form that may name it is unchecked; a plain 'See 99 Cal. App. 4th 1650.' stays not_found); a caption the extractor mangled ('City of Keller', \"Nat'l Ass'n\") is read AS WRITTEN; a government body — [its sovereign]? + department / office / bureau / commission / board … + 'of <Name>', or a title ('Attorney General') — written where the record names its bare sovereign (\"N.J. Dep't of Envtl. Prot.\" / 'State') is compatible when its named sovereign is ('Office Depot' is no body); the Oklahoma Statutes read 'O. S.' with a space too; a lettered compressed end keeps its letter ('§§ 670-79a' = §§ 670–679a, '§§ 7661-61a' = §§ 7661–7661a, '§§ 7661a-61f' = §§ 7661a–7661f, '§§ 1462a-1464', '§§ 1396a-97b' and the reversed '§§ 1395a-1394' are ranges, a family's sub-number range keeps its letters ('§§ 1320a-7a-7b'), and a former or omitted single the OLRC names ('§ 300ff-75a') is served before any range reading of it); an invisible glyph (a zero-width space, a BOM) beside a citation reads as a space in every lane, and one INSIDE a number ('347 U.S. 48​3', '§ 19​83', a run of them, a direction mark — every Unicode format character) makes the token unreadable — manual_review_required naming the glyph, never a verification of the digits before it; a pin such a glyph splits is unchecked; a glyph inside an abbreviation ('U​.S.') leaves the citation unread and the notes say how many such characters the text carries; a glyph-split writing of a section is its own review entry beside the clean writings; a subpart whose sections eCFR lists as part-level siblings (29 C.F.R. Part 1910, Subpart Z) is counted by them in document order; a FAR subpart written without its part ('48 C.F.R. Subpart 1.1') and a list of subparts ('Subparts A and Q': only the first member is read) are said in the notes; a numeric subpart before a section ('Part 15, Subpart 15.4, § 15.499') reads the section; a pin is the page written AFTER the citation ('375, 378-79'), followed by citation furniture, never a number before it ('Local No. 1189', '5 Cir. 1966'), a parallel leg's volume ('483, 74 S. Ct. 686', '483, 1954 WL 123') or the case's own year written after the comma ('483, 1954.'); a short name ('Roe', 'Cox', 'Fay') that is the whole name of a side and equals a record word is identity at any length (a particle of a longer name — 'New' of 'New Jersey' — never is), a caption role word ('Appellant', 'Petitioner') never is, the party as written is always a reading, and a role phrase is role vocabulary ('Trustee in Bankruptcy', 'Individually and as Executor') — any other comma name with a lowercase run ('Rabouin, dba Conway's Express') is the variant tier, never an accusation; a stripped reading goes through the same side test, the sovereign relation included ('Accord, United States v. Terry' is the sovereign review); one edit at four or more letters — an insertion, deletion, substitution or adjacent swap, the stopword and sovereign words included ('Wadee', 'Ohoi', 'United Staets', 'Eli Lily') — is case_name_variant; a number followed by a stray period and a reporter, spaced or not ('64, 791. P.2d 266', '477, 101. S. Ct. 1880'), is the next leg's volume, never a pin; a page INSIDE a case cited with that case's own caption before it ('Holmes v. SIPC, supra, 503 U.S. 271') is the case's pin (served as the case at its first page, so later pins are checked; a short name before 'supra,' counts; a page written WITH the writer's own pin or year is a first-page slip → review; a pin marker without the caption → review), never a first-page fabrication; a page after a subsequent-history phrase naming an ORDER — the verb and its glue ('cert. denied sub nom. X v. Y,', 'cert. den., 1970,', 'certiorari denied (1982),', 'reh. den.', 'We granted certiorari,') — is a coverage gap (the glue admits a renamed caption with or without 'sub nom.', the state courts' orders — 'appeal denied', 'review denied', 'lv. denied', 'transfer denied' — 'by Sup. Ct.', a parenthetical and the earlier legs of a parallel string, and stops at a sentence end), while one after a reported-decision verb ('aff'd,', 'rev'd,') is checked as written; a page PAST the volume's last indexed record (read from the records filed in that volume's own book — a parallel citation of a case filed in another book marks that one case's start, never coverage of the pages before it; the evidence says so when such rows lie past the record), in any request form, is an entry at the volume's end the corpus does not ingest (order lists, memorandum dispositions, tables of dispositions, court rules) or a mis-typed page — a gap whose evidence never says 'fully indexed' — unless the page lies far beyond every neighbouring volume of the reporter (more than two and a half times their largest page: the fabrication shape, which keeps the accusation); a narrated denial or grant of leave / allocatur / certification / review, or a parenthetical that is the table word followed only by disposition furniture, or '(mem.)' ('(table)', '(Table of Cases affirmed)', '(unpublished table decision)' — never '(mem. op.)', '(see Table 3)' or '(Table 2 of the regulation …)') in the citation's own trailing parenthetical chain (after its pin and a WL leg), is an order signal mid-volume that never overrides a page read inside a case (the narration's own sentence may hold any words but no caption — a caption in it is a new citation; after the sentence end only the order's caption and legs follow; 'Certification was denied.' and 'denied transfer' are narrations); a pin on an order-list entry (a one-page or derived-extent record, the signal in the citation's own sentence) is unchecked; a narration never crosses a sentence end but reaches the citation's own sentence and the one before it whatever their length (quotations folded); an author-title lead (never a procedural caption) before an F.R.D. page is a gap (F.R.D. prints articles and rules); a written abbreviation that is a corpus book's own name or a listed alias ('Tex.App.': Tex. Ct. App. and Tex. Civ. App.; 'Conn.Sup.': Conn. Supp.) resolves to the book holding the cite, the caption deciding among several and a bare cite two books hold reading as two readings — never a book whose name merely contains the written tokens; a written year decades before a shared reporter's era ('2 Doug. 655 (1781)') is a foreign-reporter signal; the writer's placeholder 'Same v. Same' is no caption; a pin past a record's last page that no indexed record of the volume holds before the next record starts lies in an un-digitized hole and is unchecked with the hole named (Bateman Eichler's pin 311), in the full-cite and short-form lanes alike, always within the served record's own book (a nominative's other-book rows never end a case); a pin below a leg's first page that lies inside a sibling leg of the same citation is that leg's pin (said, unchecked here); a caption naming the record that starts at the written pin or a pin range's start ('Robbins v. Shelby Taxing District, 120 U.S. 487, 489-90') is a first-page slip → review; a one-letter variant of a name ('Willcox' / 'Wilcox') names the containing case; 'to use of' / 'for the use of' is a designator like 'ex rel.'; a rescued nominative row is the antecedent of the short form after it only when no other citation of any class stands between; a pin LOWER than the first page of the report it is compared with is never checked against that report — it may be a parallel report's page, a slip, a footnote or paragraph number, or an Id. of another document or case (unchecked and said; the record holding it, if the caption names it, is a first-page slip → review); a caption every side of which meets the record as a variant (one of them a stem variant such as 'Railway' / 'Railroad', anchored by a written year or a parallel leg of the same case) is a variant; the caption after 'aff'd sub nom.' or 'affirmed,' names the previous citation's case too; a dismissed writ or appeal and 'as improvidently granted' are orders (never across a sentence end); a granted or denied petition or application needs no 'for' only in the reviewing court's own voice ('We granted both applications', 'this Court denied the petition') and never before a decision verb in the same sentence ('denied the petition, and we affirmed' cites the decision, checked as written); the record holding the written page or pin that the caption names (either order) is a first-page slip → review even where the page verifies another case (a mismatch) or lies inside one (a not_found) — the holder is a native record of the served book (a nominative's U.S. volume; a parallel book's pages carried by another record never hold the written page, so that mismatch stands); every citation of any class inside the previous citation's own explanatory parenthetical ('(citing A; B)', '(quoting X, 5 U.S. at 177)', '(quoting X, supra, at 177)') is a second reading of the 'Id.' after it, never its sole antecedent, while a parenthetical opened after a sentence end (a period after a lowercase word, a digit or a closer — never after an abbreviation, and never inside a closed court parenthetical) belongs to the new sentence and the citation inside it is the antecedent; under an unresolved main citation no nested authority becomes the antecedent; the hole note names every leg's reading and is written only when the short form is left unchecked; the short-form walk reads every reference the extractor does not — a record, brief, transcript, docket entry, legislative source or treatise locator ('J.A. 900', 'Tr. 900', 'A.R. at 1316', 'Pet. App. 900a', 'ECF No. 12, at 900', 'H.R. Rep. No. 94-1476, at 900', '124 Cong. Rec. 34780', '2 Wigmore, Evidence 900') — as an event in text order: an 'Id.' after one is bound to no case (unchecked and said), unless the reference stands inside the main citation's own parenthetical; an 'Id.' whose own parallel leg, written 'at P' or bare, names another citation ('id. at 423, 104 S.Ct. 3141' after Bruder) is bound to no case; an 'id.' after a volume number or a history verb ('43 id. 554', 'aff'd, id. 504') names the reporter volume, never the case; a number written right after an Id., or after the 'at' or page word of a supra reference, that a reporter follows ('Id. at 424 U.S. at 348-49', 'id. at 462 U.S. 239', 'Morris, supra, at 612 F.2d, pp. 488-489') is that leg's VOLUME, never a pin — the short form is unchecked and said ('Roe, supra, 410 U.S. at 999' writes its own short citation, which is checked as one); a short form's number that a white-space run of more than 64 characters follows is unchecked and said (what stands after the run is not read); a citation inside a quotation or a colon block ('the court said: …') is nested under the quoted authority when no sentence end separates them, and bound to nothing when one does; an authority introduced by an unparenthesised participle (', citing X', ', quoting X', ', relying on X') is nested like its parenthesised form; the legs of one short-form string share one case ('Manson, supra, 432 U.S. at 116, 92 S.Ct. at 2254' — a glued leg is unresolved when the string's case has no such leg); a pin past a parallel reporter's DERIVED span (the neighbour's start, a coverage estimate) is an unread leg — unchecked with the derivation named, and never the second signal of an accusation ('Powell, 127 U.S. 678, 8 S.Ct. 992, 1257 … Id. at 1260': the S. Ct. reporter printed the opinion in two parts); any locator-bearing text between a citation and an 'Id.' ('Joint Appendix 900', “Black's Law Dictionary 1500”, 'Record on Appeal 545', 'The record at 545', 'Dkt. #12 at 500') makes the 'Id.' that reference's — unchecked and said, never a case's; an 'id.' right after another case's caption ('Railroad Co. v. Alabama, id. 832') names the book; a footnote call, a page word or a year before 'id.' is no volume; the legs of a subsequent-history chain ('291 B.R. 727, aff'd, 439 F.3d 248. Id. at 739') share the 'Id.', which accuses only when the pin fits none — a chain never crosses a sentence end ('The judgment was reversed. See Roe …' is prose); a short form's event ends after its pin as written, footnote tail included ('Id. at 490 n. 5'), so the pin is never read as a reference by the next 'Id.'; a nameless short form ('487 U.S. at 677') is ambiguous when the document cites two cases of that volume anywhere, and its own parallel leg naming another case ('487 U.S. at 677, 108 S.Ct. 2597' where neither the group nor the antecedent's own record carries that leg) leaves it unresolved and said; a short form inside a quotation or colon block that opened before the previous event is enclosed like one that opened after it (a short form closing a block quotation never becomes the antecedent of the 'Id.' after the quotation); a colon block runs to a blank line (an unmarked block quotation's end is the layout lane's — the Id. after one is a group reading, never an accusation against the last quoted case); when an 'Id.' binds a group of several cases the evidence names every case tested and the label targets the nearest; a second report of the same case at the pin page anywhere in the SAME book with the same docket ('Morgan v. Commissioner, 309 U.S. 78, 626') is never out of range; the same-decision reading needs the same record NAME, and the same-day companion needs full dates, one court and a shared party word, a 'Cases' caption or one docket, or a served report of two pages or fewer (a short opinion may incorporate its companion's reasoning — a review that names the length); a shared party word is a distinctive one (never a sovereign, a state, an agency, an office or a corporate-form word); the companion note names the kinds served; a nominative pair the table refuses ('5 U.S. 2 Cranch 137') is two readings, never page 2 (the written nominative is quoted); a label-only parenthesised nominative ('1 N.C. (Mart.) 5') is the outer book as written, the label a second reading only when the label's volume is the official volume itself (a record under both '1 Mart.' and '1 N.C.') — never the extractor's edition choice; a parenthesised jurisdiction tag ('5 Cush. (Mass.) 295', '16 Johns. (N. Y.) 438') and a series label on the official book ('59 Mass. (Cush.) 295') are one reading; when both readings resolve to one case the citation is verified unless the caption read names another case or the written year does not fit (two paginations of one book: review); a book token's spelling is folded through the reporter database ('Metc.' → 'Met.') unless it is itself an edition; a qualifier is never dropped — part of the book's name ('4 Wash. (C. C.) 371') when the index holds that book, else a jurisdiction the served record must agree with ('1 Hill (S.C.) 1'); a held record of an order-list page is one of many entries in the secondary corpus too (a run of records of one date on consecutive pages — never two same-day opinions a few pages apart, never a neutral citation's sequence), and in a volume the index holds only through another book's parallel citations (671 N.Y.S.2d: memoranda of the A.D.2d ingest) the region is read from the STARTS — the record held at the page is two pages or fewer in its own book and, on the page or within two pages of it, several such records of ONE court-day (the same ingest, the same decision date) start on one page, or the held record fills one page or less of its own book and another short record of that court-day starts within two pages; two short opinions of other courts or days are two opinions; a caption that names another record holding the cited page is the first-page-slip review in every reporter (the volume's own records are read from the index's data, never from a table spelling: So. 2d, So. 3d, N.Y.2d); a page written with a decimal point ('623 F.2d 1.155', '16 L. Ed. 2d 31.55') is two readings — the leading page is served when the written caption or the citation's own parallel legs name its record, else a review naming both readings, never an accusation, and a gap stays a gap; a citation written with its volume and reporter twice ('311 B.R. 311 B.R. 43') reads the inner citation (a review when that does not verify); a citation written '<vol> <Rep> at <page>' is a short form in every process; an Id. after prose naming another cited case (a hyphenated or possessive short name included, never a generic institutional word, and a hyphenated compound only whole) is unbound, and every locator in the gap before an Id. is read (a year or a month-day is skipped, never a stop); a pin range whose end lies before its start with the end inside the case is a slip of the start (unchecked); a footnote marker after a comma (', n. 1') marks a pin; a short form's pin reads the same companion rule as the full citation's; a vowel-less contraction of a sovereign word ('Cmwlth.') is that word; the writer's own year is a parenthetical that closes with a year, never an explanatory parenthetical; a reversed caption is served only when distinctive party tokens cross on both sides ('Pennsylvania General Casualty Co. v. Penn.' is the same order); a popular name ending in the plural 'Cases' is a popular name whatever its punctuation (“Employers' Liability Cases”); the sovereign of 'State ex rel. X' is the State ('Gast v. Oregon' names 'Gast v. State ex rel. Stevenson'); a nominative pair whose other half names a book the index does not hold ('74 Tenn. (6 Lea) 337') is one reading — the held half's outcome, whatever it is, is served; 'United States of America' is the federal sovereign and the sovereign head of an ex rel. caption ('United States ex rel. Curry') is read even when the extractor keeps only the relator; 'Comm of Pa.' is the Commonwealth with or without the period; a standard agency acronym may take a letter from a function word ('DOI' = Department of the Interior); in a railroad's initials 'R. R.' / 'Ry.' is the one word Railroad / Railway ('B. & O. R. R.' is the Baltimore & Ohio Railroad — the letter before a spaced '&' or a comma belongs to the initialism, so 'C. & O. R. R.' is another road — a side is MET, CONTRADICTED or UNREAD, and case_name_mismatch needs a side that was READ and contradicted: initials that open the record's words in order and only SKIP words ('P. R. R.' for the Union Pacific Railroad, the form courts wrote) are a variant whose evidence names the omitted word — a word no reading of the written side opens, through its abbreviations and the words written before the initials — never a mismatch; initials the reader cannot walk but MOST of whose letters open the record's distinctive words in order ('F. S. S. R. R. Co.' for the Forty-second Street & Grand Street Ferry Railroad, 'S. D. & P. M. R. R.') are UNREAD — beside a side that names the record they are a variant saying so, never a mismatch — while initials that open at most half ('FDA' for the Federal Trade Commission) are another party and the mismatch stands; a party the extractor cut out of a longer name at a connector ('Americans United for | Separation of Church and State') is unread the same way; two NEAR sides are a variant when one is a short form (initials, a contraction, an elision), a name written with or without its last letter on the record's same side ('Shield' / 'Shields'), or the same slip on both sides ('Gullett v. Gullett' for Gullet v. Gullet) — two unrelated one-letter slips stay a mismatch; a short name one letter longer or shorter ('Amey' / 'Amy'), a two-edit scan confusion that keeps its first letter and last three ('Van Butter' / 'Van Blitter'), a scan's ligature and 'Mayor, etc.' for a city are the variant tier; a caption run never crosses the previous citation's closing parenthetical or a court-year slot ('(D.R.I.1973); U. S. v. …', 'Curtis v. U.S., D.C.N.J.1950,'); a letter may open the second element of a compound record word ('C. & N. W. R. Co.' for Chicago & Northwestern Railway, 'St. L. S. W. Ry.', 'U. S. W. of A.'); words whose initials stand in order inside an initialism of the record's side are its worded short form ('Automobile Workers' / 'Auto. Workers' for 'International Union, U. A. W. A.') — a variant; and the party before a representation joint ('Baker by Thomas v. General Motors', 'by and through', 'by his next friend') is the side's other reading — either one standing on the record's side meets it); a contraction whose last letters spell a corporate designator the record's short name dropped ('GEICO' for Government Employees Insurance) is a variant; a written token joined from two ('Barboszewska-Zajac', “Int'lShoe”) is read through the same near and elision readings as one; a year-shaped Oklahoma page with no section after it is the section, never an edition; a written year decades after every held record of the volume, with a federal-court parenthetical on a state reporter ('12 Misc. 32 (S.D.N.Y. … 2012)') — both read from the citation's own closing parenthetical, before the next citation outside its parallel group and before a sentence end, never a neighbour's — is a docket-number-shaped reference (review), while a wrong year alone stays the year lane's; a page in a span of eight or more pages the index does not hold (a table of dispositions) is a coverage gap; the nominative-first form ('7 How. (U. S.) 283', '12 How. (53 U. S.) 299') is read as the official book when its volume is written, else as the nominative book, and is an event for Id.; a parenthetical beside a volume that is a series qualifier rather than a state ('5 Neb. (Unof.) 67') names another book and is never read as the plain reporter; a nominative label beside a U.S. number that is the nominative's own volume ('13 U.S. (Wall.) 335') is read as that nominative volume (80 U.S. 335), said in the evidence; a state qualifier on a label whose held record is another jurisdiction's or absent ('1 Har. (Del.) 5') is review — a tag written with spaces ('2 Lans. (N. Y.) 506') names the same state; a number after a reporter is the NEXT citation's volume — said in verification.notes, never served as a first page — when a page list, a range or a note tail and 'of <volume>' follow it ('of 176 U.S.App.D.C., 356, 357 of 538 F.2d'; a range worded 'to' / 'through' only behind a page word: 'pages 427 to 428 of 176 U.S.App.D.C., 55 to 56 of 541 F.2d'), when another reporter AND ITS PAGE follow it ('pages 243 and 244 of 332 U.S., 67 S.Ct. at page 1604'), or when a page word and 'of' stand before the volume and a reporter follows the number ('page 17 of 221 U. S., 31 S. Ct. 502'); a full citation after 'at page N of' ('at page 105 of 310 U.S. 88, 60 S.Ct. 736') keeps its row, and so does a citation before a caption, a heading or a court slot that opens with a reporter-shaped word ('55 L.Ed. 619 U.S. v. Bausch & Lomb', '113 S.Ct. 1710 A. The …'); 'in' before a caption's plaintiff introduces it — the words before it are the speaker, never the plaintiff ('The Supreme Court in U. S. v. Breitling') — while a party's own 'in' stays the party's: on the defendant side always ('Rousso v. First Nat. Bank in Detroit'), on the plaintiff side as a second reading that can only verify ('Citizens for Ethics in Government v. X'); a written side ends at a comma followed by a volume – abbreviation – page shape, whether or not the reporter is listed or its volume is damaged ('Silverthorne Lumber Co. v. U. S., 40 Sup. St. 182, 64 L. Ed. 319', 'Wong Sun v. U. S., 317 [371] U.S. 471'), while a thousands comma ('United States v. 5,870 Bags') and a party's initials before a comma ('A. & J., Inc.') are no caption boundaries; the abbreviations of Commission, Department and Board (Comm, Commn, Dept, Bd) are read with or without their mark ('Lawrence v State Tax Comm, 286 US 276' names the State Tax Commission; 'Comm v. Mills' the Commonwealth), a marked abbreviation whose stem opens a record word (\"Comm'r\", 'Commr.' beside a Commission) is the variant tier, and 'et al.' after a sovereign is a designator ('State of South Dakota et al.' is the State); a white-space run of more than 64 characters (any Unicode white space), or a run of more than 64 invisible glyphs or stray marks, is a break no reader crosses — a citation, a caption or a pin written across such a run is not read, and verification.notes says how many runs were read as a break; every offset is unchanged; a bare digit and a comma after a first-series reporter has two readings — it is the page and its pin where the series reading is refused ('24 P. 3, 4 (1890)' is no 24 P.3d 4) or where the first-series record is the case the written caption names (the citation's own caption, or the one its parallel legs share) and the series record is not, and the series otherwise ('Ryan v. Hearst Publications, 100 P. 2, 24 (Wash. 1940)' is 100 P.2d 24); a series reading that falls in a volume the index does not hold stays that gap — the other reading is never accused, nor silently verified, beside a reading that was not read; a placeholder for a page not yet assigned ('576 U.S. -, 135 S.Ct. 2551') ends a caption side as a paged citation does; a name in a parenthetical after a citation is that case's alias for a later short form only where the parenthetical DEFINES it ('hereinafter', quotation marks, or a word of the citation's own caption) — '(Blackmun)' names an author; a side that is only designators ('Railway Co.', “R'y Co.”) names no party on either side; a court-year slot between the caption and the citation ('…, D.C.S.D.N.Y.1955, 134 F.Supp. 133') is read through; a collapsed same-case duplicate cited in full is the same case by name and date; a held record in an order-table region of the index (one-page entries on or beside the page) is one of many entries — a caption naming another is a review, never a mismatch; the number before a following reporter ('40 W. Va., 22 So., 58') is that leg's volume, never this book's page; a same-day neighbour that shares a whole party side is worded as a companion or a namesake, never as proven; one caption on two records of a volume binds a short form by the record whose extent holds the pin; a record reference written as a title and a page ('Stenographic Transcript of Testimony, 113.') breaks an Id. chain; a book written after a pin with its own page in any form (', 74 S.Ct., at 692') is a leg; a pin written 'at page P, V Rep.' belongs to the book that follows; the own-leg test reads series ordinals ('524 F.2d at 1298', '98 L. Ed. 2d at 873'); a pin inside the record adjacent to the cited one that is the same decision continued or a same-day companion of the same court ('License Cases, 5 How. 504, 582'; 'Smith v. Snow, 294 U.S. 1, 23') is a review, never out of range; a short name or 'supra' name is identity evidence only when it is unique in the whole document (a footnote printed after the body counts), and the supra's name is read from the text before 'supra' ('Ex parte Bakelite Corp., supra' names Bakelite; the parser alone hands over the last word) — a name that is only marked abbreviations ('Const., supra', 'Const. Co., supra' whose own case the parser truncated) names nothing, unresolved, never bound by those words; a supra or Id. pin that a reporter with no volume or page of its own follows ('supra, p. 1032 L.Ed.'; the dotted run read whole, so 'Id. at 9, J.A. 276' is a page of the record, not a book) is that book's page — the pin is unchecked and said, with the note stating whether the short form keeps its antecedent or is bound to no case; a reporter with a word glued to its last period ('U.S.Code Cong.', or a scan's 'L.Ed.The Court') is two readings — the pin is unchecked and the note says only that; 'n. 2C.' names no book; a year written right after 'supra,' that a full citation follows ('Keogh, supra, 1922, 260 U.S. 156, 164' — the year-first style) is its year, never a pin, while a number after 'at' or a page word is a page; a supra name that is an initialism ('N.L.R.B., supra') binds only the case whose party IS that initialism; a number after an Id. or supra that a reporter-shaped token follows which cannot be read (a glyph inside it, a book the table does not hold) is unchecked and said, never a pin; a comma-form page that a white-space run of more than 64 characters follows is unreadable, never a first page; an explanatory lowercase run after a comma (', citing X', ', quoted in X', ', following X', ', which cited X') nests the authority it introduces; a quotation or colon block encloses until it closes (a blank line ends a colon block), however long; in the locative form 'at page P of V Rep.' the page belongs to the book that follows it, and a page written after the book ('V Rep., at page P of W Rep.') is the next book's (unread); a short name before ', supra,' ('Perry, supra, 408 U.S. at 601') names the antecedent; a citation carrying its own WL or LEXIS leg is an unpublished disposition (a table entry); a containing record whose own citation is not its first page is a source-data gap; nominatives written '(4 Wheat)', '(8 Wall)', '(1 Cr.)' or with 'U. S.' are read; a list request keeps the nominative half of '6 U.S. (1 Cranch) 137' as a reading and reads a trailing pin ('347 U.S. 483, 999'); the glue is read one member at a time (one renamed caption, only before the first parallel leg, and a run of lowercase words directly before the verb — 'rehearing granted in part and opinion withdrawn in part,', also after a renamed caption — where a run that itself narrates an order ('affirmed and we denied certification,') is the order and a clause connector ends a run, but never prose introducing a caption: a new case after a finished history citation is checked on its own; a semicolon always opens a new citation); the legs of one parallel string take their caption from the text when the reporter breaks it; a parallel leg is no mismatch when the first leg's record carries the written citation (a page the reporter shares between two decisions — 'Gelling v. Texas, 343 U.S. 960, 72 S.Ct. 1002'), or when the first record says nothing about that reporter and the second leg's case is a case of the first leg's reporter too (two cases of one court may share a page); it is a mismatch when the first record's own citation in that reporter is another page ('Roe, 410 U.S. 113, 93 S. Ct. 2607') or when the second leg's case belongs to another court ('Brown, 347 U.S. 483, 125 So. 839'); the old comma form of a nominative citation ('5 U. S., 1 Cranch, 137', '14 U. S., 1 Wheat, 304', '83 U. S., 16 Wall., 130', 'Cr.' too) reads the number after 'U.S.,' as the nominative volume when the table agrees, with the pin after the page; a reporter written with a trailing comma ('347 U.S., 1999') is that reporter; a pin the caption-named slip holder holds is that case's pin, never accused beside the slip review — the label, the entry and the structure agree, and only that row's label is withdrawn; a page one or two before the NEXT record of the volume when the caption names that record ('Palmer v. Columbia Gas, 479 F.2d 152' where Palmer starts at 153; 'Bolling v. Sharpe, 347 U.S. 495' where Bolling starts at 497; 'Ogden v. Saunders, 12 Wheat. 212' where Ogden starts at 213, on a mismatch too) is a first-page slip → review, a record further on is another case; a verified leg whose caption a parallel leg verifies ('Coca Cola Bottling Works v. Hunter, 95 U.S.App.D.C. 83, 219 F.2d 765') is two readings, never a mismatch; a government body written with its sovereign after it ('Tax Commission of the State of Washington') names the record's 'State Tax Commission'; a side that is only 'U.S.A.' is the United States, and a trailing year or docket ('U. S. A., 1960', 'State, No. CAAP-…') is no party; an office written with its place ('Kent County Clerk') is an office; the caption test every within-case, slip, companion and table reader uses is the attribution lane's own (initials 'W. U. Tel. Co.', the Bluebook abbreviations 'Educ.' / 'Comm'r' / 'Ass'n', the printed M' for Mc, spaced initials 'N. L. R. B.'); a memorandum reporter's Westlaw leg is no table marker and a disposition parenthetical ('(granting petition for review)') is none; a reporter written with a scan glyph for its period ('162 F;3d 795') is read with the glyph folded and said; a mid-party 'etc.' ('Chicago, etc., Ry. Co.') is the caption's own word; the nominative written without its parentheses ('60 U.S. 19 How. 393') is 60 U.S. 393; a capital 'V' separates parties only with its period written and a Roman numeral ('Title VII,') is never a party; a caption separated from its citation by a sentence end, or two captions joined by 'and', never accuse; a caption read from the text never reaches through the previous citation; a table-marked citation whose only record is a secondary one is a review, never a mismatch; a caption a signal word introduces right after a narrated order ('dismissed the appeal …, see State v. Bonnell,') is the order's own citation, and '(dismissing appeal)' is an order marker; a '(table)' marker after the parallel leg and year counts for the first leg; the popular name's 'case' reads in either case ('the Dartmouth College case'); a party-list tail ('et al.', 'etc.', 'et ux.') is no party and never disables the caption; the glued separator reads 'v.', 'vs.', 'V.'; order phrases are an order noun with an order verb, periods optional ('cert den', 'rev allowed', “rev. dism'd”); a period-less order phrase ('rev den 295 Or 617') is an order; a caption with a glued 'v.' ('Brown v.Board of Education') is a caption; 'appeals dismissed' is an order phrase like 'appeal dismissed'; a verified leg whose caption names a candidate of an ambiguous sibling leg (two appeals reported together at one page) is a companion appeal, never a mismatch, and a secondary-corpus rescue at a page the primary index holds two cases at is ambiguous; the caption fold reads an elision whose stem opens a record word (“Comm'r” / 'Commission') and a transposed or one-letter-off acronym ('NRLB') as variants, never mismatches; the caption fold reads ligatures ('Phœbus' / 'Phoebus'), a side that is only a postal code ('PA', 'NJ') as its state, an amount in words as its digits ('Four Thousand Four Hundred Thirteen Dollars' / '$4,413'), and a vowel-less contraction as the word it abbreviates ('Wkrs' / 'Workers', 'Brhd' / 'Brotherhood'); 'Hill' names New York's reports (ingested as 'Hill & Den.') and South Carolina's, the caption deciding; a written caption never spans an 'Id. N' short form or a sentence end — a period after a word that starts lowercase, never 'St.', 'Mfg.' or 'Ry.' (the words after it are the caption); the within-case caption test walks acronyms WRITTEN as such ('Holmes v. SIPC'; 'Doe' is a surname), reads the WRITTEN caption one way in every test of the not-found branch (the extractor's parties, else the caption in the text — Bluebook abbreviations included: “Bd. of Cnty. Comm'rs”) against the record's short name as the attribution lane reads it and against its FULL name read side by side, one listed caption at a time (the written plaintiff on that caption's plaintiff side and the defendant on its defendant side, or the caption reversed as a whole: 'People v. Salem' names 'People ex rel. Detroit & Howell Railroad v. Township Board of Salem' and 'Davis v. Alabama' names an entry of a consolidated order record; a side is NAMED only by a distinctive party word with every party word of it on the record's own side, and a sovereign or other generic side is a party only where a listed party begins with it; a caption that shares one co-party with a long caption, or pairs parties of two listed cases, names nothing; a caption named on one side whose other side stands only inside a party's description ('German Alliance Ins. Co. v. Kansas' for … v. Lewis, Superintendent of Insurance of the State of Kansas), or one met only by a generic word or a one-letter slip while a party word stands nowhere on the record's side, is two readings — a review, never the pin reading and never not_found; and the VERIFIED pin reading needs EACH written side positively met on the record's same side (its short name or one listed caption of its full name, either order): every party word met by a word of that side — a NAME by a name (equal, a particle written apart or with an apostrophe, one edit — never against a sovereign or a common word, and never where the same volume holds another case under the caption AS WRITTEN: 'United States v. Burns, 934 F.2d 149' inside United States v. Burks, the volume holding a United States v. Burns, is review; never a longer or shorter name that shares its opening letters: 'Williams' is not Willard, 'Long' is not DeLong), an abbreviation only where the text marks it with a period or an apostrophe ('Com.' is no Cowman, 'St.' no Schmidt), initials only where the text writes initials and only for consecutive words of the record, a municipal word ('City and County of') passed over only beside a municipal record or a record that is the bare place ('City of New Orleans' for New Orleans — never a private party named after the place), and never through a role word, 'et al.' or a given name of the record's full name — a bare sovereign ('United States', 'State', 'People', a state's name) only where a listed party IS a sovereign of the same family or its officer or body named WITH the sovereign word ('Ohio State Auditor'; never 'Mayor of New York'), never a company or agency named after one ('United States v. Johnson' inside Johnson v. Georgia Highway Express) and never the other family, and an officer title ('Commissioner') only where it stands on that side; a ONE-party record ('In re Johnson') is a two-party caption's own only through a caption its full name lists; a caption that names one side of the case and something else — or nothing the readers can meet — on the other ('Gunter v. Union Pacific Railroad' inside Gunter v. Atlantic Coast Line Railroad, 'Lin v. Holder' inside Zelaya v. Holder, 'Sidaris v. Commissioner' inside Sidaris v. Runyon) is the same two-readings review, never the pin reading and never not_found; a ONE-party caption ('Ex parte Young', 'Matter of Smith') written before a page inside a two-party case never borrows the document's full citation of that case — it stays not_found), reads a written reporter abbreviation that is the leading part of another held book's name as two readings when that book holds the page under the caption ('United States v. Hall, 2 Wash. 366' — 2 Wash. C. C. 366), reads 'the X Case', a procedural one-party caption of a one-party case ('Ex parte Hawk', 'in Re Jacobs'), the caption written before 'sub nom.' as the record's other name, a name particle on either side ('de Lavrier' / 'deLaurier'), and a reversed caption of the containing case (a slip → review), treats a generic side ('Bank') as untestable and a name particle as no identity ('Connor' / \"O'Connor\"), and the document's own full citation of the containing case — anywhere in the document, a parallel leg of the same string included; the SAME record, or the same name decided the same day; a case of the same name decided another day (a rehearing or the decision below printed apart, or a namesake) is never the pin reading — at most a two-readings review naming the record the document cites; and never when the citation is written under a two-party caption that names another case, which is not dropped — or 'id.' is a second signal (the caption is read as written: the extractor's parties, else the caption in the text, where a lowercase 'v.' separates the parties and a capital 'V.' among initials is an initial — 'Yazoo & M. V. R. Co.' — and the next-record first-page slip reads it the same way); a not-found page in a volume the index holds without page extents is a coverage gap, and so is a page in a hole of a regional volume whose own records cover a minority of its pages (the other states' cases live in their own books, or, for a single-state reporter, its other cases are held in other books or not ingested — 'fully indexed' is said only of a volume its ingest tiles; a record whose last page precedes its first is no coverage); the first page of a hole of three or more pages (an opinion withdrawn after review was granted) and a page between two order-table pages (pages of one-page entries) are coverage gaps; a page after a record whose indexed extent is unusable (a last page '?', a garbled range) is a coverage gap, never 'fully indexed'; a page equal to a record's LAST page where no record starts is a boundary (a case may begin on the page where another ends) — a coverage gap, never 'falls within'; an accusation at a citation where the secondary corpus holds a record without a case name is two readings → review; a caption naming any case the corpus holds at an arbitrated page (either order, both sides) is never accused on its name and that record is SERVED; a pin equal to the next case's FIRST page is a companion string cite → review (it demotes the document; a pin inside the next case is out of range); a stray period or comma after the next leg's volume is never a pin, and only a REPORTER's leg refuses a pin ('No.', 'C.A.', 'Fn.' leave it a pin); a leg absent from its volume beside a sibling leg that verifies the string's caption is review (the legs share the first leg's caption; in a run of more than 25 comma-joined citations the sibling legs read for a leg are the twelve on each side of it, the caption still the run's); a record side that is an OFFICE ('Attorney General', 'Collector') is untestable only against a PERSONAL written name that could be its holder ('CIO v. Wemyss', 'Acme, Inc. v. Wemyss', 'Cent. R.R. v. Wemyss' are mismatches); a pin a stray character splits ('999~5', '49~5 n.3', 'at 49~5') is unchecked and said, and a glyph after a range end the walk read splits the range; a list request resolves '5 U.S. (1 Cranch) 137' as a text request does; a companion-pin review demotes the brief as it demotes the check; a visible OCR character inside a number ('73Ú', '240.10b~5', '§ 78~j', '2°6 U.S.C.') is review naming the character in every lane; a section in a parenthesis after a Part / Subpart head is read, 'Subpart E of Part 164' is the part with its subpart (a lettered list reads its first), a '§'-marked section of another part after a singular head is read beside the part as two readings with its list members read or said, 'Section N of the Agreement' after any head is prose while the head's own section is read whatever follows it, a marked section after a singular part or a subpart-of-part head ('Part 164 and § 164.9999', 'Subpart E of Part 164; § 164.9999') is read or said with every member after it, a head subpart before a section ('Part 164, Subpart Z, § 164.502') is said as context, every subpart of a list of a part ('Subparts D, E, and Z of Part 164') is a unit and a range names both ends, every part named after 'of Parts' is said, an edition year written '(Supp.1960)' or behind a hyphenated pin counts and the year after a list's last member is the list's, a member written with the part elided ('§§ 703.003, .115', '.301-.312', '.301 to .312', '.212-5', '.402(g)-1') takes the head's part as the whole written token with its range, pins joined before the edition year by one or more joiners ('(b) and (c) (1960)', '(b), (c), and (d) (1960)', '(b) and (c), (1960)') are skipped, a ';'-opened member after a plural head is a member whether or not it carries its own mark (read or said), and the year after a said member — through an 'et seq.' tail too — is the group's, a ';'-opened marked list after any head reads every member, and an 'id. §' statute short form is said; a one-page-everywhere volume has no pin extents; two long words sharing a stem ('Railroad' / 'Railway') are the variant tier; the words after a sentence-opening comma word are the caption's own party ('As in Gideon, Smith v. Wainwright' is the variant tier) and a variant reached through a second reading says so; a soft hyphen in a section number is two readings whenever both verify ('§ 19­83'); a government body passes only on the record side at its own position, and with no named sovereign only a state-executive body (department, agency, commission …) against a state-level side — 'Board of Education' keeps its name check, and a body written where the record's sovereign sits on the OTHER side ('Jones v. Department of Corrections' for 'State v. Jones') is the reversed-caption review; a record's own abbreviations spell a written side ('Railway Co.' / 'Ry. Co.'), two record words joined spell a written word ('AntiFascist'), and a comma-ended name before a role phrase ('Butner, Trustee in Bankruptcy') is the party; a period after a WORD ('California.') is a sentence's end, never an abbreviation; a record carrying a popular name in the Reporter's form ('Lottery Case', 'Legal Tender Cases') beside a written 'X v. Y' is a case_name_variant advisory, while a one-sided record ('Estate of Snively', 'The Antelope') is compared on its own words; a state's Bluebook abbreviation ('Pa.', 'State of NY', 'W. Va.') is the state, and a state after 'of' qualifies the institution ('Board of Trustees of the Univ. of Ill.'). A soft hyphen (U+00AD) has two readings — nothing, or a dash — and a citation is never accused on the dash reading when the string without it exists ('§ 240.10­A-3'). A section token that mixes capital and small letters ('§ 7St(f)', a scan of § 78t) is manual_review_required, never statute_section_not_found. A citation year that disagrees only with a SECONDARY court-records record's date is manual_review_required (a review finding that moves the document status and counts in attribution_flags), never citation_year_mismatch; an amended-opinion advisory (a year later by 2–3) is read first. A caption written after ANY leading word ('Unlike X v. Y', 'Contrast X v. Y') whose tail is the resolved caption word for word is never a name mismatch. When the record’s own caption writes a relator (‘State ex rel. Curry v. Morrow’), a written relator must name it — the sovereign head alone (‘State ex rel. Zzyzx’) no longer passes the side; against a record with no relator the head alone still does. A closing parenthesis after a lettered section is read as a lost paragraph opener only when it is UNBALANCED at the citation’s own position — ‘(citing 42 U.S.C. § 1983Z)’ closes its own parenthetical and the absent section is accused as written; an absent number-plus-lowercase-letter section whose bare section exists and holds that letter as a top-level subsection (‘29 U.S.C. § 187a’ beside § 187(a)) is manual_review_required naming both readings — a run-in subsection designation or a lettered section — never statute_section_not_found. A Code EDITION written in the slot after ‘U.S.C.’ (‘18 U.S.C. 1952 ed., § 371’, ‘(1958 ed., Supp. IV) § 371’, ‘1964 ed., Supp. V’) is an edition, never a section: every ‘§ N’ the head introduces (a list included) is a citation to that edition, the evidence names the edition, and a section the current Code lacks is manual_review_required naming the edition (it may have been renumbered or repealed since), never statute_section_not_found. The edition rides every occurrence of the section in the document, whichever is written first. A section whose own text prints a numbered level inline above its marked-up subdivisions (‘42 U.S.C. § 4332(2)(C)’ over (A)…(L)) leaves every paragraph pin unchecked, never ‘does not exist’ and never found (a chapeau that merely cross-references a designation — ‘section 3324(a)’ — is no inline level); an ambiguous FIRST pin token (‘(v)(l)(A)’, ‘§ 3505b(l)’) takes the class of the section’s own top level, so a scanned ‘l’ reads as 1 wherever the section starts at (1). Every paragraph item listed after a pin by list glue (‘§ 1101(a)(43)(F), (Z)’, ‘(F) & (Z)’, ‘§ 230(c)(1) and (e)(3)’), of any designation class, is named in verification.notes as written and said to be unchecked — never dropped; a paragraph range joined by ‘through’ or ‘to’ (‘§ 1101(a)(43)(F) through (Z)’) is read exactly as the dash form is. A dash string whose compressed end names a real section BELOW the start (‘§§ 481-57’, ‘§§ 1968-61’) is a range written in reverse — manual_review_required when both ends exist, never statute_section_not_found — under a single ‘§’ too (‘§ 1968-61’), with a one-digit end under ‘§§’ (‘§§ 1968-6’) and with a full-width end below a lettered or hyphenated start (‘§§ 1983-1981a’, ‘§§ 2000e-5-1981’); under ‘§§’ a hyphenated start followed by an end shaped like a section of the title (‘§§ 78u-4-79’, ‘§§ 701-1-709b’) is the range to that section as written, never a family sub-number.","docket_numbers":"since 1.20.10 (2026-09-29): a citation-shaped string written after the docket word 'No.' / 'Nos.' ('United States v. Sprei, No. 26 MJ 91 (E.D.N.Y.)', the magistrate docket form) is a docket number, never a reporter citation — it is named in verification.notes and no label is served for it; before 1.20.10 it was read as 26 M.J. 91 and accused not_found_in_covered_volume against the case at page 90 (verdictnews 2026-09-29).","pin_extents":"since 1.20.10 (2026-09-29): pin_cite_out_of_range is raised only against a page span that is the cited reporter's own — a row whose first page is not the page the citation names, or whose span ends before it starts, leaves the pin unchecked (pin_cite_checked=false; measured on 200 real opinions: 18 of 30 full-citation pin accusations rested on such a span, landmark cases among them). A short form WITHOUT a reporter ('Id. at 688', 'supra, at 694') names the case in whichever reporter's paging the writer uses: it passes when the page lies inside that case's span in a reporter the document cites FOR THAT CASE (in full or as a short form of it); a span of a reporter the document uses only for another case leaves the pin unchecked; an Id. after a short form binds the parallel group its antecedent was cited in, and a group with a leg the index cannot read ('41 U.S.L.W. 4213') leaves the pin unchecked; a short form that names its reporter ('93 S. Ct. at 928') is checked against that reporter only. Over the same 200 opinions the label fell from 395 to 138 with no citation newly accused; the remainder is mostly a short form whose true antecedent is not a citation the reader extracted (a record or brief page, an unindexed reporter) — read a short-form pin_cite_out_of_range as a prompt to check the antecedent, not as proof of a fabricated page. The next case’s FIRST page, one page past the cited case, is manual_review_required (a companion string cite written as a pin, or a pin one page off) for every form alike — the full citation’s pin, an Id., a supra and a volume short form; an old-style ‘id. N’ with no ‘at’ after a semicolon in a string list (‘10 Wheat. 1; id. 51’) names page N of the same book, a citation of its own, and is never range-checked as a pin; a nameless volume short form (‘483 U.S. at 640’) whose text since the volume-matched case names ANOTHER cited case, with the page outside the matched case, is unresolved (both readings named), never out of range; a record locator whose last word merely begins like a month (‘Declaration 45’, ‘Decree 12’) is a locator — a month is a whole word; a state or sovereign word written as prose (‘California law’) never unbinds an Id., but written in a case-name frame (‘In Washington, the Court …’, ‘Washington’s holding’) and naming a party of another cited case it is a second reading — the Id. is unresolved, said as ‘may name’, never accused; the other case may be cited by a parallel reporter (‘107 S. Ct. 3034’ for a ‘483 U.S. at’ short form), and a case present in the gap only as its own citation and caption is no naming (only the words the case’s own name carries are blanked, never the prose before them); a state or sovereign word is prose only when a generic noun follows it (‘law’, ‘courts’, ‘Supreme Court’, ‘legislature’) — written any other way and naming a party of another cited case (‘Under New York,’, ‘As Massachusetts recognized’, ‘In New York v. United States,’) it may name that case and the Id. is unresolved; a second reading is refused only on a POSITIVE contradiction (the other case’s record lists the short form’s reporter at a different volume); an Id.’s own parallel leg (‘Id. at 47, 359 A.2d at 206’) that no cited case carries and the antecedent’s record does not list is unread and said — the Id.’s pin is still checked against the antecedent, and only a leg that names another cited case, or a leg whose reporter the antecedent’s own record lists at another volume, contradicts the binding; a word of a case’s caption is blanked only when the case’s RECORD name carries it (abbreviations read — ‘Bd. of Educ.’), and a case the index does not hold is blanked only up to the first word naming another cited case; a state word is read per occurrence — one generic use (‘New York courts’) never silences a naming use (‘Under New York,’); a case name of any length the lane tokenises (‘Lee’, ‘Roe’, ‘Doe’, ‘Ker’) and a lowercase-led name written as the caption writes it (‘eBay’, ‘al-Kidd’) name a case as a longer name does; the blanked caption of a case cited in the gap consumes the record name’s words in order, so a prose naming that shares a surname with the next caption (‘followed Anderson. Anderson v. Liberty Lobby’) is never blanked; an Id.’s pin is the pin the text writes after it (‘Id. at 490–92’, ‘Id. at 490 & n.4’), never only what the citation parser read; an unread parallel leg’s note says ‘checked’ only when the antecedent’s extent is held; a citation-shaped short form the parser did not tag (‘(Anderson, at p. 1138.)’, ‘Anderson, 483 U.S. at pp. 640-641’) is a citation event the lane never checks a pin against: unless it names exactly the antecedent’s own case, the Id. after it — and the Id. chain that follows, until the next tagged citation — is unresolved and said; a written pin’s page word may be ‘p.’, ‘pp.’, ‘page’, ‘pages’ or ‘pg.’ for an Id., a supra and a reporter short form alike, but with no ‘at’ and no comma before it the word must be lowercase (‘Id. Page 52 of the trial transcript …’ opens a new sentence), and a page ‘of the record / transcript / appendix / brief’ is never the case’s pin; an Id. whose pin lies outside its antecedent but inside the case the short forms before it were bound to or named (a citation inside a quotation or a colon block took the binding) is unresolved, two readings said.","start_page_hit":"since 1.20.10 (2026-09-28): a case citation whose page is the FIRST page of a case in the reporter's own indexed volume is verified_exists even when the source's citation row carries another series label ('76 A.D.3d 144' whose row reads '76 a.d.2d 144'; 'So.2d' for 'So. 2d') — the evidence names the label error; before 1.20.10 such a cite was accused ('no case starts at page N … falls WITHIN <that very case>'). A page INSIDE a case is still the pin-page accusation. A citation that is the SOURCE'S OWN wrong-series label of a real case ('76 A.D.2d 144' for Mixon v. TBV, an A.D.3d case) is manual_review_required naming every such case, the cited volume's own case containing the page, and the citation to use — a writer who copied the source's label neither fabricated nor cited a pin page. A review-labelled CASE row counts like an ambiguity in the document status (never a plainly 'verified' document) and is a verdict_check_brief flag; verdict_get_case answers such a citation with the review evidence (an invalid_input error), never 'case not found'. EVERY page of a span of three or more pages the index does not hold between two indexed records (‘289 A.D.2d 1100–1102’) is outside_indexed_volumes — a coverage gap — as its first page is, never ‘fully indexed’. When one half of a nominative pair is a coverage gap (‘74 Tenn. (6 Lea) 339’, Lea unheld), the served half is the row’s target and the written pair is named in the evidence.","edition_lag":"since 1.20.7 (2026-09-14): a C.F.R. SECTION or PART absent from eCFR's latest edition at the DEFAULT date is accused only after a second signal. eCFR's edition dated D trails the Federal Register — a final rule published or effective on or after D can ADD a section that edition D cannot hold, and a partner's controlled pair (an agency order published the day after the edition, codifying a new section; the same article with a part-level wording verified) was accused. Now the Federal Register is searched for final rules published within 120 days before D or since, or effective on or after D: for a section, rules its index matches on the quoted number are read one by one until one's text names the section AS A SECTION (a bare decimal in a rule is not a section); for a lettered part, rules matching 'Part X' are read for the part; for a numeric part, any rule the Register records as affecting that part counts, unread. A rule that names it makes the absence a coverage GAP (reporter_not_covered, 1 credit) whose evidence names the document and the date to re-check, with data.unit.federal_register_document and data.unit.recheck_after (the later of its publication date and its indexed effective date; null when neither is on or after the edition — the effective date is then stated only in the document); an unreadable Register, more than 24 candidate rules, or an index page over 50 is a gap with the reason; no candidate, or candidates read that do not name it, leaves the accusation standing with the check recorded in verification.notes. A Register transport failure is not retried for a few minutes, during which every absence in that title and part is a gap. An explicit as_of is the caller's own date and is answered at that date WITHOUT the Register — pass one only when you mean that date. Items and appendix designators are unchanged. A cited number lying inside a subpart's listed range is not evidence the section exists — numbering gaps are ordinary. Cost: an absence at a default date now includes a Register round-trip (seconds; longer while candidate rules are read; cached for the day). Residual: a rule the Register indexes by neither a publication date within the window nor an effective date on or after D is not found, and so is a section created by REDESIGNATING a range (its new number is printed nowhere)"},"quote_attribution":{"since":"1.20.8 (2026-09-25)","refined":"1.20.20 (2026-10-21): the attribution reader was reviewed through passes 153–240 and a confirmation pass, and a pairing is served only when the words are shown to be the cited case's own. NESTED QUOTATIONS: words that carry a quotation of their own (another mark family inside them, '“… ‘…’ …”') are the HOST's — the citation the document gives for the whole; a '(quoting X)' member names only the inner quotation's source, so the outer words are checked against the host, never against X (unless the sentence names X's court and not the host's); a first quotation whose sentence holds a later nested quotation before the host's quoting member is two readings — the host's or the quoted source's — and is review naming both. QUOTING: '(quoting X)' and ', quoting X' are ONE statement in both spellings, after a full citation, a short form, an Id., a record, own-opinion, law or treatise citation alike: the words are X's; where X is cited to no case the lane resolves, the quotation is review naming both; 'citing X' attributes nothing and never becomes the antecedent of a later Id. HISTORY: a subsequent-history citation (aff'd, cert. denied and the like, in every spelling the lane reads) is never the quoted source and never the case a later sentence narrates ('The court there explained') — the merits citation is. RECORD AND OWN OPINION: a record citation and the document's own opinion citations (Maj. Op., Diss. Op., Conc. Op., ante, post, 'Dissent at 5', 'Op. at 12') are never a case: words they attribute are disclosed in data.not_checked with a reason that names them ('… this opinion's own dissent or majority (post/ante), not a cited case'), and an Id. after one never reaches the case cited before it (review). NON-CASE SPEAKERS: words the sentence gives to a party (an agency that argues included), a witness, the record, an instrument (an agreement, a will, a letter, a policy), a law or a secondary work are disclosed as not checked, the reason naming that speaker and its verb ('the document attributes the words to The officer (testified), not to a cited case — not checked'); a case cited BETWEEN that speaker and the words still owns them. COURTS: a sentence that names a court other than the cited case's — another level, another court system, or a court of another decision ('the court in a later case') — is review ('the citation after the quotation is from another court than the one the sentence names'); each citation's court is read from its OWN parenthetical, never the next citation's. POSITION (owner decision, 2026-10-18): a citation in the sentence BEFORE the words pairs by position only when the words' sentence continues the citation's — an empty or lowercase continuation, introductory phrases naming no source, or sentences in which the cited case or its court speaks; otherwise two readings, review naming what stands between ('the words stand in a sentence with a subject of its own … two readings, so the citation of the sentence before is not assumed by position'). REPEATED QUOTATION (owner decision, 2026-10-18): a See-signalled citation after words this call already attributed to a non-case speaker or to another case is two readings — the source or support for the writer's use — review naming the earlier attribution. The nested-quotation, quoting and history rules REDIRECT a pairing to the citation whose words they are (a quote_not_found can follow only against that case); every other rule here demotes a pairing to review or refuses it, and a review or a refusal is never an accusation. MATCHING (1.20.20, the check side): the normalized tier now also folds the inner quotation-mark FAMILY (single inside double, curly or straight, at any nesting level) and a terminal punctuation mark written inside the closing mark where the source continues — the same words verify as quote_normalized_match; a quotation that OMITS the source's inner marks is a word_split review whose note names the omitted marks, and the 'page header split a word' note is served only when a page header or a split word was actually set aside. MATCHING (1.20.21, 2026-10-22, the check side; pass 243 and its review): every reading below runs only after every tier above missed, so nothing that matched before moves, and none is ever an accusation. ALTERATION FOLD: a bracketed affix of one to three letters glued to a word ('court[s]', 'require[s]', '[P]laintiff[s]') is read as the editor's alteration of the bare word the source prints, at a word boundary ('analyze[d]' never matches 'analyzing'); a match at a verified tier serves quote_normalized_match with the alteration named in detail.note, a match at a review tier stays that review with the alteration named; a bracketed word or phrase ('[the court]', '[plan]') and an affix the alteration tier already reads ('consider[ed]') stay the alteration review, and so does a quotation's own free-standing bracketed insertion the source does not print ('[a] defendant'). FOUR NEW REVIEW REASONS (label manual_review_required, the value in data.quotes[].detail.reason; never verified, never an accusation): omission_between_sentences — every sentence of the quotation (two or more, each of at least four words, no ellipsis) matches the source alone and in order, each starting within 400 characters of the end of the one before, and the source prints words between them that the quotation omits without an ellipsis; detail.note names the omitted source text (up to three passages, e.g. 'ibid.' or a citation string); source_alteration — the quotation matches only once the SOURCE's own bracketed alteration or insertion (at most 40 characters) is set aside or read as the one to four underlying words the quotation gives in its place ('his' for the source's '[the defendant's]'), or once both sides' brackets are; the bracket is named; text_layer_letter_spacing — a DOCUMENT gate on text input (verdict_check_quotes, verdict_check_citations, verdict_check_brief, verdict_check_ai_output): when single-letter alphabetic tokens are at least 0.30 of the document's alphabetic tokens over at least 500 (clean court opinions measure at most 0.189 over 2,267 real opinions; the two letter-spaced filings measured 0.53–0.54), the text layer is letter-spaced or otherwise character-damaged and nothing in it is accused: every quotation of the document is review with detail {reason, letter_spacing_ratio, bound} and is not looked up; every accusing citation label (case, statute, regulation, rule) reads review with resolution.reason text_layer_letter_spacing, resolution.read_as (the label it would have carried) and the ratio, the attribution and pin findings beside it move to resolution.withheld_findings; and every CASE citation's verified_exists also reads review (read_as verified_exists), because a letter-spaced citation ('1 4 2 S. Ct. 1 7 0 8') is read as a fragment that resolves to another case ('2 S. Ct. 1') — all 13 such verifications in the two filings were another case; a statute, regulation or rule verification is not withheld. The ratio is read over the whole document: a terse transcript or a caption and contents page standing alone can reach it (review, never an accusation), one damaged block inside a long clean brief can stay under it, and a document under 500 alphabetic tokens is not read; text_layer_ocr — the quotation missed, the DOCUMENT's own text layer shows OCR confusion damage (three or more distinct damaged tokens), and the quotation repaired with the document's own vocabulary (a damaged token read as the one common word the document prints elsewhere, or a dropped space beside a function word: 'supp011ed' read as 'supported', 'ofa' as 'of a') matches the source at any tier: detail.repairs [{printed, read_as}] names the repaired tokens and detail.match_tier the tier reached; a repaired quotation that still misses stays quote_not_found; text input only (a quotes list carries no document). HTML: a '<' in a source's prose is text — a tag needs a name after '<' (a letter, '/', '!' or '?'), no line break and at most 250 characters; before this a stray '<' deleted the source's text up to the next '>' and a correct quotation inside it could read quote_not_found (23 real rows now verify). MEASURED on the real-opinion sample (29,239 quotation rows checked against their cited opinions): quote_not_found 6,893 → 6,582, 0 rows moved into quote_not_found or any accusation, 0 verified rows lost, 129 newly verified (106 alteration folds, 23 by the HTML fix) — every one read against the source and every one the court's words; the other moved rows are review (67 alteration, 53 omission_between_sentences, 38 ellipsis, 18 word_split, 9 source_alteration). On a partner's four hard-case filings: the two letter-spaced filings 216 of 216 quotations review and 0 accusations (6 had been quote_not_found against fragment citations); the OCR filing 4 of 6 quote_not_found rows review text_layer_ocr, 2 stay, correctly (the words are not in the cited opinion). The envelope did not change shape at 1.20.21: the four reasons are new values of detail.reason, and detail.letter_spacing_ratio / bound / repairs and resolution.reason / read_as / letter_spacing_ratio / withheld_findings / withheld_reason are optional keys inside dicts that already carry extra keys; classify by label and label_classes. Nothing in the envelope changed shape: a review is a data.quotes row with label manual_review_required, no citation, and the reason in its evidence after 'no attributing citation: '; a refusal is a data.not_checked row (verdict_check_brief: data.report.quotes_not_checked) with its reason; reason wording is evidence text, not an enum — branch on the label, the citation and the list a row is in","rule":"verdict_check_quotes / verdict_check_brief pair a quotation with the citation the DOCUMENT attributes it to, never merely the nearest one. A law the document names as the words' source before the quotation ('Rule 8(a)(2) provides that …', '42 U.S.C. § 1983 states:') owns them, whatever is cited after as support; so does a law that OPENS the quotation's sentence ('Under Rule 56(a), summary judgment is appropriate where \"…\" See Celotex' — the words are the rule's, Celotex is support) unless a case is cited between the law and the quotation. A source the sentence itself names that is not a cited court — 'the district court held', 'the ALJ found', 'Plaintiff alleges', 'Defendants argue', 'the statute provides', 'the Agreement states' — owns the words too (disclosed, not checked) unless a case is cited between that source and the quotation; a party, a witness, the record and an instrument keep their words whatever citation follows the closing mark. A RECORD citation after the closing mark (Tr. 45:3, Dep. 12, Ex. A § 4, Dkt. 22 at 4, ECF No. 7, Compl. ¶ 12, AR 25, JA 100) is the stated source, as is an Id. that follows one; so is an authority outside the case corpus (a treatise, a law review, a Restatement, legislative history, a slip opinion or database cite, the Constitution) — all disclosed, never checked against a case. Otherwise the next case citation after the closing mark within 250 characters is the source unless the gap crosses a paragraph break, closes a citation's parenthetical (the quote belongs to THAT citation), carries a non-attributing signal (see also, cf., but see, accord, see generally, compare, contra — support, not source), or belongs to a LATER sentence — a citation sentence that follows the quotation's own attributes only when it opens bare or with See / e.g. / quoting ('\"…\" Tr. 45. Under Twombly, that testimony is insufficient' attributes nothing); otherwise the citation of the same sentence or the sentence before is the source. A SHORT FORM ('Anderson, 477 U.S. at 250'), a supra and an Id. attribute like the citation they resolve to (short forms by volume, reporter and the name in their own caption — two candidates is no candidate; Id. to the citation before it, a statute included) — never to whichever full citation came last. A quotation the document attributes to a LAW — a statute, a code section, a federal rule (the citation form 'Fed. R. Evid. 403;' / '5 U.S.C. § 552(b).' / 'Cal. Civ. Code § 1542.' first after the closing mark or opening the citation sentence that ends the quotation's own; an 'Id. § 552(b)(6)' after a statute; never a rule merely mentioned in prose such as 'under Rule 12(b)(6) stated in Twombly') is NOT a case quote and is not checked — it is DISCLOSED in data.summary.not_checked (a count) and data.not_checked (the spans with the reason), as is a span that crosses a paragraph break, a bracketed placeholder in quote marks, or an unbalanced quotation mark. A quotation no citation attributes is manual_review_required with the reason in its evidence ('no attributing citation: the nearest citation follows a non-attributing signal', '… is in another paragraph', 'no citation in the quotation's sentence or the one before', 'a short form that resolves to no single case'). quote_not_found still means: the words are not in the text of the opinion the document attributes them to.PAGINATED TEXT (text that carries form feeds, as a PDF text extraction does): A PAGE BREAK IS A REVIEW BOUNDARY. The lane does not know what stands at a page's foot or top — a footnote body, a running head, a filing stamp — so a pairing that depends on text across a page break is not assumed: the quotation is manual_review_required with the reason in its evidence when (1) a page break falls between the words and their citation, (2) a page break falls inside the quotation's sentence, (3) the citation is an Id. whose antecedent citation stands across a page break (a NAMED short form — 'Bond, 572 U.S. at 858' — identifies its own case and still pairs), (4) the citation, that antecedent, or the full citation a short form or a bare name resolves to stands in the first three short lines of a page (read in the caller's own lines), where a running head may stand, or (5) the sentence leans on the sentences before it ('It held that …') and a page break falls within them. The rule only ever demotes a pairing to review; it never creates or redirects one, and it is never an accusation. Under a page break only known page furniture that carries a digit (a filing stamp, a page number) is removed, never a line with a quotation mark and never inside an open quotation. NUMBERED PARAGRAPHS are not modelled either: a quotation whose sentence leans back, in a paragraph that opens with a paragraph number, is review. A caption-party surname the document ALSO narrates as an actor ('Smith slipped on the ice. Smith v. Jones, … Smith said that \"…\"') is review: whether the words are the opinion's or the party's own is not shown. When the grammatical speaker is a LAW or a secondary source ('Congress enacted the Act in 1970. It provides \"…\" Id.'), the words are the law's and are disclosed as not checked, whatever Id. follows. REVIEW TIERS on the check side: a quotation carrying a bracketed ALTERATION of more than one letter ('[plan]', '[sic]') is matched with the bracket read as the words it replaced and reported as review tier 'alteration'; a quotation carrying an ELLIPSIS is matched piecewise in order and reported as review tier 'ellipsis'. a quotation that matches only after an inline FOOTNOTE CALL glued to the punctuation is removed — in the source ('…by counsel.1 If it had been') or pasted inside the quotation — is review tier 'footnote_call' (pass 143; before it a verbatim two-sentence quotation across a call read quote_not_found). None is ever `verified`: exact and normalized are the only verified tiers. SPANS are CODE-POINT offsets into the caller's text after newline normalisation (CRLF / CR → LF). A consumer that indexes UTF-16 code units (JavaScript `String.prototype.slice`) must convert: a character outside the Basic Multilingual Plane before the span shifts a UTF-16 index by one per such character. The served words are the caller's characters; only white space is the lane's. STRUCTURE the lane reads as boundaries, never as speakers: a heading (a title-case or numbered line without terminal punctuation) between a sentence and the pronoun that continues it is stepped over — the words stay with the speaker before the heading, or review when none is shown; a two-column caption, a table of authorities and a front-matter heading end the search for a speaker (the body begins below them). Each of these only ever demotes a pairing or refuses; none creates one. ","why":"a partner ran 27 court-filing templates: 26 quote_not_found rows, 0 true — every one a quotation attributed to the nearest citation rather than the one the sentence named (rule text beside a 'see also' case, a quote inside another citation's parenthetical, a short form resolved to the wrong case in the same volume), and a regex that let a short defined term's closing mark re-open as a quote. Court filings cite densely and use exactly the constructions that break nearest-citation attribution","measured":"the same 27 templates after the change: quote_not_found 26 → 2 (an ellipsis-elided quote and a quote that silently omits an internal citation — lane limits, disclosed, neither a fabrication; the third, the Sixth Amendment's text quoted to Klopfer, is now disclosed as the Constitution's words), real quotations verified 2 → 11 (a twelfth, statutory words under 'Under 33 U.S.C. § 1365(a) …', is now disclosed as the statute's), 22 quotations with no attributing citation reported as review with the reason, 14 spans disclosed as not checked; controls: fabricated words attributed to a real case still quote_not_found, a real quote still quote_exact_match; seven pre-deploy review passes turned 66 constructed shapes into permanent tests (a rule mentioned in prose, a statute cited in passing, an Id. after a statute, a short form in a shared volume, a footnote mark after a period, a rule parenthetical inside another, a record quote followed by a case in the next sentence, the summary-judgment sentence under Rule 56(a), 'the district court held', an Id. after a transcript cite, a treatise before a See, Bluebook abbreviations in captions). One corpus, one genre, labels counted; the harness (27 bodies, both control classes) is committed locally and re-run on every change to the lane152 adversarial review passes over the lane (pass 152 the last before this release), every reviewer probe kept: 1,179 permanent test rows in 528 tests, 30,828 pinned reviewer probes re-run on every change (0 outcome changes at ship), head/tail invariance over 30,357 probes, a quote-text invariant (served words equal the caller's characters) over 32,680 served quotations, 37 wrap-width sweeps (25–110 columns), 19 metamorphic families (about 1.9 million transformed documents at each close run). The page-break rule was MEASURED before it replaced a page model five consecutive passes had found accusing defects in: three real paginated opinions (43 pages, 110 quotations) — 0 outcomes moved; 441 paginated probes — 284 pairings became review, of which 133 were pairings the reviewer wanted (the cost), 88 accepted either outcome, 65 were wrong or unwanted pairings now closed. 1.20.20: review passes 153–240 (the chain bounded after pass 240) and a confirmation pass; 48,775 pinned reviewer probes re-run on every change (0 pinned outcome changes at the ship cut); two audits read every open reviewer expectation of the earlier probe modules (781 rows in the second alone: 162 defects fixed, 596 design rows decided). A REAL-OPINION GATE since pass 177: 2,329 quoted opinions (a fixed 1-in-40 sample of the case-text cache), 81,162 attribution rows, 0 crashes and 0 timeouts at the ship cut, every pairing gained or lost read at each cut. The two design demotions were priced on real rows first: the sentence-before position read (40 of 1,830 position pairings read: 12 the cited case's own words, 18 another source's or the writer's, 2 another case's, 8 ambiguous) moved 1,560 of 1,821 position pairings to review, 258 still pair; the See-cite over a repeated quotation (30 rows inside another speaker's quotation read: 14 the case's own words, 10 another source's, 6 ambiguous) moved 148 rows to review, narrowed to the two priced classes (36 pair again). The nested-quotation host rule moved 599 real rows from the quoted source to the host — a correct quotation of the host's words had been checked against the inner source, where quote_not_found could follow. A random read of 150 served pairings at the bounded cut: 143 right, 3 wrong, 4 doubtful (95.3 %, Wilson 95 % interval 90.7–97.7 %; wrong 2.0 %, upper bound 5.7 %); one of the three wrong classes (the nested first quotation) was then closed — 44 sample pairings became review. Eighty served reasons sampled: 79 true, one known false. ","blind_spots":"an ellipsis inside a quotation is not bridged; a quotation that omits an internal citation without '(citation omitted)' reads as not found; a parenthetical longer than 40 characters before the quote is not recognised as one; 'in accord with', 'compare that' and 'contra proferentem' are prose, not signals — only a signal that opens a citation clause is read; a stated non-case source with no citation between it and the quotation refuses the check even when the words are a case's ('Plaintiff contends that Twombly requires \"…\"' is disclosed, not checked); 'Under Rule 56(a), the Court in Celotex explained \"…\"' is Celotex's only because Celotex is cited between; '\"…\" The Court so held in Twombly, 550 U.S. at 570' is review (the citation sentence after a quotation attributes only when it opens bare or with See); a record citation is recognised by its Bluebook abbreviation — an unabbreviated 'Transcript at 5' is not; a block quotation set in its own paragraph with the citation in the next paragraph is review, not checked; an unbalanced '(' left in prose before a quotation can make an earlier citation read as the quote's owner (an unbalanced quotation mark is disclosed, an unbalanced parenthesis is not); a quotation more than 600 characters into a parenthetical is not seen as inside it; an Id. after a citation whose parenthetical cites another case resolves to the inner case; in verdict_check_brief a document whose only quotations were disclosed as not checked certifies 'Quotes: 0 found' with the not-checked count on the same line and a review checklist item — read that linepaginated text loses pairings to review by design (above) until a measured structure pre-pass ships; text pasted from a PDF WITHOUT form feeds is not paginated text to the lane — a running head or a footnote body inside a paragraph is read as prose; a quotation split across a page with a page number inside it is disclosed as not checked (the span crosses a paragraph break); a named judge as the speaker with a signal-led citation right after the words ('Judge Chen held that \"…\" See Smith') pairs with that citation — the conventional form for the opinion the judge wrote, not verified as such. 1.20.20 residue (reachable in filed prose, each pairs the citation and none is a fabrication finding by itself): a case as the subject of a verb that is no saying verb ('Cruz altered our understanding of a \"…\"') pairs that case though the words may be the writer's label; a law as the subject of a verb outside the listed saying verbs ('Section 1983 subjects to liability \"[e]very person who …\"') pairs the case cited after; a negated case sentence ('Hale nowhere adopts the \"…\" standard. Hale.') pairs Hale; a party's hedge ('On Apex's reading,') or a party's position predicate followed by a See-cite ('Appellees' position is \"…\" See Anderson') pairs the See case; 'Smith and its progeny hold' pairs; a plural court subject ('The courts of appeals have uniformly held'), 'Our high court held' and New York's Appellate Division against a Supreme Court (trial-level) citation are not read as another court; 'reconsideration denied' is not read as history; a document inside the cited case ('The expert report in Hale stated') and a lower tribunal inside it ('In Smith, the ALJ found', 'In Smith, the district court found') pair the case; 'Hale, and later the Supreme Court, held' pairs Hale; a decision noun after a lower court's act ('The district court dismissed the complaint. The opinion explained that \"…\" See Hale') pairs Hale; 'the Court' in a Supreme Court brief is read as the generic court; a STATE place opener ('In California, courts have held that \"…\" See Hale (9th Cir.)') pairs; a jurisdiction's law as the speaker ('Ohio law requires that \"…\". Smith') pairs Smith. "},"treatment_boundaries":{"since":"1.20.4 (2026-09-09); non-citable by rule added 1.20.5 (2026-09-10); how NOT to judge it, 1.20.6; the shapes a bar takes, 1.20.7","rule":"What NO citation graph can see, named beside the 'no claim check' boundary because partners measured it in the field: a judgment VACATED AS MOOT (no merits ruling overrules it); a case SUPERSEDED BY STATUTE or by the abolition of the body whose scheme it construed (the legislature acted, no opinion did); an opinion DEPUBLISHED or decertified on review (it stays real, findable and non-citable); agency GUIDANCE WITHDRAWN while its text did not change by a character (a pin check, a caption check and a text-stability re-fetch all pass, correctly); and an opinion NON-CITABLE BY RULE — barred from citation in any unrelated case from the moment it issued, by the issuing court's own rule (a memorandum opinion under a state appellate rule that says it 'shall not be cited or relied on for any reason in any unrelated case' is the shape). The fifth is different in kind: it is not displaced law at all — the opinion is correct, on point and scores HIGH on every relevance instrument BECAUSE it is on point — so a high relevance score is the aggravating factor, not a defence, and there is no later opinion for any graph to find. A treatment block with no signal on such a case is a coverage statement about a citation graph, never clearance — and for these five mechanisms a perfect graph would also be silent. Non-citability is judged against the issuing court's RULE, court by court, with the date the rule turns on: an 'unpublished' designation alone does not make an opinion non-citable (the federal appellate rule forbids courts to restrict citation of dispositions issued on or after 2007-01-01, and some circuits' local rules permit citing earlier ones for persuasive value — non-precedential, which is a status, not a bar). A publication-status field in a case database is NOT the test: it records whether the SOURCE regards an opinion as issued, not whether the ISSUING COURT permits citing it — a partner measured such a field at 0 of 37 recall on a known non-citable class (every barred memorandum opinion read 'Published'), and it fails silently in the reassuring direction. The evidence is the issuing court's own designation on the face of the opinion, read with the rule — never a quotation of the rule about another case (4 of 43 raw text hits were such quotations), and never a detector carried across courts: one court stamps an affirmative legend on a barred opinion, another's bar attaches to the ABSENCE of a certification, and the false-positive class differs with it (opinions filed unpublished and later certified). Read the rule, then the residue, per court. The signal is the RULE's operative sentence, never the document's label: across fourteen courts a partner read at source (a partner's reading, not ours), a detector keyed on the word 'memorandum' would be right in three and wrong in eleven — in Texas the memorandum opinion is the DEFAULT FORM of a fully precedential civil opinion (Tex. R. App. P. 47.4, 47.7(b)), and one numbered rule can treat docket types differently (47.7(a) governs criminal appeals, 47.7(b) civil), so read the subsection that governs the docket. A bar has SHAPES, and a per-case answer exists for the first three (from the opinion, its filing date and its authorship) and for none of the fourth: ABSOLUTE (decided by the opinion and the rule alone — Tennessee, California, Oregon); DATE-SPLIT (decided by the filing date; the legend is identical on both sides of the split — Hawaii, Massachusetts, Indiana); DATE-AND-AUTHORSHIP (Wisconsin: an unpublished opinion issued on or after 2009-07-01 may be cited for persuasive value only if AUTHORED — a per curiam or memorandum opinion is expressly not authored, so a date gate alone wrongly keeps every per curiam); CONDITIONAL (Nebraska: citable only if no published opinion would serve as well — a question about the rest of the corpus, so the same case is barred on one page and citable on another and no per-case verdict can be right). Beside bar-or-permit there is a DUTY: Georgia requires a physical-precedent opinion to be MARKED as such when cited — citable, but only with the disclosure. A jurisdiction whose rule nobody has read is UNEXAMINED, never 'permits'. A related mechanism a graph CAN in principle see but a text search cannot: ABROGATION — a higher court adopts the contrary rule, typically resolving a split between coordinate divisions, without expressly overruling the displaced decision; the displaced opinion's own text is silent about it by definition, so a judgment made from that opinion alone is structurally blind, and no phrase exists to search for. The mechanism a graph CAN reach is a later opinion that reverses, overrules, distinguishes or declines to follow, when the citing opinion is in the corpus and its edge is held (the secondary corpus graph's own coverage is stated in every block's window note)","example":"a page serving the entire losing side of a pre-2018 circuit split (four opinions, two later Supreme Court decisions) read well-sourced, on-topic and above every automated relevance floor; only a reader of the later decisions could see it. And three state memorandum opinions, correct and on topic, were being served as authority at relevance 0.60 while the rule they issued under forbade citing them in any unrelated case. A reader's finding is not a citator finding and is never served as one"},"disposition_consumption":{"gate_on":"in_part — a ruling is mixed only when an outcome carries in_part:true","never_gate_on":"len(outcomes) > 1 — MULTI-ACTION IS NOT MIXED. Real single dispositions parse to multiple actions: 'dismissed as improvidently granted' is ONE act (served as [dismissed]); 'affirmed; all pending motions denied' is an affirmance plus housekeeping; 'MODIFIED AND AFFIRMED' is one composite ruling","text_is_primary":"the verbatim text field is the authority; outcomes[] is a parsed convenience and may be empty when the recorded text does not parse","credit":"the gate-on-in_part rule and its examples were measured in the field by verdict.news"},"language_evidence":{"since":"1.20.9 (2026-09-25)","rule":"the negative_treatment_language signal (engines graph-v1-cl+language-v1 and graph-v1+language-v1-fused) orders its citing_opinions[] by whether the quoted sentence ASSERTS the treatment — the citing court states it, with no question, reported argument, condition, negation, modal or hedge on the way to the verb — then by year; evidence.assertive_citers counts the distinct citers that assert it. STRONG (beta) needs at least five ASSERTING distinct citers (before 1.20.9: five citers of any kind); two or more DISTINCT citers of any kind serve as review; one citer stays dark. The review note says when none of the quoted sentences asserts the treatment, or how many of the citers do","what_stays":"a citer that questions the treatment ('whether the Court should overrule X'), reports it ('petitioner argues that X was overruled') or negates it ('cannot conclude that X has been overruled') is still counted in citer_count and is shown only when fewer asserting citers exist than the envelope carries (three at strong, two at review); it never raises the tier. No field is removed. A secondary-corpus DUPLICATE cluster (one opinion imported under two ids: the same sentence, year, caption and decision date) counts once in citer_count and assertive_citers; a target whose only two rows are one such opinion is a single citer and stays dark, as depth-1 always has (before 1.20.9 the pair read as two citers and served review — 846 such targets)","direction":"demotion-only: a doubtful sentence reads as not asserting, which orders it later and can move a target from strong to review, never the reverse; measured over the 74k stored citer quotes before the change (46-49% assert — a quarter of the stored quotes begin mid-sentence, a mining artefact the next re-mine removes, and a fragment with no readable head never asserts; 994 targets moved to review (counted through the serving accessors, the duplicate-cluster collapse applied); the landmark overrulings kept STRONG and now serve the asserting sentence first). Absence of a signal is never clearance","cap_mined_lane":"the older per-target CAP-mined lane (engine graph-v1+language-v1, serving where the fused lane has no rows) reads the same way: asserting citers first, evidence.assertive_citers, STRONG only when at least one citer asserts (its single-citer STRONG stands on that citer's assertion — a per-target tight grammar), review when none does; 48 of the 69 targets it served alone moved to review","target_object":{"since":"1.20.11 (2026-10-05)","rule":"a quoted sentence counts as asserting only when the thing its verb treats IS the cited case: its name, its short form, 'the <Name> rule / test / decision', or a pronoun or 'that case' whose antecedent the sentence (or the sentence before it in the quote) names. A sentence that treats something else no longer counts in evidence.assertive_citers: the case's name used as a label ('the Terry stop was abrogated', 'his Miranda rights'), another case as the thing overruled ('Pennhurst concludes that … Tindal … was overruled'), the cited case as the one DOING the overruling ('Buckhannon overruled Baumgartner'), a ruling in the litigation ('Point fourteen is overruled', 'overruled Berry's request'), a pronoun whose antecedent is another case","direction":"demotion-only: the read is a second condition on an asserting sentence, so it can move a target from strong to review and never the reverse; a sentence the reader cannot place reads as not asserting. citer_count and the stored quotes are unchanged; a citer whose sentence no longer counts stays in citer_count and is ordered after the asserting ones, so WHICH quotes an envelope shows (three at strong, two at review) can change","measured":"four blind panels of three independent judges over 1,200 stored quotes; on the held-out panel, 89.9% of the sentences the reader stops counting are not about the cited case (134 of 149, Wilson floor 84.1%) and 99.3% of the sentences it keeps are (150 of 151, floor 96.3%). Over the stored quotes 3,740 of 35,468 asserting citer rows stopped counting (about one in ten); 40 more targets read review instead of strong — 1,034 baseline-strong targets now serve review (994 at 1.20.9), 575 serve strong; the older CAP-mined lane reads the same way (8 of the 276 targets it serves moved from strong to review)","limits":"about one in ten of the sentences the reader stops counting IS about the cited case (an antecedent two sentences back, an unusual list, a spelling the caption does not carry): the cost is a strong signal served as review. A few sentences still count that name another case sharing a party word with the cited one ('State v. Terry' for Terry v. Ohio). A caption with no readable party word is not read this way. BETA as before: verify the citing opinions before relying","why":"a partner's reader-adjudicated cases (2026-10-04) showed strong signals on decisions its readers found to be good law — Terry v. Ohio on 'the Terry stop was abrogated' and on a 'this case … is overruled' about another case","refined":{"since":"1.20.12 (2026-10-07)","rule":"the same read, with the sentence shapes a blind panel refused on 2026-10-07 now read as not asserting: a DERIVATIVE of the cited case (its progeny, a companion case, 'the Lochner era decisions', cases founded on or relying on it) is not the case; a REPORTED reading ('the district court agreed with X that Apprendi overruled …', '(suggesting Y overruled …)', 'Lewis implies …', ', the Board determined,', ', they say') is not the court's own; a CONDITION or a CALL ('[even if] X were overruled', 'until X is overruled', 'it is time for this Court to … overruling X') is not a treatment; a ruling on a filing ('Application for rehearing was overruled', 'Grounds of error #1 and #4 are overruled') and a docket fragment ('No. 00-21151 overrule X') are not about the case; the cited case named as the AGENT ('… was overruled in Plyler') is not the patient; and a treatment UNDONE on the case in the same sentence ('X was overruled in Guy, but was reinstated in Foremost', 'overruled Ringer, but only temporarily') is not negative treatment; a relative whose antecedent is a filing, a statute or a doctrine ('a plea of privilege which was overruled', 'a statute that was superseded') is not about the case, while 'a case which has been overruled' is; a clause-initial 'In X (cite), this case was overruled' names X as the venue. Shapes that now COUNT where 1.20.11 read them as glue: a rule established by or applied in the case ('the standard established by Albritton was abrogated'), a case listed with others ('those cases, together with Curd v. Brown, were overruled'), a relative with the court's own parenthetical ('which we have recognized was abrogated'), an appositive set off by dashes","direction":"demotion-only in effect: over the 93,901 stored citer rows 214 stopped counting and 54 started (every one of the 54 read — a word-boundary defect in the attribution read had matched 'say' inside 'hearsay' and 'Lindsay'); served tiers: two primary-corpus targets move from strong to review, one secondary-corpus target from review to strong","measured":"the panel that found the shapes: three independent judges over 1,014 asserting quotes of the 141 secondary-corpus records the 2026-09-30 graph reached for the first time — 751 unanimous assertions all still count, the 31 a majority refused all stop (pinned as tests); the whole-table diff against 1.20.11 was read row by row before the cut, and three independent review passes replayed it","limits":"a partial treatment stated as a foundation ('this standard, based on Conley, was abrogated') reads as not about the case; a sentence the miner glued to a page or footnote number may still read either way; the next re-mine removes the glue"},"why_refined":"the pre-ship panel of 2026-10-07 found a landmark (Almendarez-Torres) reading STRONG on a docket fragment, a conditional and a district court's reported view","single_row_fixes":{"since":"1.20.13 (2026-10-07)","rule":"six sentence shapes the 1.20.12 whole-table read listed as still asserting about a case that is not the verb's patient now read as not asserting: the cited case as the VENUE of an open relative clause that holds the verb ('Potter v. Froment, in which this doctrine has been repudiated', 'Miller v. Hurley, in which the contention there made was overruled'); a relative clause of ANOTHER case whose own verb takes the cited case as its object ('the panel opinion in Hernandez, which extended Bivens and which was repudiated'); the citation convention 'In X (1983), overruled Y (1986)', where the LATER case did the overruling (Y is the agent, X the patient); a treatment undone on the case itself ('part of Thomas was overruled by Crago, this portion has been restored to good law', 'the viability of those two cases was restored'); and a procedural subject whose complement is a common noun ('The action of the school board in discharging Mrs. Givens is overruled')","direction":"demotion-only: over the 93,901 stored citer rows 85 stopped counting and none started; every one of the 85 was read (the venue, the later case as agent, a derivative's overruling); no served tier moved — each was a single row under its target","measured":"the 1.20.12 panel's 751 unanimous assertions still count and its 31 refusals still stop; a true assertion the first cut lost ('by which cases the prior case of Beverley v. Brooke has been overruled') is pinned as a control with the six rows, and an independent review pass's 140 probes (a string cite's earlier captions, a bare short-form subject, 'wherein', a restore of something else) are pinned with their fixes"}},"why":"a consumer's own client read three certiorari QUESTIONS as the evidence for one landmark and negated sentences for another while the citer count was right (2026-09-23)"},"citability":{"since":"1.20.14 (2026-10-07)","refined":"1.20.17 (2026-10-10): two recall holes of the first build closed on the heads corpus — Alabama's own 'AFFIRMED … NO OPINION.' line is read whatever its separator (a semicolon, a colon, a comma, a dash) and with a scope parenthetical after AFFIRMED ('AFFIRMED (as to the appeal from conviction); NO OPINION.'), never inside quotation marks and never a writ denial, a rehearing overruling or a narration of another court's no-opinion affirmance; Tennessee's own 'MEMORANDUM OPINION' designation carrying its Rule 10 footnote CALL ('MEMORANDUM OPINION1') is read where the record's text as extracted dropped the footnote itself, never a workers'-compensation panel's memorandum or a trial judge's memorandum quoted in the opinion (80 records: 71 Tennessee, 9 Alabama — the last two the same Alabama line straddling the 1,200-character window edge). Indiana's amended Rule 65(D) was read at source (effective 2023-01-01: a memorandum decision issued on or after that date 'may be cited for persuasive value to any court by any litigant' but 'is not binding precedent'): the 2023+ class stays stamped_bar_unread with the court's own legend served verbatim until a status value for 'citable, not precedent' is decided","field":"data.case.citability on verdict_get_case / verdict_get_cases for secondary-corpus records of the covered courts (BETA)","rule":"a decision is served as non-citable ONLY when the ISSUING COURT'S OWN RULE bars citing decisions of its kind and the court STAMPED this decision as that kind — the stamp quoted verbatim from the record's text, the rule as read at source. Covered: Indiana memorandum decisions before 2023-01-01 (Ind. App. R. 65(D)); Tennessee memorandum opinions (Tenn. Ct. App. R. 10 — the court's capitalised designation AND its Rule 10 footnote); Oregon nonprecedential memorandum opinions (ORAP 10.30); South Carolina memorandum and unpublished opinions (the court's own MO / UP number or Rule 268(d)(2) legend); Alabama no-opinion affirmances (Ala. R. App. P. 53(d) / 54(d)); D.C. Memorandum Opinion and Judgment headings (D.C. App. R. 28(g)); California opinions stamped 'NOT TO BE PUBLISHED [IN THE OFFICIAL REPORTS]' with no later certification recorded (Cal. R. Ct. 8.1115). The Tennessee Court of Criminal Appeals (its own Rule 20) is NOT covered. 'Memorandum' and 'nonprecedential' are NOT detectors: the word means opposite things in different courts (a West Virginia memorandum decision is citable precedent; a Minnesota 'nonprecedential' legend is a permission); every marker is the court's own stamp on THIS decision, never a mention of the rule about another case","statuses":"stamped_non_citable (the stamp is read; the bar applies, subject to the rule's stated exceptions — law of the case, the same parties, …); stamped_superseded (the stamp is read but a publication of THIS decision is recorded — the record's own certification or publication order, a reporter citation the record carries, or the same docket's record filed the same day that carries the court's certification — named in `superseded`; the bar is NOT served); stamped_bar_unread (the stamp is read, the bar for that date was not read at source: Indiana memorandum decisions from 2023-01-01); no_stamp_read (no stamp in the head and tail of the record's text — NEVER 'citable': the stamp may sit elsewhere, the text may be absent, use-level and page-level bars are not read). A record of an uncovered court carries NO block","direction":"additive and demote-only: the block states what the court said about its own decision; it changes no verification label, feeds no accusation, and a missing block says nothing","measured":"the 2026-09-30 corpus (build r2c1, 2026-10-10): 545,526 records in the covered courts, 93,242 stamped non-citable (California 60,386; Indiana 14,281; South Carolina 13,141; Oregon 2,951; Tennessee 1,885; Alabama 598; D.C. 0 — its one heading in the build was ordered published), 1,169 stamps SUPERSEDED by a recorded publication of the decision (1,161 California: 396 by a reporter citation the record carries, 629 by the same docket's record that carries the court's certification of the opinion filed that day (59 of them partial), 132 by a certification in the record's own head, 4 modified re-posts of a certified decision; 8 Indiana by the record's own order) and 19 Indiana stamps dated 2023 or later. Three blind panels of independent judges on 2026-10-07 (207, 253 and 236 items; every rule, the superseded class and controls): the first two found the shapes the detectors then learned (a South Carolina opinion number in the body names ANOTHER decision; a publication order supersedes a legend; California's standard notice is not an order; a slip and its certified twin are two records); the third read 150 of 150 stamped records as the court's own stamp (Wilson floor 97.5 %), all 19 Indiana 2023+ stamps as stamps, 32 of 33 superseded records as the court's own stamp, and refused 34 of 34 controls; two adversarial review passes read 40 rows per rule and every supersession path against the opinion text, and ten certified docket twins read by hand were all the same case. On a partner's 15,564 served records in these courts the partner reproduced the Indiana count exactly (455). The round-2 blind panel (2026-10-10; 313 items: every rule, the superseded class, controls and the 78 records newly stamped by the first cut as a hidden stratum; three independent judges) read 120 of 120 stamped records as the court's own stamp (Wilson floor 96.9 %), the 78 new records 78 of 78 (floor 95.3 %; Tennessee 71 of 71), all 28 superseded stamps as the court's own, and refused 67 of 68 controls","limits":"the stamp is read in the first 6,000 and last 3,000 characters of each opinion (a stamp past that window is a no_stamp_read, never a wrong statement); a later publication order outside that window is not read; a decision later ordered published by a separate order the record does not carry still reads as stamped; the registry covers 7 of 53 jurisdictions' rules — 50 jurisdictions' citation rules are unread, so absence of the block says nothing about them; BETA: verify the stamp in the opinion before relying","why":"a partner's served set carried 455 Indiana memorandum decisions the issuing court itself marks non-citable; none carried a treatment signal, so nothing in the envelope said so (lane 4 of the currency measurement, 2026-10-07)"},"history":{"since":"1.20.15 (2026-10-08)","refined":"1.20.16 (2026-10-10): the reader of the higher court's tail was rebuilt over twenty-seven review passes (three Opus, twenty-four Fable, the last returning no reachable defect) after the 1.20.15 ship — a case REFERENCE in the tail is a property of the caption (another case's quash / disapproval / overruling never acts on this decision), the higher court's own stay / writ / transfer orders are set aside, 'for the reasons stated by the <court>' reads the court's reasons as ADOPTED (a named affirmance), an 'erred' read needs the citer as the asserting subject, an AGREEMENT reference ('like the Court of Appeals, we reverse the trial court') is the named court's affirmance, the citer's own verb counts against the named decision only on that decision (a lower court's or an agency's judgment, a conviction, an award is never its part; a SCOPE on the named object — 'insofar as', 'that portion of' — is a partial), companion causes and districts in opposite directions are set aside, a date is the patient's own, Indiana's scoped summary affirmance is the App. R. 58(A) vacatur of the rest; a tail CUT before the subject of a reversal beside a named affirmance is set aside (which judgment that reversal reaches is not read), the named court's act in ANOTHER case ('the decision of the Court of Appeals in Downey is reversed', 'In Downey, we reverse', New York's 'In People v Stanley: Order reversed', 'Affirmed in Hunt', 'in the third case, Travis v Batchi') is that case's and never this record's, an abstraction reversed ('the roles are reversed', 'the burden of proof is reversed') and a reaffirmed STATEMENT ('we affirm that either will suffice') are no dispositions, a parenthetical describing the patient ('(which affirmed that of the district court)') is not its disposition, a footnote marker glued to a period or a volume before a reporter abbreviation never misplaces a sentence boundary; the table rebuilt (r90 — same_court_round 164,927 / no_disposition 85,167 / not_this_decision 32,777 / reversed_or_vacated 29,415 / affirmed 38,341; the Federal Circuit's rounds over district records read again, agency and other-district patients set aside)","field":"data.case.history on verdict_get_case / verdict_get_cases for secondary-corpus records (REVIEW-only; BETA)","rule":"the LATER decisions of the SAME litigation among the record's citers, in a HIGHER court of the same system (or the Supreme Court of the United States), each served with its court, date, caption and the VERBATIM disposition sentences of its opinions, plus the detector's `read` of what that court did to THIS decision — served as DATA for the reader, never as a verdict on the record: `status` is always review, no verification label changes, nothing accuses. Reads: reversed_or_vacated (the higher court's tail names the intermediate court it reverses or vacates, in whole or in part; or it reverses the trial judgment this decision had affirmed); affirmed; no_disposition (a later round whose tails carry no readable disposition); not_this_decision (a candidate set aside, with the reason: a grant order, a stay or other self-vacatur, a judgment of another date, a later round of the record's own court between the two, a review more than 1,850 days later); same_court_round (a later round in the record's own court — listed, never read against the record: its disposition acts on the trial court). The same litigation is read from the caption (each side shares a distinctive word; initials match exactly; a sovereign's or an officer's opponent must be the same person; procedural words, roles and state names never bridge)","direction":"REVIEW-only and additive: the block is a reader's lead to the later decisions, read from the sentences shown and the decisions themselves before relying; an ABSENT block says nothing — the reversing decision often does not cite the reversed one, and a decision reviewed under a different caption is not found","measured":"the 2026-09-30 corpus (build r7, 2026-10-08): 350,627 rows over 300,467 records — reversed_or_vacated 27,800, affirmed 39,250, no_disposition 85,183, not_this_decision 33,327 (a candidate set aside with its reason), same_court_round 165,056 (listed within 1,850 days, never read). A blind panel of three independent judges over 110 served pairs, re-scored against this build: reversed_or_vacated 46 of 49 with none contradicted (3 unsure: trial-court records whose later judgment may differ), affirmed 14 of 14, the pairs the same litigation 100 of 100 on the read classes; four adversarial review passes (the third with six blind judges over 1,027 served rows, the fourth — the confirmation — with six over 521 and a last 160) read the served table by read AND reason; Wilson 95 % floors of the correct rate per class at 1.20.15: a read naming the reviewed court ≥ 0.87 (reversed) / ≥ 0.91 (affirmed), a trial-court record's reversal ≥ 0.87, a reversal without the court named ≥ 0.85 of the readable rows (bare 'Reversed.' lines need the reader's confirmation), reversed in part ≥ 0.83, affirmed in part or as modified ≥ 0.76; at 1.20.16 the cut-16 blind panel (536 items = 8 per moved class + 60 controls, nine independent judges three per chunk, re-scored against the shipped table r90): affirmed 127 of 140 agreed, 0 contradicted (floor 0.95); reversed_or_vacated 192 of 199, 1 contradicted — a trial-court record's affirmed-in-part / reversed-in-part row (floor 0.95); by reason — a read naming the reviewed court ≥ 0.92 (reversed, 68 of 69) / ≥ 0.93 (affirmed, 53 of 55), reversed in part ≥ 0.94 (56 of 56), a reversal without the court named ≥ 0.77 (25 of 29), a trial-court record's reinstated judgment ≥ 0.83 (19 of 19), bare affirmed ≥ 0.83 (28 of 33), affirmed in part ≥ 0.87 (25 of 26) / bare affirmed in part ≥ 0.77 (20 of 25); the standing panels 42 of 42 and 53 of 53 on the reversed read — overall the reversed_or_vacated read is wrong in roughly 2–3 % of rows and affirmed in roughly 2 %; the reads that inferred the effect from the target's own direction were wrong in a third of rows and were WITHDRAWN to set-asides; about 7 in 40 same_court_round rows with differing captions are other litigations; a partner's 66 reader-found misses: 56 cleared; RECALL against a partner's reader-found reversals about one third","limits":"courts come from the docket records of the corpus (a record without a court is matched only by the Supreme Court); the reviewing decision must CITE the record; same-court rounds are not read; a decision reviewed after more than 1,850 days is set aside; a tail naming only the trial court is NOT read against the record (both tails are served for the reader); the first panel read the affirmed class at 10 of 20 before the chain, 14 of 14 after; a reversal of a companion case named in the sentence under another record reads a bare reversal; the Maryland 'remanded with directions to reverse' form reads no_disposition; a sentence of the higher court's OPINION vacated ('we vacate the sentence above') reads a partial; BETA: never cite a record as reversed or as affirmed on this block alone","why":"a partner's served set carried decisions a higher court had reversed in the same litigation while no treatment signal said so (lane 1 of the currency measurement, 2026-10-07); founder decision 2026-10-07: review-only after the 66-case cut, improved over time"},"batch_usage":{"endpoint":"POST /v1/cases/batch (MCP: verdict_get_cases)","max_ids":500,"treatment":"since 1.20.3 treatment=true fans the SAME beta reliance-signal block verdict_get_case serves out to every FOUND record (data.cases[].case.treatment), billed at treatment_enrich per record a citation graph could compute (zero-rated principals: metered at list, charged 0). The ESCROW is taken at the ceiling — ids × (get_case + treatment_enrich) — and settles to found × get_case + data.treatment_enriched × treatment_enrich, so a 500-id treatment batch needs 2,000 credits on hand even when it settles far lower; verification.method names verdict_citation_graph when any record computed, and a treatment batch weighs twice a plain one against the rate bucket. Before 1.20.3 the batch omitted the block by design and a 62,726-record sweep needed 62,726 single calls on a shared rate bucket (a partner measured it at 35 hours). Every block names the graph window it derives from — the CURRENT secondary court-records corpus release for verdict:cl: records, the frozen corpus through the CAP window for verdict:cap: records — and a block with no signals is never clearance","id_grammar":"verdict:cap:<n> | verdict:cl:<n> — ASCII digits, no leading zeros, <=18 digits; non-canonical spellings reject the whole call with a 400 naming them (nothing is silently respelled, merged, or dropped)","dedupe":"exact-duplicate ids are merged; data.requested counts distinct ids","billing":"per FOUND record only — coverage gaps are never billed","replay":"an identical READ request replays the cached envelope only within 1m (transport-retry dedupe); after that it RE-EXECUTES against current data and bills again — reads stay fresh across corpus ships. Verification tools (citations/quotes/briefs/ai) replay for 24h. EVERY replayed envelope carries meter.replayed:true — when verifying that a corpus change landed, a replayed envelope is not evidence; vary the request (or wait out the window) to force a fresh execution"},"delta_fold":{"since":"1.20.25","coverage_statement":"secondary corpus release 2026-09-30 plus a weekly delta fold; the fold's reach is served as coverage.corpus_recent_through on every envelope that carries a coverage block (the envelopes the delta_fold 'when' entry names) and on /healthz as cl_delta_fold_date","what":"a WEEKLY fold of the secondary court-records corpus's daily delta (the records it created after its quarterly release: the record id, the court of the record's docket, the court its caption names where the source read one, the case name, the filing date, the docket number and the record's citations) into the local verification index — no citation graph, no treatment, citability, history or disposition lane: those stay quarterly","when":"cut on Sunday mornings from the delta days published through the Saturday run; the fold's reach is an INSTANT, records_created_through (UTC): per walk family (nightly, historical, the closed seam window) the end of its latest walk — a completed walk's end, a partial walk's cursor — and the MINIMUM over the open families; the fold reaches only the walks that ran (a family absent from every folded day is named absent in the fold's own record, never silently treated as complete); the fold holds the records the source CREATED up to that instant (a record created later is not folded) and a DELETION at the source after the fold is not detected. coverage.corpus_recent_through is the LAST COMPLETE UTC DAY of records folded: the calendar day before the instant's UTC date, unless the instant is at or after 23:59Z of its day; absent = no fold is served. It is carried by exactly these envelopes: verdict_check_citations, verdict_check_brief, verdict_check_quotes, verdict_get_case / verdict_get_cases for verdict:cl records, and the three case-law search envelopes (verdict_search_catalog, verdict_fetch_cluster, verdict_lookup_case — where data.data_recent_through is the search service's own recent index, a different date). verdict_coverage (data.corpus) and /healthz state the same day as cl_delta_fold_date and the instant as cl_delta_fold_created_through; each folded resolution carries records_created_through, and fold_date there is the latest delta DAY folded (a day name, not the reach)","order":"the one guarded case lookup consults the fold only when every quarterly step (the primary index, the nominative form, the secondary-corpus rescue, the SCOTUS supplement) left a coverage gap, a not-found OR a quarterly ambiguity (whose candidates the fold's records then join), and BEFORE the recency lookup service — a citation the fold answers is never sent to the recency lane; a citation matches the fold only on its exact key (a suffixed page such as '1258(A)' or '240578-U' never verifies on the bare number)","served":"a hit is verified_exists exactly as a secondary-corpus hit, with resolution.source 'delta-fold', canonical_id verdict:cl:<id>, records_created_through, fold_date (the latest delta day folded), delta_day, court_id, docket_number (caption_court_id where read) and evidence naming the fold; verification.method carries verdict_cl_corpus; data.summary.verified_delta_fold counts them; the attribution lane reads the record's case name and filing date as it reads any secondary record (a year disagreement is review, never an accusation); a treatment block on such a record is NOT computed (it postdates the quarterly graph) and names the corpus lower bound as its window","several_records":"records are ONE decision recorded more than once only when caption, court, filing date, docket number (equal, or null on both) AND their own citations agree (equal, or one set extending the other where a non-null docket number agrees — the delta adds citations later) — such records are served once, the lowest id, the others named, and a precedential status they disagree on is served as 'disagree: A | B'; distinct decisions at one citation are never picked between: a PARALLEL leg of the same citation that resolves to exactly one of them settles the whole citation (one canonical id per parallel citation) when the caption written names that record or names none of them — a caption naming OTHER records than the leg is the review, the leg's record named; else the caption written is read against them by ONE rule shared with the recency lane's read of a citation several of its records share (that read CHANGED with 1.20.25: it reads the caption by the same identity-word rule) — a TWO-PARTY caption names a record only when EACH side matches that side of the record (either party order), by the kind of the record's side: a PERSON side ('Spencer Lee Jefferson'; the person before a comma when the rest is a role — 'Abby Leigh Atkinson, as Personal Representative …') matches on its SURNAME, and a written state name against a person's side is a surname ('People v. Washington') — a bare place reads as the State only against a government side, and the court-state filter applies to government sides of STATE courts only (never a federal court) ('Jefferson v. Scott Equip. Co.' names 'Spencer Lee Jefferson v. Scott Equipment Co., LLC'), and a given name or initial written must agree with one the record states (a different one contradicts, an absent one does not — 'Anthony T.' is not 'Anthony R.'); an ORGANISATION or GOVERNMENT side matches only when EVERY identity word of both sides is named by the other — generic party, organisation and procedural words ('LLC', 'Ins.', 'Co.', 'Holdings', 'Bank') set aside, abbreviations read either way through the one Bluebook table (with the T6/T10 forms 'Cnty.', 'Twp.', 'Vill.', 'N.Y.', 'N.Y.C.', 'N.J.', 'U.S.', 'Dept.', 'Preserv.', 'Hosps.', \"Comm'r\", 'Auth.', 'R.R.', 'Is.', 'Natl.'), an acronym naming the words it spells ('MTA', 'LIRR', 'HPD'), either name of a 'd/b/a' — so 'Zorvath Side Marquis LLC' never names 'West Side Marquis LLC' and 'Capital Grp.' never names 'Capital One Bank'; a government side may leave out the place it names ('State' names 'State of New York'; 'Bd. of Educ.' names 'Board of Education of the City of New York') but never names another government ('City of New York' is not 'State of New York') (never in reversed party order when a side is a government — 'State v. Driver' never names 'Driver v. State of New York'); a word the caption writes that the record does not carry ('Bank', 'Dept.') never matches; an agency ('City of N.Y. Dept. of Educ.') is not its city; a hyphenated name is one word ('Lackey-Garcia' is not 'Garcia'); a party in brackets ('Super. Ct. (Garcia)') is not the side's own name; a side written only in initials ('R.R.', 'J.D.') names the same initials; a side of generic words alone ('People') matches only beside a side that matches by identity words; a placeholder ('Redacted', 'Anonymous') never; a side the row reader cut at an abbreviation or inside a street address ('Enters.', '211 W. 14th Prop. Owner') is read from the caption written whole (the caption walk, else the span back to the sentence start when it ends at the citation's comma); a later parallel leg is read with the caption its first leg was read with, whichever order the legs are written in; every review quotes the caption as read; before a caption is read, a citation signal or a narrated court is read away (See, Accord, Contra, Compare … with, But cf., Under, Unlike, Following, Pursuant to, 'the Appellate Term in'), a sovereign's full style reads as its short form ('The People of the State of New York' = People, 'State of Ohio' = State, 'CNMI' = Commonwealth, 'United States of America' = United States), a role phrase (', as Trustee …', ', individually and …') is not the party's name, and the abbreviations of the whole Bluebook Table T6 read either way; a ONE-PARTY caption ('Matter of X', 'In re X') names a record only when its significant words (the generic party, organisation and procedural words set aside) are the record's own — a one-party record's, or the first side of a proceeding caption ('Matter of X v. Y') that is no sovereign — never on a shared word: it settles on the equal full name (abbreviations read through the Bluebook table) or the record's own significant words, never on a placeholder ('Redacted'); the recency lane reads a one-party caption by the same rule and never takes the first of the records it shares; a citation the QUARTERLY corpus already reads ambiguous gains the fold's records as candidates under the same read (a nameless quarterly record the fold names takes the fold's name; a quarterly record and a folded record of ONE decision — the same case name and filing date, the folded record's citations including the quarterly record's (the delta adds citations later), e.g. one slip opinion in both corpora — count once and verify the quarterly record, the folded one named). One match — folded or quarterly — verifies; several distinct matches read ambiguous_multiple_matches; no match reads the merged review when the quarterly corpus already read the citation ambiguous, and otherwise (the quarterly corpus left a gap) goes to the recency lookup service, unless that lookup could read only part of the page (a '(A)', '(U)' or '-U' suffix); when that lane is off, answers nothing or would misread the page — and for a not-found row, which is never sent to it — the citation reads ambiguous_multiple_matches. The review names up to 8 candidates, the caption's matches first (a candidate the caption names is always listed), candidates_total the count, and says which of three reads applied: no caption could be read, the caption names N of them, or the caption as read (quoted) names none. A record settled on such a page (a parallel leg) whose caption the attribution lane finds naming another case reads that review — 'N records share this citation; the caption names none of them' — never the accusation that the citation resolves to the other case; on a parallel citation the caption is compared against the FIRST-written leg's record only, as on any secondary-corpus record: a single-record first leg whose caption names another case raises case_name_mismatch — unless another leg's record names the caption written, when no attribution flag is raised (the lane reads the caption as that record's; a missed accusation, never a false one); a later leg is not compared; ONE decision recorded more than once at a citation (a quarterly record and its fold twin) is served as that decision whatever the caption says — the attribution lane reads the caption — and a parallel leg reaching the twin — the fold's record, or the quarterly record of one of the folded decisions — settles the citation and is served as the same id (one id per parallel citation) ONLY through the fold's own twin map or a citation the two records share that EXACTLY ONE decision holds (a citation the source assigns to several records is each one's own; a multi-decision table page such as one '(A)' page of several slip opinions, or a shared slip number, never counts) — never on name, date or court; otherwise the legs are served as each resolves (two ids); two citations of ONE reporter joined by a comma, and two NEUTRAL citations of any series (a decision carries one neutral citation: '2026 IL App (4th) 250954-U, 2026 IL App (3d) 250063-U'), are a STRING CITE of two decisions, never one parallel citation — each verifies as its own record, as two quarterly legs always have, and a string-cite leg without a caption of its own reads the same sentence's caption (its first leg's) for the caption pick; two citations of one reporter (not both neutral) join only when one leg's own record carries the other leg's citation and that citation is held by one decision alone; two legs in DIFFERENT reporters (not both neutral), at least one resolved THROUGH the delta fold, on records no such citation joins that DISAGREE — different courts (a quarterly record's court, which the quarterly corpus does not serve, read from the jurisdiction of its own reporter), different filing dates, or different docket numbers (a placeholder docket — 'Redacted', 'Sealed' — is unread) — where the parallel mismatch cannot fire (a shared name word or date), read manual_review_required ('the legs of this parallel citation resolve to two records that share no citation of one decision alone', both records named with court, date and docket and the field that disagrees) — never two clean verifications, never 'the same decision'; a case_name_mismatch on the first leg stands beside it (the review is the pairing's, never in place of the accusation); verdict_get_case(citation) on one decision recorded twice serves that decision and names its other record, as check_citations does; two QUARTERLY legs are read exactly as before 1.20.25 (neither the review nor one id); the proceeding opener ('Matter of', 'In re', 'Application of') is never a party's or an agency's word; a recency answer whose records include one without a case name, with no caption read, is the review counting every record","attribution":"on a fold record the caption is read as on any secondary-corpus record (the attribution lane's token comparator, unchanged; a reversed caption stays the caption_order review); among several records sharing a citation the caption settles one only by whole identity, never on a shared word. A caption keeping one true party, or only another record's generic words, on a SINGLE-record citation — fold, quarterly or CAP — can still read verified: a filed limitation","quotes":"verdict_check_quotes and verdict_check_brief: a quotation whose cited source verifies through the delta fold or the secondary corpus — and whose text the recency lane did not read — is NOT CHECKED (its text is not held): manual_review_required with quote_not_checked_reason 'source_text_not_held', counted per reason — summary.source_text_not_held / source_ambiguous / source_not_read, their sum summary.not_checked_source_total — never 'not resolvable' and never a coverage gap; a source several folded decisions share reads the review (quote_not_checked_reason 'source_ambiguous'). Such a quotation DEMOTES the brief exactly as an unresolvable one does (partially_verified at best, completeness partial). The recency lane's text is read for a quotation only when it answers the record the citation lane settled, and a page with a '(A)', '(U)' or '-U' suffix is never sent to it","parallel_mismatch":"the parallel mismatch ACCUSATION (parallel_cite_mismatch, label class extra_accusation) now also fires on two legs that are not CAP records, under this exact gate: both legs verify on named secondary-corpus or delta-fold records, at least one through the delta fold, the two records share no name word, their dates differ, and the first leg's record carries its own, different citation in the second leg's reporter; for two quarterly secondary-corpus records the check stays silent, as before, pending a measurement; a CAP leg beside a fold leg follows the CAP rule (a CAP record on both legs), so that pair stays silent. Measured: 1,994 correct parallel citations built from the fold records' own citations emit 0","read_failure":"a fold that cannot be READ for a citation (disk I/O, a lock, an unopenable or malformed file) is not examined: an accusation there reads manual_review_required ('the delta fold could not be read for this citation; not examined'), a gap stays a gap, and that envelope states no corpus_recent_through when EVERY fold read of the request failed (when some succeeded, the field stays — the fold's reach for the citations it read — and summary.fold_unread (check_citations / check_brief) counts the citations not read; a verification note says which, and the brief certification adds 'N citation(s) could not be read against the delta fold (retry)'); a malformed served file is withdrawn. Every reader says so: verdict_get_case answers the structured internal error (REST: HTTP 500, kind internal, a request id; MCP: the SDK withholds the text — the caller sees 'Error executing tool verdict_get_case' and retries; never not-found), a verdict_get_cases entry carries that note, a case-law search result reads 'verification not read (delta fold unavailable)', and a quotation's source reads not read (quote_not_checked_reason 'source_not_read', never a coverage gap); verdict_get_cases (data.fold_unread counts the ids not read) and the three case-law search envelopes read the same rule: corpus_recent_through is dropped only when every fold read of the request failed","brief_status":"since 1.20.25 verdict_check_brief demotes on REVIEW-class case citations (ambiguous_multiple_matches / manual_review_required) as verdict_check_citations does — a brief whose only citation is ambiguous reads unverified, completeness partial; before, it read verified while its own check read unverified (a status honesty fix, no label change); verified statutes, regulations and rules count as something verified beside such a finding, as they do for the check (partially_verified); a flagged brief that verified nothing reads unverified, as its check does","review_marker":"when the record's docket names one court and its caption another (court_conflict), the citation reads manual_review_required with court_conflict: true, court_id and caption_court_id — the record exists; which court decided it is unsettled in the source; it demotes the document and never accuses","miss":"a citation the fold does not hold falls through to the recency lookup service exactly as before 1.20.25; a miss in the fold is a coverage statement, never evidence that the case does not exist","records":"verdict_get_case / verdict_get_cases serve a folded record by its verdict:cl id (source 'delta-fold'; its BETA blocks are empty — no lane is built over it yet, never 'none exists'); an id the source RETIRED by a merge answers its single survivor with data.alias {requested, survivor, reason, fold_date} — an alias, never a not-found; a retired id whose opinions went to SEVERAL records is refused naming them (never a pick)"},"retrieval_resale":{"since":"1.20.22","status":"BETA (currently testing)","tools":{"verdict_search_catalog":"POST /v1/search/catalog","verdict_fetch_cluster":"POST /v1/search/cluster","verdict_lookup_case":"POST /v1/search/lookup"},"lookup":{"since":"1.20.23","what":"a keyed case LOOKUP by exactly one of citation, docket or name over EVERY case in the search service's bulk build — including about 2.1 million cases the catalog search cannot rank (2,124,954 of the 10,174,551 cases in the 2026-09-30 bulk, the search service's count on release r3, 2026-10-26: opinions too short to embed and courts outside the search partitions); a lookup, never a relevance ranking: use it to find a case you can name, the catalog search to find cases about something","fields":["citation","docket","name","name_mode","court_ids","date_after","date_before","jurisdictions","limit"],"limits":{"exactly_one_of":["citation","docket","name"],"key_max_chars":{"citation":200,"docket":100,"name":200},"name_mode":["all","phrase"],"name_mode_rule":"a name lookup only; the search service's default is 'all'","court_ids_max":500,"jurisdictions_max":10,"limit":"1-100 (the search service's default 20)","dates":"date_after / date_before as YYYY-MM-DD"},"result":"each result as the store returns it — cluster_id, case_name, case_name_full, court_id, date_filed, precedential_status, citation_count, docket_number, citations[], partitions[] (every partition that lists the case), served[] (the partitions with searchable text; EMPTY = the case exists but has no searchable text) — PLUS `verification` exactly as on a search result; top-level `by` (the key that answered) passes through, and `recent` passes through when the search service sends it (not in RETRIEVAL.md §2 for /lookup as of 2026-10-26)","gaps":"coverage.gaps always says: no match is never a statement that no such case exists; a retired cluster id is not yet redirected to its survivor (the search service adds that at the next quarterly)"},"what":"case-law search served through api.verdict.io by the Verdict ecosystem's search service: public court opinions from the secondary court-records corpus (the search service's corpus build, folding and index), rebuilt quarterly from that corpus's bulk data, plus a daily RECENT index of cases added since the release — data.data_release names the release and data.data_recent_through the newest delta day the recent index holds (null when none); NOT real-time; data.source names the provider in the envelope","coverage":"the 50 states + D.C. (state courts and federal district / bankruptcy courts), the federal appellate courts, and per-circuit federal district partitions","request":{"search_fields":["query","vector","mode","jurisdiction","jurisdictions","k","offset","keyword_terms","boolean_query","date_after","date_before","published_only","exclude_non_citable","court_ids","scopes","fold","opinion_types","author_ids","author_name","author_inferred","case_name","case_name_mode","docket_number","nature_of_suit","include_facets"],"limits":{"jurisdictions_max":5,"k":"1-50","offset_max":150,"offset_plus_k_max":200,"query_max_chars":2000,"vector_length":768,"cluster_ids_max_metadata":1000,"cluster_ids_max_text":10,"entry_chars":{"jurisdiction":64,"court_id":64,"keyword_term":100,"opinion_type":32,"nature_of_suit":100},"opinion_types_max":12},"cluster_ids":"strings of ASCII digits (the secondary court-records corpus's cluster ids); text defaults to true","unknown_fields":"a request field outside search_fields (or outside cluster_ids / text on a fetch, or outside lookup.fields on a lookup) is REFUSED on REST (HTTP 400, nothing sent, nothing charged) and is dropped by the MCP argument model before the tool runs — a note, a document or an account field is never read and never forwarded","boolean_query":"\"quoted phrases\", bare words, uppercase AND / OR / NOT, parentheses; keyword or hybrid mode only"},"carve_out":"ONLY the query text, filters, facets and paging (or the vector you embed yourself, or the cluster ids, or the lookup key and its filters) are sent to the search service — never your identity, account, documents, or anything else; verification runs on Verdict's side","in_transit":"In transit, search queries pass through a CDN / security provider that terminates TLS in front of the search service; it is configured with no request-body logging and no caching of searches.","result":"each result as the search returns it — cluster_id, opinion_id, case_name, court_id, date_filed, citation_count, precedential_status, chunk_text (the matching passage), score, folded[] (duplicates / same dispute folded into it), citability (the issuing court's own non-citability stamp; null = nothing said, NEVER 'citable'), the facet fields you opted into — PLUS `verification`: Verdict's block for the cluster where it is in the verification corpus (canonical_id verdict:cl:<id>; existence; treatment — get_case's block, clearance always false; citability; history = later decisions, REVIEW-only), or null with verification_note 'outside the verification corpus' (never a statement about the case)","notes":["BETA (currently testing): case-law search served by the Verdict ecosystem's search service; the results are the search service's, the verification blocks are Verdict's.","Freshness: the corpus is rebuilt quarterly from the secondary court-records corpus's bulk data (not real-time); data_release names the release this answer came from; data_recent_through names the newest daily delta day the search service's RECENT index holds (a search may return cases added after the release; null = no recent index served).","Coverage: the 50 states + D.C. (state courts and federal district / bankruptcy courts), the federal appellate courts, and per-circuit federal district partitions.","Facet coverage is partial (an author id on a minority of opinions; dissents only when filed as their own opinion; nature of suit mostly federal district; docket numbers on most cases): a null facet value means the bulk data records nothing, never 'none'.","Two dates: data.data_recent_through is the search service's RECENT index; coverage.corpus_recent_through (when present) is the VERIFICATION corpus's weekly delta fold — the last complete UTC day of records it folded, which the per-result verification blocks read.","'Nothing found' is never 'no such law': an empty result means no match in this corpus.","No result is 'good law' because the search returned it: the per-result verification block (existence, treatment signals, citability stamp, later decisions) is Verdict's, served as evidence and never as clearance; absence of a signal is not clearance."],"prices":{"catalog_search":10,"cluster_fetch":2,"cluster_fetch_text":5,"case_lookup":2,"unit":"credits per call (one page of search results; one batch of clusters; one case lookup)","verification_included":"the per-result verification block — existence, the treatment SUMMARY (get_case's block), citability, later decisions — is included in the call price for every result: no per-result add-on, and get_case's treatment_enrich row does NOT apply here","note":"these figures are today's price table (pricing.current_table); an early-access account within its lock-in is charged by its own table — GET /v1/account data.pricing names it, and every charge names it in meter.price_table"},"caps":{"per_account_daily":200,"global_daily_pool":5000,"rule":"every search-service call (search, cluster fetch or case lookup — a lookup counts against the SAME two counters as a search) counts once against your account's per_account_daily ceiling and the service-wide global_daily_pool, per UTC day, counted when the call is sent to the search service (a search-service failure still counts; a replay inside the 1-minute read window does not); ecosystem (zero-rated) principals count against both caps like every account — the pool protects the search service, not the ledger; past either: HTTP 429 error.kind capacity_reached with Retry-After to 00:00 UTC, nothing charged"},"envelope":"verification.status is not_applicable on every search-service answer (a search is not a verification of a document; the evidence is the per-result verification block, and verification.labels is always empty); coverage.completeness is always partial (one page of one index, never the law), coverage.jurisdictions_searched echoes the request, coverage.corpus_as_of is the data release's date, and coverage.gaps names a keyword-budget timeout (the page is INCOMPLETE) and an empty page (never 'no such law')","errors":{"partner_unavailable":"HTTP 503: the search service could not answer (busy, down, timed out); never charged; error.retry_after_s (also retry_after and the Retry-After header) says when to retry — never read it as an empty result","capacity_reached":"HTTP 429: a case-law search day cap is spent; never charged; Retry-After to 00:00 UTC","not_configured":"HTTP 503: the case-law search service is not configured on this server; never charged","invalid_input":"HTTP 400: the request broke a limit above, or the search service refused it (the reason quoted); never charged"},"never":"a result is never 'good law' because the search returned it; 'nothing found' is never 'no such law'; a facet is never complete (a null value = the bulk data records nothing)"},"response_matching":{"raw_semantics":"data.citations[].raw depends on the REQUEST MODE. Text mode: raw is the eyecite-CORRECTED citation string ('606 U.S. ___ (2025)' in prose comes back '606 U.S. ___') and rows carry a span into YOUR text — match by span there. List mode (citations=[...]): raw is your input string VERBATIM and there is no span — equality against your own list is safe there. Text-mode shape-scan advisory rows carry the harvested original substring. The failure to avoid: equality-matching TEXT-mode raw against stored strings silently mis-buckets answered citations as unanswered (consumer-measured, 4 cites in one sweep).","mechanism_vs_source":"resolution.source names the CORPUS OF RECORD ('cl-corpus'; 'scdb-supplement'; 'delta-fold' since 1.20.25; absent = CAP, except tier-2-upgraded rows which carry tier:2 and a verdict:cl canonical id with no source key) — it does NOT name the deciding MECHANISM. Duplicate collapses and SCDB arbitration ride the corpus row they resolve and are named only in the evidence string, so crediting mechanisms from source under-reports them (a consumer measured an arbitration lane as 0 wins that way).","ambiguity_candidates":"an ambiguous_multiple_matches resolution carries candidates[] (each with canonical_id where the record is retrievable), candidates_total, and candidates_truncated — large collision sets list the FIRST candidates only, so a name-in-candidates gate must treat candidates_truncated:true as 'candidates may be missing', not as an exhaustive roster. A correct citation to a LISTED case should never be rewritten away."},"advisory_labels":{"rule":"a labels[] entry with advisory:true is INFORMATIONAL — never a fabrication claim, never a blocking signal. Current advisories: shape-scan entries; case_name_variant (a near-identical caption spelling or a short form of the record's words — edit distance 1, a known OCR confusion, an acronym or a contraction — most often a typo: verify the spelling, never treat as a wrong-case claim); and citation_year_later_than_decision (a text year 2-3 years AFTER the recorded decision date — the amended/reissued-opinion shape; the year the citing court itself used is not a fabrication); and, since 1.16.0 with propositions=true, supports (a lexical match of what the sentence says a case stands for with a court-authored description — moved OUT of label_classes.success at 1.16.0: a clean verdict never rests on it). related_insufficient stays in the ADVISORY vocabulary for stored records but has not been emitted since 1.16.3 (blind panel: the non-match was a paraphrase two times in three); it lives on only as data.citations[].proposition.verdict"},"provenance_naming":{"rule":"1.13.0 (founder decision 2026-09-06): served ENVELOPE prose names the secondary court-records provider wherever a record, a date or a verification comes from it — evidence strings, notes, the coverage report's recency line — so a certification quoted in a filing carries its attribution. Public surfaces (this guidance, the manifest, tool descriptions and the instructions string, site copy and docs) do not name it pending the data agreement; they say 'the secondary court-records corpus' / 'the recency lookup service'. Opaque identifiers are unchanged: verdict:cl:<id> canonical ids, date_source 'cl-corpus', lane names. verification.method carries 'recency_lookup' (the one schema value; the pre-1.12.0 value is normalised to it wherever a stored record is validated) only when a tier-2 hit VERIFIES a citation; a consult with no match, or a record with no caption, adds a note and not the value. Since 1.20.16, when the recency lookup service holds SEVERAL records under one citation, the caption written in the document is matched against each record's caption: the record it names verifies with the evidence saying so, and a caption naming none of them is ambiguous_multiple_matches with every candidate named — never a case-name mismatch finding and never the first record by default (a partner measured the old first-record read as an accusation on a correct citation, 2026-10-08). Secondary-corpus case records carry full_text_url null + text_availability 'not held locally' (their opinion text is not held here). negative_treatment_language signals served under the engines 'graph-v1-cl+language-v1' (secondary-corpus targets) and 'graph-v1+language-v1-fused' carry, per citing_opinions[] entry, the citer's caption (the current record's when the mined snapshot had none — the opaque id only when no record carries one) AND canonical_id; the CAP language lane's entries carry caption+date+quote without canonical_id; cautionary_treatment_language entries carry treatment+quote (no citer identity). A tier-2 record with NO caption never verifies a citation OR a quote's source: the row stays a coverage gap / manual review and a note counts it","stored_records":"verification logs, reports and IDEMPOTENT REPLAYS are served AS ISSUED (the audit contract): records are never rewritten. Records issued under 1.12.0-1.12.7 use the neutral descriptors in their evidence; earlier and later ones name the provider — both read true. A replayed record without audit.guidance_version (stamped since 1.12.2) says so in verification.notes; its audit.request_id is the as-issued record at /v1/logs"},"post_finalize_failures":{"rule":"if a verification ran and was charged but its stored envelope cannot be served (a drifted enum) — on the original call OR on an idempotent replay of it — the answer is STRUCTURED: error.kind 'internal' with the request_id (MCP: an isError result carrying the envelope; REST: HTTP 500 kind internal), and the charge is refunded to your balance (Terms §5: a failed execution attributable to the Service). ONE contract (1.12.7, refined 1.13.1): when the refund is SETTLED — credited now, credited earlier, or nothing was ever held — meter.credits_charged is 0, the message names the amount credited back when credits were held ('no credits are held for it' otherwise), and a retry re-runs the verification (a refunded record never replays; it may fail the same way until the fix ships). When the refund could NOT be settled (the ledger was unreachable) the envelope carries the charge the ledger holds; the message states how much of the tool's replay window is left (24h for verification tools, 60s for reads, from the recorded call): inside it an identical retry replays this answer and re-attempts the refund, after it a retry re-runs and is charged, and support is named ONCE with a handle that resolves (the request_id, else the ledger entry) for the refund of that charge — the only case to escalate (1.13.2: no automatic sweep refunds it). On a replay the error envelope carries meter.replayed true. The record is retrievable as issued via verdict_verification_log(request_id) whenever the message says so (the ledger's request_id, never a corrupted record's — when no id can be trusted audit.request_id is empty and the message names the ledger entry); the log's credits_charged says what the call charged and its state 'refunded' says the credits came back. A stored record that no longer decodes or is not a record is served through the same lane: refunded, error.kind 'internal', not retrievable, not counted as a call, and the NEXT retry executes fresh. If the ledger cannot RECORD a finished verification: a record that did land is served as issued (without balance_after); a call that outlived its escrow lease (30m; every running call refreshes its lease every 5m, so a live call is not reaped while the ledger is reachable, up to an absolute ceiling of 6h from its start — the lease never extends past it; 1.14.0: ONE predicate decides — an escrow whose lease is alive can be extended or settled, an escrow whose lease lapsed cannot, whatever the reason, and the ledger returns the estimate the moment it refuses, so a lapsed lease never charges; the sweep releases anything the refusing lane did not, within about 10m; a synchronous call completing later is served free, a large brief whose report was already failed is not re-opened — re-submit) is served FREE with meter.credits_charged 0, a basis and a note saying it is not in the verification log and that an identical resend re-runs and is charged; otherwise the answer is error.kind 'internal' with credits_charged 0 when the estimate was released, the actual charge when the record was written but could not be confirmed, or the held estimate (basis 'estimated_escrow') while the escrow is still held (when the ledger shows no escrow but the record cannot be read, the message states both readings and credits_charged is the actual charge at most; when the ledger could not be read at all the finalize and engine lanes carry the held estimate with basis 'estimated_escrow — unconfirmed', and a stuck report answers a transient view that names no figure) — a held estimate is released by the ledger sweep or on an identical retry after the lease, within about 40m, and an identical retry inside the lease answers idempotency_conflict. A large brief whose record could not be written — or whose engine failed — is a FAILED report whose result is that envelope (never a traceback); a large brief whose job could not be started answers error.kind 'internal' at submit with the escrow's true state (released, or held until the sweep) and no report_id; a report still pending 30m past the ceiling resolves from the escrow's ledger row: a record the runner had written is served as the COMPLETED report (its charge stands); anything else reads as failed with the escrow's state (released; or held — released by the sweep within about 10m or at once by an identical retry, which re-runs; a written record nobody can serve is REFUNDED by the sweep); the ledger sweep persists that resolution on its next tick and reading a report never writes or moves money; a terminal report state never changes (an unreadable ledger answers a transient view carrying a stable request_id that names no record); a job that fails outside its own handlers is a FAILED report carrying that envelope too; failure envelopes carry list_credits for zero-rated principals where the ledger row records it, and insufficient_credits envelopes the tool's own unit; a result finishing after the 6h ceiling is never charged (the ledger refuses to settle it); the job lane bills exactly as the synchronous lane (class multiplier, rate bucket, text cap); a failed replay is not counted as a call. Failure results of the compat tools search and fetch — returned OR raised (not found, bad id) — keep their output shape (results: [] / a placeholder document) with the error, audit and meter blocks beside it and isError set; error.kind gains not_found and invalid_input as schema values for them. REST validates typed arguments against the tool's schema before any charge (400 invalid_input), exactly as MCP clients are validated"},"argument_parsing":{"rule":"1.12.5 (moved here in 1.12.6): a string sent for a string-typed argument (text, citation, canonical_id, jurisdiction, as_of, request_id, …) is taken AS THAT STRING on both transports — a document that happens to be JSON (an LLM's structured output, the literal 'null') is verified as text, never decoded. A JSON-encoded string for a list argument (citations, canonical_ids) is still decoded for clients that stringify lists"},"rest_error_statuses":{"rule":"REST maps a caller error to HTTP 400 kind invalid_input; HTTP 404 kind not_found ONLY for a missing resource the caller addressed: GET /v1/cases/{id} (case not in the corpus), GET /v1/cases/{id}/text (verdict:cap ids only — a verdict:cl id answers 400: its opinion text is not held locally), GET /v1/reports/{id} and GET /v1/logs/{request_id} (not this account's), DELETE /v1/account/keys/{prefix} (no such active key). Text you supply inside a citation or as_of never changes the status (since 1.11.7; the 404 body kind is not_found since 1.11.8 — earlier builds said invalid_input with a 404)"},"account_surfaces":{"billing_mode":"since 1.20.1 GET /v1/account (verdict_account) states billing_mode: 'test' | 'live' — the deployment's declared Stripe mode, which the secret key must agree with or billing answers 503 and nothing is sold or set up. In 'test' a Checkout purchase moves no real money and the credits it grants are TEST-MODE credits; in 'live' a purchase is a real charge (billing address collected at Checkout, Stripe Tax when the account has it on). An event whose livemode disagrees with the mode grants nothing and records no card; ecosystem (zero-rated) principals are unaffected either way","free_tier":"since 1.20.19 (2026-10-21): an oauth account receives 100 credits ONCE, at first sight (signup_credits; 500 before 1.20.19). From the first calendar month after signup, the balance is raised back TO 500 credits (refill_floor) once per calendar month — never subtracted from, a balance above the floor untouched — ONLY while a payment method is on file (card_on_file). Without one, nothing is refilled and nothing is removed; the month's refill applies at the first call after a card is saved. GET /v1/account data.free_tier states signup_credits, refill_floor, refill_requires_card, next_refill_period (YYYY-MM; null where the lane has no refill), card_setup (the route below), monthly_floor_credits (kept for 1.20.18 readers: the floor on the oauth lane, 0 elsewhere) and, when no refill can apply, refill_blocked_reason ('no payment method on file'; that the principal's lane has no refill; or that the refill floor is 0 on the deployment). A zero-rated principal needs no card (refill_requires_card false); API-key and service principals have no refill. Accounts that existed before 1.20.19 keep their balances; their next calendar month's refill requires a card. Both numbers are read from the live pricing table","card_on_file":"since 1.20.19 GET /v1/account data.card_on_file true|false and data.card_on_file_since (ISO 8601 UTC, or null): the account has a saved payment method on its Stripe customer — recorded when a Stripe Checkout in mode setup completes on the account's customer with the card saved, or when a purchase completes (the purchase Checkout rides the same customer and saves its card there; a purchase proves a card). card_on_file_since is the first recording since the last clearing; a later session moves nothing. CLEARED (back to false, null) when a payment method is detached from the customer and none is left, or the customer is deleted — the balance is never touched, and the next refill waits for a new card. Nothing else sets or clears it","setup_session":"since 1.20.19 POST /v1/billing/setup-session (REST, authenticated like POST /v1/account/checkout, rate-guarded, free) answers {url, mode: 'setup', billing_mode}: a Stripe Checkout that saves a card on the account's Stripe customer (the one the purchase Checkout uses) and charges NOTHING (Stripe verifies the card). Open the url; on completion card_on_file reads true within the webhook's delivery, and a refill that was due applies at once. 503 not_configured when billing is not configured or the key disagrees with billing_mode. Not an MCP tool (the payment checkout is not one either)","class":"since 1.19.1 GET /v1/account (verdict_account) states the principal's CLASS: account_class standard | discounted | internal, zero_rated true|false, and class_note in words. A zero-rated (ecosystem) principal is metered at list price on every call — meter.list_credits — and charged 0, so its credits_balance is a budget that never moves and the funded rate tier applies whatever it reads; a zero balance on such an account has never meant 'no credits' (a partner read it that way, 2026-09-09). Partner principals were granted a visible budget the same day","pricing":"since 1.20.26 GET /v1/account (verdict_account) data.pricing states the account's price table, true by construction from the resolver its charges read: price_table (the effective date of the table this account is charged by), early_access_account (true|false — created before general availability; every account while it is not announced), locked_until (null until general availability is announced, and null for an account that is not early-access; otherwise the last UTC day the account's table holds), current_table (the effective date of today's table) and note. The same date is on every charged envelope as meter.price_table (early_access.mechanism)","note":"GET /v1/account/usage?window=7d|30d|90d (REST-only, free, rate-guarded): zero-filled daily[] time-series, by_tool[] with MCP tool ids, refunds{} — refunds are credit-backs: engine-failure/stale pre-auth returns (at the estimate) and, since 1.12.5, serve-failure refunds after finalize (at the actual charge); NEVER an error metric; verification outcomes live in the per-request log, not in account aggregates"},"label_classes":{"rule":"success = the label itself asserts a POSITIVE verification outcome (the only labels that may reach a 'clean' verdict); gap = honest coverage gaps (review-never-block, never fabrication evidence). Labels in neither class: use label_metrics recommended_tier, advisory markers, and the accusation category — an unclassified label must land in a LOUD bucket, never a silent pass. verified_statute_exists is a success label for EXISTENCE; if you need CURRENT law additionally reject status in {repealed, recodified, omitted, eliminated, former, historical_range, transferred, renumbered} (null, range, reserved and vacant are not currency findings) — that set is the STATUTE lane's; verified_rule_exists (1.20.0) carries its own currency statuses abrogated | reserved | transferred | renumbered, all to reject for current law, including 'reserved' (see court_rules.status). One existence caveat lives in evidence text: a hyphenated section number inside an OLRC repealed/omitted/transferred GROUP verifies with 'membership ... UNCONFIRMED' — the group row does not enumerate its members","classify_by":"MEMBERSHIP IN THESE LISTS, never by the PRESENCE OF A FIELD. A consumer whose detector branched on `typed_hallucination_category in (citation, statute)` dropped every manual_review_required label in SILENCE — that label carries no accusation category (it makes no fabrication claim, by design) and no advisory marker (it moves the verdict), so it fell through every branch of a detector written to catch exactly it. A partner reproduced this on a paragraph-pin finding and reported it (1.19.4). Read `label` against success / gap / review / advisory and treat anything in none of them as LOUD. Any field may be absent on any label; the label value never is","in_band":"since 1.20.8: EVERY verification.labels entry carries `label_class` — success | gap | review | advisory | accusation — the same class the lists below give, stamped from the engine's own vocabulary as the envelope is served (stored records too), so no consumer needs a copy of these lists at all. A partner's agent that held no copy refused to classify a flag it could not place and named the citation verbatim instead of assuming it benign — that is the right behaviour, and this field is what makes it unnecessary. `accusation` is new as a list: not_found_in_covered_volume, the attribution-class extras (pin_cite_out_of_range, case_name_mismatch, citation_year_mismatch, parallel_cite_mismatch, statute_section_not_found) and the quote lane's quote_not_found — the labels whose evidence makes a claim against the citation; read each with its label_metrics tier (case_name_mismatch is review-tier). The taxonomy is now EXHAUSTIVE over every label the engine produces: `review` gained ambiguous_multiple_matches (the corpus holds the citation more than once — you pick, never an accusation) and superseded_or_outdated (a REAL C.F.R. unit removed by the asked date — a currency finding, not a fabrication claim); negative_treatment_flag, in the published enum since the first envelope, is produced by nothing (reliance signals are served as the treatment block) and is listed as reserved with contradicts and wrong_jurisdiction. An entry with NO label_class carries a label in no class; treat it as LOUD. A response replayed from before this version lacks the field for at most 24 hours. The lists are also served at /.well-known/verdict-consumer-guidance.json (this document; no authentication) — there is no /v1/guidance route, and a 404 there means the wrong URL, not that the question cannot be answered","review":["manual_review_required","ambiguous_multiple_matches","superseded_or_outdated"],"advisory":["case_name_variant","citation_year_later_than_decision","supports","related_insufficient"],"success":["verified_exists","verified_statute_exists","verified_rule_exists","quote_exact_match","quote_normalized_match"],"gap":["outside_indexed_volumes","reporter_not_covered","statute_title_not_covered","statute_section_beyond_snapshot"],"accusation":["not_found_in_covered_volume","reporter_or_volume_nonexistent","pin_cite_out_of_range","case_name_mismatch","citation_year_mismatch","parallel_cite_mismatch","statute_section_not_found","quote_not_found"]},"volume_holes":{"rule":"since 1.14.2: a cited volume that is MISSING from the corpus's volume table but lies inside the reporter's recorded range is a coverage gap (outside_indexed_volumes; 1 credit; counted in coverage gaps, not in summary.suspect; typed_hallucination_category null) whose evidence names the recorded volume spans around it ('recorded around it: 1-287, 361'), never an existence claim. Reporter volumes are numbered consecutively, so a number between two recorded volumes normally names a published volume; the corpus cannot tell such a gap from a series boundary, so it never accuses on a missing volume. The claim is scoped to this reporter's own volume table: a case in a missing volume may still resolve through a parallel citation ingested under another reporter. reporter_or_volume_nonexistent is NO LONGER EMITTED (it remains a schema value for stored records issued before 1.14.2); a volume beyond the recorded range stays outside_indexed_volumes with the recorded spans named. In verdict_check_citations the secondary court-records corpus and SCDB rescue lanes run before a gap is served; in verdict_check_quotes / verdict_check_brief the gap is the index's own answer and the recency lane is the only rescue","since":"1.14.2 (verdictnews field report 2026-09-07: '335 Or. App. 201', printed six times by the Oregon Supreme Court, had been accused because the corpus records Or. App. vols 1-287 and 361)"},"proposition_support":{"since":"1.16.0 (2026-09-08)","opt_in":"propositions=true on verdict_check_citations (text mode), verdict_check_brief, verdict_check_ai_output","status":"STANDING advisory label since 1.16.6 (founder decision 2026-09-08): 'beta' is off `supports` after two calibrated blind panels over two corpora (51/52 pooled, Wilson [0.90, 1.00] — label_metrics.supports). What it asserts is unchanged: a lexical match of the sentence's proposition to a court-authored description, quoted — never a legal conclusion; a non-match is never a finding. Nothing else in the lane carries a label","rule":"For every VERIFIED case citation in text the lane reads what the document says the case stands for — the document's own parenthetical right after the citation ('(holding that …)'), else the clause from the sentence's signal phrase ('held that', 'stands for', 'establishes' …) to the citation, else the citing sentence, else the preceding sentence when the citation stands alone — and compares its CONTENT TERMS with court-authored descriptions of what the case was cited for (citing opinions' parentheticals: the holding tier, panelled 98.6%, and the screened describing tier). data.citations[].proposition = {engine, proposition, proposition_source, compared_terms, descriptions_consulted, verdict, best_match{description, holding, citing_count, overlap}, other_descriptions[], evidence, note}. verification.labels gains an ADVISORY entry ONLY for supports (>= 60% of at least 4 content terms found in ONE description — a lexical match with the description quoted, NEVER a legal conclusion). A lexical non-match is served in the block as verdict related_insufficient with the closest description quoted but emits NO label since 1.16.3: a blind panel found a paraphrase of the proposition in 41 of 60 such emissions on real prose — the non-match is not evidence of a wrong attribution. No label and verdict 'unmeasured' when the case has no mined descriptions or the proposition is too short: absence is never a statement the case does not stand for the proposition. contradicts is never emitted in v0.","catches":"a real case cited for a holding no court has attributed to it (the commonest hallucination shape around a correct citation)","does_not_catch":"a wrong sentence about a REAL holding whose words agree with a description ('mandates fees' against 'in the case of any successful action') — that is a claim check against the source text, not a citation check; and paraphrases too far from any description read as related_insufficient, never as a finding","status_and_coverage":"advisory only: the supports label never moves verification.status or coverage, and the block's verdict is never a label class","pricing":"proposition_check credits per MEASURED citation (a computed block); unmeasured blocks are free; the escrow estimate prices every citation and finalize refunds the difference; verdict_check_brief and verdict_check_ai_output add the same per measured citation","data_edge":"descriptions come from the frozen citation graph; a case described by no later opinion in it is unmeasured"},"regulations_in_prose":{"since":"1.15.0 (2026-09-08)","tools":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"rule":"Every C.F.R. citation in `text` is examined through the regulation lane (the same eCFR reading verdict_get_law serves, date = the title's latest issue date): data.regulations[] (data.report.regulations[] for the report tools) carries one entry per DISTINCT unit (since 1.20.10 two spellings that RESOLVE to one eCFR unit — '§ 102-71.10' and '§ 102-71.10.3' under a hyphenated part — fold into one entry: spans and occurrences combined, the other spellings in also_cited_as[], one charge) — {raw, occurrences, spans, resolution{label, evidence, status, citation, url, canonical_id, as_of, range_end?, category?, unread?}} — and verification.labels carries the law lane's own label for it: verified_exists (served), not_found_in_covered_volume (absent at the date — an accusation with evidence, after the version-history and, at a default date, Federal Register checks), superseded_or_outdated (a unit rescinded from the CFR — a real citation, not a fabrication), manual_review_required (a reserved position, an appendix designator under a rescinded unit), reporter_not_covered (NOT EXAMINED: a regulation shape the lane does not read, the lane switched off, eCFR unreadable — evidence names which; resolution.unread=true). Forms read: '42 C.F.R. § 483.12', the bare '42 C.F.R. 483.12' and '14 CFR 250.8', ranges, lists after a plural mark ('§§ 483.10, 483.12 and 483.25'), appendices; a bare form is never a case-citation shape any more. Whole-part and subpart citations ('45 C.F.R. Part 164', 'Part 164, Subpart E', 'Subpart E of 45 C.F.R. Part 164') are EXAMINED units since 1.17.0 (part_reading), and a paragraph pin on a section is checked against eCFR's own ids (paragraph_pins); before 1.17.0 a part cite was listed as an unexamined gap. Since 1.20.10 (2026-09-28) title 41's CHAPTER-PART numbering is read: '41 C.F.R. § 102-71.10' and '41 C.F.R. Part 102-74' are units (before 1.20.10 the section was not extracted at all and the part was read as the RANGE of parts 102–74 and falsely accused).","coverage":"coverage.completeness counts regulations: an unexamined regulation is a gap in coverage.gaps and the document is never 'complete' over it. Before 1.15.0 verdict_check_citations did not examine C.F.R. citations at all and could report 'complete' over a text whose only authority was regulatory — treat pre-1.15.0 records that way.","status":"a regulation counts as LAW beside U.S.C. sections: a document with only regulations is 'verified' when every unit is served, 'unverified' when none is, 'partially_verified' otherwise; a rescinded, absent, review or unexamined unit demotes a document exactly as a statute finding does. Since 1.16.1 verified LAW also counts as something verified when every CASE citation is a gap or an ambiguity: one ambiguous case beside verified regulations and statutes is 'partially_verified', not 'unverified' (before 1.16.1 the case rows alone decided that step)","depth":"section, plus the pinned paragraph since 1.17.0 (paragraph_pins); whole parts and subparts since 1.17.0 (part_reading)","pricing":"each DISTINCT regulation unit is priced once like a citation: the definitive rate when examined (served, absent, superseded, review), the gap rate when not examined; U.S.C. sections stay free; verdict_check_brief counts distinct units toward its included citations"},"line_structure":{"since":"1.16.5","rule":"line endings are normalised first ('\\r\\n' and '\\r' become '\\n'; every served offset indexes that text, which is the text as sent for LF input). Citation EXTRACTION (cases, U.S.C. sections, C.F.R. units, the shape scan, caption hints, parallel and short-form checks, the quote pairer, the escrow estimate) then reads a flowed copy in which a lone newline (a hard wrap: non-blank lines on both sides) is one space; a blank or whitespace-only line stays a paragraph break. A citation split across a line break ('347 U.S.\\n483', '347\\nU.S. 483', a pin or year on the next line, in a filed brief or PDF text) is therefore a counted, verified citation with its pin read, instead of an advisory shape-scan row; its raw string (and a quotation's served text) carries the wrap closed and its span locates it in the normalised text. A CRLF brief and its LF copy give the same result","propositions":"the proposition lane reads the ORIGINAL line structure: a newline ends a sentence only after a line that reads as a heading or a list item — it ends in ':' or in sentence punctuation that is not an abbreviation, is mostly upper-case or mostly Title-case words, carries a list marker, or is a short block-opening line much shorter than the next; a line ending mid-sentence (a function word, a comma, a digit, a caption's 'v.') never is. A heading glued onto a clause by a re-flow ('COUNT V: RACE DISCRIMINATION Separate but equal…') is stripped when its designator and separator can be told from the clause; one that cannot stays unmeasured. The same words at any wrap width from 25 to 110 characters give the same citations and the same verdicts (tested; a partner measured the drift, 2026-09-08)","measured":"A partner disproved the first version of this claim on a real 14-citation motion (1.17.1), and a full review of that fix disproved the SECOND version: the number published at 1.17.1, '23 of 23', was measured over the 23 documents that had drifted BEFORE the fix, a population selected by the old failure. The honest figure, over every template document that produces proposition blocks: 53 of 61 give identical verdicts at every wrap width from 25 to 110 when the re-wrap keeps the document's line structure. 8 still drift at some widths; the known causes are a degenerate proposition when a boundary lands mid-sentence, and a hyphenated word broken across lines. Flattening a document to ONE line first is a lossy change to the text rather than a re-flow: a document whose list items ARE its lines reads differently once they are merged, and that is correct. What the lane reads is CONTENT, never where a wrap fell. The heading test consults no length (a hard wrap makes short lines by construction); a Title-case or one-word heading counts only where it opens a block; an INSTRUCTION marker ('[If applicable.]', '[Insert date]', a leading footnote number) is stripped, while a Bluebook alteration ('[T]he Constitution …'), a party substitution, a term placeholder and a form checkbox are CONTENT and stay; a short-form citation ('550 U.S. at 570') is never part of a proposition; and the multi-case ambiguity window is the citing sentence itself","before":"records before 1.16.5 could carry a wrapped citation only as a shape_scan_strings row (free, uncounted)"},"statute_depth":{"depth":"section, and the PINNED PARAGRAPH too — C.F.R. since 1.17.0, U.S.C. since 1.18.0 (paragraph_pins)","rule":"verified_statute_exists / statute_section_not_found decide the SECTION a citation names. A paragraph pin below the section — '(b)(4)(ii)', '(a)(2)(A)' — IS read now: for C.F.R. since 1.17.0 against eCFR's own paragraph ids, and for U.S.C. since 1.18.0 against the U.S. Code's own USLM hierarchy (see paragraph_pins). Before those versions a pin that did not exist under a real section still verified silently; records issued earlier read that way","applies_to":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output","verdict_get_law"],"since":"1.14.1 (disclosure; the section gate has always been section-level); 1.17.0 (C.F.R. paragraph pins read); 1.18.0 (U.S.C. paragraph pins read)"},"sovereign_mismatch":{"since":"1.20.2 (2026-09-09)","tools":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"rule":"A citation whose document caption names a SOVEREIGN of a different FAMILY from the reporter's caption side — a federal party ('United States', 'Government') where the reporter names a state ('Ohio', 'State', 'People', 'Commonwealth'), or the reverse — is a manual_review_required finding carrying the structural marker sovereign_mismatch: true, and the document is demoted exactly as a reversed caption is. Until 1.20.2 that citation verified CLEAN: 'United States v. Terry' cited for Terry v. Ohio — a federal prosecution written where the reporter shows a state stop-and-frisk case — passed as an untestable sovereign party, while 'Ohio v. Terry' was review and 'Texas v. Terry' an accusation (the severity inversion a full review found). ONE relation now serves the suppression and the verdict: SAME (equal sovereign words; 'Government' is the United States), COMPATIBLE (a generic state designator against a state name — 'State v. Smith' and 'Ohio v. Smith' cannot be told apart without the court, so that alias stays clean), FAMILY (federal against state-level — the inversion class), DIFFERENT (two different state names). A bare state name ('Texas v. Terry') was an accusation before and stays one through the name lane (case_name_mismatch, measured) — and that lane passes two states sharing a token ('North Carolina' for a South Carolina case), before and after this change: a disclosed gap, not a guarantee. A state named WITH a generic word ('Commonwealth of Pennsylvania', 'District of Columbia') cited for another state's case verified clean before and is a review finding now, not a new accusation. A reversed caption stays the caption_order review","marker":"labels[].sovereign_mismatch: true is STRUCTURAL — counted in summary.sovereign_mismatches, counted into attribution_flags (the document demotes), carried by verdict_check_brief's flags and verdict_check_ai_output's review count; never advisory; the label VALUE is manual_review_required until the panel says otherwise, and the marker survives either way","pricing":"no price: the finding rides the citation's existing charge","class":"REVIEW, never the fabrication class: the case exists and the citation resolves. Its blind panel (label_metrics.sovereign_mismatch) returned 5 of 7 findings unanimous with every control unanimous — a Wilson floor of 0.36 at n=7 — so the class stays review; the two disputed findings are the DISPLACED shape (a sovereign written where the reporter has a private party), which the evidence now names as such","not_read":"a sovereign party whose caption has NO sovereign side at all ('United States v. Smith' cited for Smith v. Jones) is judged by the name lane as before; abbreviated sovereigns ('U.S. v. Katz') are not sovereign vocabulary and are not read here; a short non-sovereign party (three letters) is never judged, so 'Texas v. Doe' for Ohio v. Doe keeps its existing case_name_mismatch and 'State v. Doe' for Ohio v. Doe stays clean"},"court_rules":{"since":"1.20.0 (2026-09-09)","tools":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"rule":"A citation that NAMES a federal ruleset — 'Fed. R. Civ. P. 12(b)(6)', 'Fed.R.Civ.P. 56', 'Federal Rule of Evidence 403', 'FRCP 59(e)', 'Fed. R. App. P. 4(a)(1)(A)', 'Fed. R. Crim. P. 11', 'Fed. R. Bankr. P. 3001', and a plural list or range under one marker ('Federal Rules of Evidence 104(a), 401, 402, and 403'; 'Rules 801-807') — is checked for EXISTENCE against that ruleset as the Office of the Law Revision Counsel publishes it in the U.S. Code appendix (Civil, Criminal, Evidence, Appellate, Bankruptcy; 528 rules across the five, plus the two supplemental sets that are not read; refreshed with the quarterly Code cycle). data.rules[] carries one entry per distinct rule: {raw, resolution: {label, cite, ruleset, ruleset_name, rule, evidence[, heading, status, release, pin, pin_status, range_end, elsewhere, suffix_unread, cited_as]}}. A held rule is verified_rule_exists (a SUCCESS label; counts as verified law for the document status exactly as a U.S.C. section does). No charge","status":"a rule number the appendix prints as [Abrogated …], [Reserved], [Transferred] or Renumbered is a REAL rule number and verifies, carrying `status` abrogated | reserved | transferred | renumbered as a CURRENCY advisory (null = live) — the same shape as a repealed U.S.C. section. A consumer that needs current law rejects a non-null status","absent":"a rule number ABSENT from the ruleset named is manual_review_required (the REVIEW class; the document is partially_verified beside verified law, and unverified when the absent rule is the ONLY law unit — nothing verified is never 'verified' by the absence of fabrication). Never the fabrication class, and this is a rule, not a hedge: the ruleset held is the CURRENT edition and no renumbering history is held, so a rule renumbered or abrogated before this edition (the 1983 Bankruptcy Rules were renumbered wholesale) is absent here too. The evidence names the ruleset's numbering and, when the number exists in a SIBLING ruleset, says so and lists it in `elsewhere` ('Fed.R.Civ.P. 702' -> 'Rule 702 exists in the Federal Rules of Evidence'): a harvest of 2,977 judicial opinions found that shape to be the commonest real error","not_read":"a BARE 'Rule 12(b)(6)' (7,426 in the harvest beside 2,467 named forms): which ruleset it names depends on the document, and a state court's Rule 12 is not the federal one — unread, never guessed. 'Civil Rule 6.3', 'Evidence Rule 808', 'Bankruptcy Rule 927' WITHOUT a federal marker: the harvest found them naming LOCAL and STATE rules. The Supplemental Rules (Admiralty A-G, Social Security 1-8). A PIN below a rule ('(b)(6)'): carried with pin_status 'unread' — the appendix text is held without paragraph structure, so no pin is checked and none is accused. A DECIMAL the ruleset does not number under a rule it does ('Fed. R. Civ. P. 56.6'): only Rule 56 is verified — the entry's cite and rule name Rule 56, `cited_as` keeps the document's '56.6', and `suffix_unread` carries '.6', which in extracted text is a footnote marker glued to 'Rule 56.' more often than a rule; the suffix is never verified and never accused. Under a PLURAL spelled-out marker ('the Federal Rules of Evidence 2 years ago') a number outside the ruleset's own numbering is prose and is not read; a singular marker ('Fed. R. Civ. P. 99') reads its number and lets the ruleset decide","off":"VERDICT_COURT_RULES=off, or a deployment without court_rules.sqlite, answers every named rule citation statute_title_not_covered ('the appendix is not held … unexamined'): a coverage gap, never a finding, and the document is never 'complete' over it","measured":"MEASURED BEFORE SHIPPING, every figure from a committed, seeded harness run through the shipped path. Emission rate on real prose: 12,000 judicial opinions, 9,328 named federal rule citations, 9 findings (0.10%); 3,000 opinions, 2,223 citations, 3 findings (0.13%). Every finding was read: eight name a rule number the ruleset does not hold because the writer named the WRONG ruleset (an Evidence rule cited as Civil or Criminal, and the reverse) or a number that never existed (Evidence Rule 108; Bankruptcy Rule 7011 for 9011), and one is a 1982 citation to Bankruptcy Rule 752 under the pre-1983 numbering — correct when written, absent from the CURRENT ruleset, which is exactly the currency class this lane discloses. Blind panel, three judges on three models, no label and no detector answer, asked the legal question: 12 of 12 correct-order controls and 12 of 12 constructed absents unanimous, 8 of 9 findings unanimous, Wilson floor 0.56. The ninth (Bankruptcy Rule 7011) all three judges called held, inferring it from the 70xx numbering; the Office of the Law Revision Counsel's published Part VII runs 7010, 7012 … and has no 7011, and that publication is what this lane reads — the disagreement is recorded, not resolved in the lane's favour. Recall: 60 of 60 constructed absent numbers fire through the shipped path with 0 false fires on the 60 originals. The grammar was corrected twice by these harvests before shipping (local and state rules, proposed rules, the number-first form, reporter and page numbers after a marker, a footnote marker glued to 'Rule 56.'); every artifact class is a permanent test. Review class, by the floor and by construction: the store holds the current ruleset and no renumbering history, so one signal never accuses"},"paragraph_pins":{"since":"1.17.0 (2026-09-08) for C.F.R.; 1.18.0 (2026-09-09) for U.S.C.","tools":["verdict_get_law","verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"usc":"Since 1.18.0 a U.S.C. citation whose section number is IMMEDIATELY followed by a paragraph path ('15 U.S.C. § 1692g(a)(1)'; '§ 1692g (a)' with a space is prose) has that path checked against the U.S. Code's own USLM hierarchy — the OLRC's marked-up structure, held locally and refreshed with the quarterly corpus cycle. data.statutes[].resolution.pins[] carries {pin, status found|not_found|unread, evidence[, paragraph_text, level, code_release]}: found serves the paragraph's own text and names its level (subsection, paragraph, subparagraph, clause…); not_found names the deepest existing prefix and its subdivisions; unread covers a section whose structure is not held, a section with no marked-up subdivisions, subdivisions the OLRC prints inline instead of marking up, and a string that is not a paragraph path — this lane never accuses on an empty or inline tree. INLINE means the paragraph's own text carries the deeper levels as printed markers rather than as marked-up children: a marker that OPENS a segment ('A State shall—(A) buy; and (B) sell'), or a marker whose SUCCESSOR appears later in the same text ('shall (1) publish and (2) certify'). A lone parenthesised CROSS-REFERENCE ('described in clause (i)') is neither and does not suppress a finding. Two provisions named in ONE cross-reference do read as inline and suppress it — deliberately, because a suppressed finding still reaches you as unread with its reason, while the other error would report a paragraph the Code really publishes as missing. `code_release` names the U.S. Code release point the section was read against (the OLRC's own release id, '119-102'), stamped per SECTION from the row that answered — it is a release id, never a date, which is why it does not travel in the C.F.R. half's `as_of` (an ISO issue date). It is absent when the edition is unrecorded. A pin is attached only to a section the statute lane VERIFIED (a pin on a section that does not exist is moot). Since 1.18.1 the two halves are the SAME: a not_found U.S.C. pin adds a manual_review_required label targeting the citation WITH its pin ('15 U.S.C. § 1692k(a)(4)') and the unit is partially_verified, exactly as a C.F.R. pin does. 1.18.0 served it as a field only, so a document whose every statutory pin was invented read status 'verified', coverage 'complete' and success labels, with the failure visible only inside the block — a partner reproduced it the same day and it is fixed. The label is REVIEW, never the fabrication class: the section is real. U.S.C. sections are free and the pin check adds no charge","rule":"A C.F.R. section citation that pins a paragraph ('45 C.F.R. § 164.508(b)(4)(ii)') has the pin checked against eCFR's OWN paragraph identifiers for that section at the issue date (the renderer marks every paragraph with its full path; no reconstruction). data.unit.pins[] (verdict_get_law) and data.regulations[].resolution.pins[] (the prose tools) carry one entry per pin: {pin, status found|not_found|unread, evidence[, paragraph_text, as_of]}. found = the paragraph exists and its OWN text is served (up to 1200 characters) so the claim can be read against it; not_found = the section exists but no such paragraph does — the evidence names the deepest existing prefix and its subdivisions, a manual_review_required label is added on the citation and the unit is partially_verified (a pin is a pointer: the review class, never the fabrication class); unread = eCFR's paragraph structure could not be read or the pin is not a paragraph path — a note, nothing labelled, never accused","one_unit":"one section pinned several ways in a document is ONE unit and one charge; every distinct pin is checked (up to four), every check rides the unit's pins[] with `cited_as` = the spelling that carried it, and a not_found pin's manual_review_required label TARGETS that spelling","status":"a served section with a not_found pin is a VERIFIED section carrying a review flag: it counts as verified law for the document status (a regulations-only document is partially_verified, never unverified) and once in summary.regulations_review, not in regulations_partial; verdict_check_brief lists the pin label among its flags","does_not_catch":"an EXISTING pin on the wrong content — the sentence says (b)(4)(ii) stands for something it does not; only the served paragraph text makes that readable (the anchored-claims ledger is the tool for that). Since 1.18.0 U.S.C. pins are read too (see `usc`), with the same limit","pricing":"no price: the pin check rides the section's unit","off":"VERDICT_CFR_PINS=off on a deployment leaves pins unread (a note); no label, no demotion","unchecked":"a pin is served 'unread' with the reason whenever it could not be checked: a section eCFR served no text for, a paragraph structure that could not be read, a failed lookup, or more than four spellings of one section in a document (the fifth and beyond are named, never dropped). A pins[] entry names the spelling that carried it in `cited_as`; a merged spelling's review label targets that spelling, which may not be the entry's own `raw`"},"part_reading":{"since":"1.17.0 (2026-09-08)","tools":["verdict_get_law","verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"rule":"A whole PART, a SUBPART or a RANGE of parts is a unit read from eCFR's structure tree at the date — '45 C.F.R. Part 164', '45 CFR pt. 164', 'Part 164, Subpart E', 'Subpart E of Part 164', '40 C.F.R. Parts 1500-1508'. The tree is complete for parts and subparts, so existence, heading, section count and state are decided without a fetch: a live part or subpart is verified_exists with data.unit = {kind: 'part', title, part[, subpart], heading, sections, sections_reserved, subparts[] (id, heading, reserved, sections), section_range, url, text: ''} — NO TEXT is served for a part (cite a section for text); a [Reserved] part or subpart is verified_exists at partially_verified (a position with no sections); a part listed only inside a reserved part range is manual_review_required; a part absent from a served tree is checked against eCFR's version history first — a rescinded part is superseded_or_outdated (40 C.F.R. Part 1500 after the 2025 CEQ rescission), otherwise not_found_in_covered_volume with the neighbouring listed parts named. A cited RANGE's end is settled whatever the start's state — an end the tree does not list is the accusation, naming the end. Subpart identifiers are read as eCFR spells them: a letter ('E', 'AAA'), a letter range ('Subparts A-C'), a hyphenated id ('Subpart D-3') and a numeric id ('Subpart 15.9'); the plural word with no identifier ('the subparts of Part 164') is prose, and one subpart spelled two ways is one unit and one charge. HYPHENATED PARTS (since 1.20.10, 2026-09-28): title 41 numbers its parts CHAPTER-PART ('Part 102-74', 'Part 301-10', 'Part 51-9'); 'Part 102-74' is read as that PART — listed (served), inside a reserved range of hyphenated parts ('Parts 102-72—102-73 [Reserved]': review), or absent (not_found_in_covered_volume naming the hyphenated neighbours) — never as the range of parts 102–74 (before 1.20.10 that reading falsely accused 'has no Part 74 — the end of the cited range'); in every other title 'Parts 1500-1508' stays a range. A part eCFR lists as a SHELL — a heading with no sections and every subpart [Reserved] (41 C.F.R. Part 102-74 after GSA's 2025-12-16 rescission; 48 C.F.R. Part 525) — is decided by the version history: every unit removed = superseded_or_outdated (evidence says it is listed only as a heading), otherwise verified_exists at partially_verified (a listed position that carries no rules); before 1.20.10 a shell served verified_exists 'with 0 section(s)'. A section under a hyphenated part is read against the same inventory ('41 C.F.R. § 102-74.265' — removed 2025-12-16 — is superseded_or_outdated; '§ 102-74.9999' is accused). A DOTTED TAIL after a section number ('§ 1604.11.3', on a start or a range end) is served as the listed section with a note that the tail is a footnote marker or prose (one unit, one charge), unless eCFR lists the dotted id itself ('41 C.F.R. § 109-42.1100.51'), which is then the unit; a SUBPART id cited as a section ('41 C.F.R. § 51-9.1', '48 C.F.R. § 15.3') is manual_review_required naming the subpart, never an accusation; a part form followed by a dotted tail ('Parts 160-164.3 The rule', 'Part 102-74.265') has two readings (a range plus a footnote marker, or a section written with the part word) and is manual_review_required naming both, never silently dropped; a section listed after a hyphen-shaped pair of parts and admitted only by the PLAIN range reading ('Parts 102-71, Section 105.2 of the Lease') is manual_review_required when it does not exist, never an accusation; a CHAPTER cited as a part in a CHAPTER-PART title ('41 C.F.R. Part 60' — OFCCP's parts 60-1 to 60-999) is manual_review_required naming the chapter's parts; a section list after a plural range of parts ('45 C.F.R. Parts 160-164, §§ 160.103, 162.100 and 164.502') reads every member whose part lies between the ends. A SUBPART is never accused: one absent from a live part is manual_review_required with the part's subparts named (a subpart has no version history of its own — the second signal a section or part absence has), one inside a reserved subpart range ('Subparts A-C [Reserved]') is a reserved position (partially_verified), and one cited under a [Reserved] or rescinded part is manual_review_required with the part's state or removal named. 'Subpart E of 45 C.F.R. Part 164' reads the same as 'Part 164, Subpart E'. A range of parts serves its START with range_end = the cited end; both ends must be listed, a reserved end demotes to partially_verified, a rescinded end is manual_review_required naming the range, and the parts between are not checked","prose":"the regulation lane examines whole parts now (before 1.17.0 they were listed as unexamined gaps): each distinct part or subpart is one unit at the definitive rate; a document whose only regulatory authority is a part cite can now be 'complete'","pricing":"one unit at the law rate (verdict_get_law) / the definitive rate (the prose tools)"},"caption_order":{"since":"1.19.0 (2026-09-09)","tools":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"rule":"A case citation whose two party names match the reporter's caption in the OPPOSITE order ('Campbell v. State Farm Mut. Auto. Ins. Co.' for a citation whose caption is 'State Farm Mutual Automobile Insurance v. Campbell') carries an extra manual_review_required label naming both captions, and the document is partially_verified. Until 1.19.0 both orders passed every check: each party was tested for evidence ANYWHERE in the resolved caption, so party ORDER was never compared and a reversed caption verified clean","class":"REVIEW, never fabrication — the case exists and the citation resolves. Two legitimate causes the label names in its evidence: the parties really were reversed, or the caption of a different stage of the same litigation was used (parties routinely swap between trial and appeal, so a district-court reporter and an appellate one carry opposite captions)","bounds":"it fires only where BOTH party names are readable from the document's own caption and both match the opposite side; one-sided evidence is the ordinary abbreviation case and never fires. Since 1.19.2 SOVEREIGN captions are read too ('United States v. Katz' for Katz v. United States; People, State, Commonwealth and state names) with the identity required to AGREE — the same state on both sides, or the same designator with no state conflict — so 'Texas' never names 'Ohio', 'People' never names 'State', 'North Carolina' never names 'South Carolina', and a state never names a city of the same name. Measured recall 0.968 on CORPUS-FORM captions (label_metrics.caption_reversed). It does NOT claim to catch every reversal, and these are SILENT by construction: an abbreviated sovereign ('U.S. v. Katz', 'Com. v. Smith'); a caption the extractor does not split into two parties at all ('United States of America v. Katz', 'State of Ohio v. Terry'); a party naming a sovereign that is not word-for-word the caption's ('State' for 'State of Ohio'); initials-only names and very short surnames. The recall population is corpus-form captions and is structurally blind to the abbreviated ones, so END-TO-END recall on running prose is LOWER than 0.968. A parallel-reporter caption is never a finding: when the index holds a sibling row for the same court, date and docket carrying the document's own order, the writer is citing the other reporter's caption and nothing is emitted","marker":"the label carries `caption_order: true` — a STRUCTURAL marker, so a gate never reads the evidence prose. It is NEVER `advisory` (an advisory label is informational and never moves status; this one demotes) and never carries a typed_hallucination_category. `data.summary.captions_reversed` counts them and verification.notes says so in words","pricing":"no price: the finding rides the citation's own check"},"pin_check":{"rule":"since 1.14.6 a CASE citation with a page pin carries pin_check on the citation entry's resolution (data.citations[] for check_citations; data.report.citations[] for check_brief / check_ai_output) = {pin, page, within_case, first_page, last_page[, reason][, containing]}: within_case true = the page lies inside the resolved case's page span; false = it does not (the pin_cite_out_of_range label still fires) and `containing` names the case in the same reporter volume whose span holds the page ({cite, case_name, canonical_id, first_page, last_page}) or null when no indexed case does; null = the check could not run (reason: 'page extents unusable' for tier-2 or range-page rows, 'not a page pin' for footnote, paragraph and star pins, 'resolution era-suspect' when the resolution itself is era-impossible and every check on it is withheld). Absent when the citation carries no pin, did not resolve verified_exists, or is a shape-scan row. Additive; labels and classes unchanged","applies_to":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"since":"1.14.6"},"page_structure":{"rule":"since 1.20.10 a CASE citation's resolution may carry FACTS about the cited page — what the index holds there, never a label and never a class: `within` = {case_name, first_page, last_page, cite_norm, cap_case_id}, the record whose pages hold the cited page (on not_found_in_covered_volume: a page inside a case cited as a first page; on an outside_indexed_volumes boundary gap: the record that ENDS on the page); `boundary`: true = the page is a record's last page and no indexed record starts there (a case may begin on the page where another ends — a coverage gap); `past_tail`: true = the page lies past the volume's last indexed record (a coverage gap; the evidence names that record's page); `order_signal` on not_found_in_covered_volume = whether a subsequent-history order phrase, a narrated order or a table marker stands by the citation; `extents_derived`: true = the record's page span was derived from the next record's start, not read from the record — a pin past such a span is never accused; `pin_hole` = the reason a pin could not be range-checked because it lies in a span no indexed record tiles (pin_check.within_case is null and pin_check.reason carries the same words). Every key is optional and additive; classify by label_classes, never by these keys","applies_to":["verdict_check_citations","verdict_check_brief","verdict_check_ai_output"],"since":"1.20.10"},"era":{"era_note":"on a verified_exists resolution: its decision date cannot belong to the cited reporter edition or volume — the corpus row, not your text, is suspect; in check_citations, when your own year corroborates the row (+-1) the note is dropped, otherwise year/attribution/pin/parallel checks are WITHHELD (era_conflict: true)","era_notes":"on a case RECORD: per-citation notes; a record carries era_note only when its date fits NONE of its citations' volumes (misdated row) — otherwise the outlier is a spurious parallel citation on that record, or — when it is the record's own source volume — a late-printed or compound (original + rehearing) record; the note says which","date_repair":"1.11.0+: when the secondary court-records corpus's citation table binds the same citation to the same case (party-wise name match) with a date strictly inside the volume window (within 90 days of its printed range, and of the record's own source volume's), decision_date is that corpus's (date_source: 'cl-corpus'), cap_decision_date keeps the CAP row's date, date_repair names the secondary-corpus record (verdict:cl:<id>) and window, era_note states the repair, era_conflict is NOT set and every check runs (era_repaired: true marks such a resolution or record); a document year agreeing with EITHER date is never accused, a later year by 2-3 draws the non-accusing citation_year_later_than_decision advisory, any other year citation_year_mismatch naming both dates. Absent when no independent record settles the date (the withholding stands; where the secondary court-records corpus's record bears on it, era_note adds the fact: the same date; a different decision at that citation; or a late-printed/compound record (the record was ingested from the cited volume yet its date fits its other citations) — the latter two carry era_other_decision: true and survive corroboration as review pointers)","successor":"on verified_statute_exists: ALWAYS a list of citation strings; a range successor reads '41 U.S.C. §§ 6306-6308'"},"law":{"tool":"verdict_get_law","coverage":"federal REGULATIONS via eCFR, date-versioned (as_of, default = eCFR's up_to_date_as_of for the title, refreshed daily; an as_of before 2017-01-03 is a coverage GAP, never a miss). U.S.C. citations here are an honest coverage gap; their existence, currency and successors ride the statute lane inside verdict_check_citations","shapes":["part | subpart | parts range (since 1.17.0): '45 C.F.R. Part 164' | '45 CFR pt. 164' | '45 C.F.R. Part 164, Subpart E' | 'Subpart E of Part 164' | '40 C.F.R. Parts 1500-1508' — decided from eCFR's structure tree (existence, heading, section count, state); no text served (see part_reading)","section: '29 C.F.R. § 1604.11' | '20 CFR 404.1520(a)'","section forms (since 1.14.3, INVENTORY-driven): the string after the section mark is read against eCFR's own list of section ids for the title at the date, and eCFR's exact id is served — so hyphenated numbers ('17 C.F.R. § 240.10b-5', '26 C.F.R. § 1.132-1', '48 C.F.R. § 252.204-7012', '27 C.F.R. § 21.94-T'), parenthesised numbers ('26 C.F.R. § 1.501(r)-4', '1.501(c)(3)-1', '17 C.F.R. § 275.206(4)-1', '29 C.F.R. § 2584.8477(e)-1') and range-shaped ids ('48 C.F.R. § 9904.412-60.1') resolve whatever their title; a parenthesis that is NOT part of an id is a paragraph pin or an official-interpretation comment ('12 C.F.R. § 1024.41(b)(1)-4' serves § 1024.41 as partially_verified — the comment is not checked); typographic dashes and letter case are normalised to eCFR's ids; an em-dash after the number is prose","section RANGES (since 1.14.3): '29 C.F.R. §§ 785.27-785.32' | '42 C.F.R. 483.10-.12' | '45 C.F.R. §§ 164.400-414' | '§§ 785.27 through 785.32' | '26 C.F.R. §§ 1.132-1–1.132-9' | '21 C.F.R. §§ 210.1-211.208' (a range may span parts). BOTH ends must be in eCFR's structure at the date: the START section's text is served, data.unit.range_end names the end, unit.citation reads 'T CFR start-end', the canonical id is verdict:ecfr:T-start-end@date, and a note says the sections between were NOT individually checked. A range whose end is not in the structure is read under its other readings and, when none exists, is not_found_in_covered_volume naming the readings tried","part appendix (since 1.11.2): '20 C.F.R. Part 404, Subpart P, Appendix 1' — also pt./subpt./app., '§ 404, App. 1', hyphenated/prefixed designators ('Appendix A-1', 'Non-Mandatory Appendix Q'), 3-letter or mixed-case subparts ('Subpart AAA', 'Subpart Cb'), and eCFR's own 'Appendix 1 to Subpart P of Part 404'. Resolution is against the title's complete eCFR structure tree; the subpart may be omitted when the designator is unique in the part; roman and arabic readings are BOTH tried and distinct appendices under the two readings are reported as ambiguous — or settled by a cited item that only one of them carries (the note says so) — never guessed; a designator with a family in scope ('Appendix A' where A-1…A-8 exist) is ambiguous too; eCFR's own 'Appendix A to Part 61' is read as explicitly part-level","appendix ITEM (since 1.11.2): '..., Appendix 1, § 12.04' | 'Listing 12.00B' | 'Part A, Listing 12.04' | 'Appendix 2, Rule 201.14' — the item number is read wherever it stands after the designator. A heading is qualified by TAG: a prose paragraph or a table row whose first cell IS the number; a table cell or formula line that merely starts with a number is never an item. A Listing is its paragraph run to the next heading; a Grid rule is its table row, verbatim, with ditto_resolved alongside ('do' = same as the row above; [n] marks a footnote). '12.00B3' serves paragraph 12.00B partially_verified with the numbered sub-item flagged for review; further items in one citation are noted, not resolved","section appendix (since 1.11.2): '29 C.F.R. § 1910.1001, Appendix A' | 'Appendix A to § 1910.1200' — OSHA-form appendices live inside the section's own text and are isolated from it"],"section_reading":"since 1.14.3: a section string is settled against eCFR's structure tree (the inventory of section ids for the title at the date), never by a grammar guessing where a number ends. Readings, in order: the whole string is an id; two ids joined by a dash are a range; the longest id that is a prefix of the string is the section and the rest is a pin or comment — unless ids of the rest's shape exist under that base ('275.206(4)-9' beside 275.206(4)-1…-8), which is an ABSENCE naming those siblings; a [Reserved] id or a string inside a reserved span is a reserved position: status partially_verified, no text (label verified_exists for a listed reserved id, manual_review_required inside a span). A listed id whose text endpoint answers 404 is served from the structure as partially_verified, never accused. An absence in the inventory is confirmed by fetching the string itself before any accusation, and eCFR's own 'No matching content' is quoted in the evidence; an unrecognised eCFR answer is a gap. If the structure tree cannot be read the section is fetched as before. Since 1.14.4 a SECTION-APPENDIX citation ('29 C.F.R. § 1910.1001, Appendix A') settles its section the same way (the tree's part and exact id; a listed section whose text endpoint answers 404 is manual_review_required, never not_found) and a cited range whose START the tree does not list is confirmed end by end before any accusation. Since 1.14.5 sections of a LETTER-part ('26 C.F.R. § 15a.453-1', '29 C.F.R. § 4041A.21', '12 C.F.R. § 269b.110', '42 C.F.R. § 51c.101' — a part carries at most one letter, any case) and temporary Coast Guard ids ('33 C.F.R. § 165.T01-0149') are read like any other section; a word after the part number ('§ 1604.See') is still prose, never an id. data.unit.url carries eCFR's own spelling of the section id (case and spacing kept, spaces percent-encoded). Since 1.14.6 a section or part that is absent at the date is checked against eCFR's VERSION HISTORY before any accusation: a unit that once existed and was removed from the CFR (the 2025 NEPA rescissions, e.g. '40 C.F.R. § 1502.14') is superseded_or_outdated — status verified, no hallucination category, the evidence names the removal date — never not_found_in_covered_volume; a unit re-added by the date, one first promulgated after the asked date, or one with no history is decided on the edition alone; an unreadable history makes the absence a GAP, never an accusation; an appendix under a rescinded part is review. Since 1.14.7 the section outcome vouches for the SECTION alone: an appendix designator under a rescinded section, or a joined range whose rescinded start leaves its end unconfirmed, is manual_review_required with the removal named (never a verified status over an unchecked designator or end); a paragraph pin on a rescinded section is noted, not checked; when the only earlier history row is eCFR's 2017 baseline import the evidence says 'unamended since its baseline import', not 'last amended'. Since 1.16.2 a section inside a reserved span whose START eCFR mistyped ('§§ 97-97.106 [Reserved]': 34 C.F.R. § 97.105) is a reserved position (partially_verified), not an accusation","section_text":"since 1.15.1 data.unit.text carries EVERY block of the section in document order — paragraphs, flush paragraphs, headings and TABLE ROWS (cells joined by ' | ', one row per line); before 1.15.1 only <P> paragraphs were served, so a section whose figures live in a table (40 C.F.R. § 19.4's penalty amounts) read as prose without them","whole_appendix_cap_chars":120000,"on_cap":"an appendix longer than the cap is served with truncated: true, total_chars and items (its qualified headings) — cite an item for its full text","labels":{"verified_exists":"unit retrieved; evidence names the eCFR as-of date and, for an item, the isolation","superseded_or_outdated":"since 1.14.6: a SECTION absent at the date that eCFR's version history shows was in the CFR and REMOVED on or before that date (status verified, no hallucination category; evidence names the removal date and the last amendment). A citation naming an appendix under a rescinded PART is manual_review_required with the removal in the evidence (the designator cannot be checked); an unreadable history makes the absence a gap, never an accusation; a unit re-added by the date, one that only ever existed as a [Reserved] placeholder, or one first promulgated after the asked date is decided on the edition alone","not_found_in_covered_volume":"a CFR title outside 1-50; a missing PART (absent from a SERVED eCFR structure tree — a served tree proves the date); a missing SECTION or section appendix that eCFR itself reports as 'No matching content found' at the current edition or at YOUR as_of (at a default date, after the Federal Register check of absence_semantics.edition_lag); a section-appendix LETTER absent from a section whose text carries other appendix headings and no family of that letter; a missing ITEM (its number appears nowhere in the resolved appendix's text) — each as of the date named in the evidence (corpus_as_of carries the same date). A part-level APPENDIX that matches no structure node is never accused (identifiers are heterogeneous; section-level appendices have no node). Since 1.11.4 eCFR's 404 body decides the cause: a 404 eCFR attributes to the DATE, a 404 with no recognised reason (never evidence of absence), or ANY absence — section, part, item, letter — at a default date taken from a cached index (the current edition may be newer) is reporter_not_covered, never an accusation (1.11.5). An absence is a claim only at YOUR as_of, or at the current edition when eCFR's version history shows no removal and, for a section or part, the Federal Register records no later rule naming it (absence_semantics.edition_lag, 1.20.7)","ambiguous_multiple_matches":"unverified: the citation fits several appendices — cite the subpart or the full designator","manual_review_required":"unverified, never an accusation: no appendix node matches the citation (evidence carries the part's inventory: the cited subpart's appendices in full, or the first 12 with '(+N more)', and a near miss such as 'Appendix 1 to Subpart P is under Subpart P, not Subpart B'); a designator found only under the other numbering (roman/arabic) with no item to settle it; a [Reserved] appendix; an item number present in the text but not as a heading; the same number leading several times (bundled documents); a lettered paragraph absent from a found item (evidence names the paragraphs found); an appendix with no numbered items; material after the appendix the grammar cannot place ('Method 5', '§ 4.1.2'); a section with no appendix headings","reporter_not_covered":"coverage gap (1 credit): a U.S.C. citation; an as_of before eCFR's point-in-time coverage, past the title's latest issue date (eCFR's own 'past issue date' answer included), or on a reserved or unlisted title; a title with no structure tree at that date; an eCFR 404 with no RECOGNISED reason (a gateway page is not evidence of absence — at any date); and, at a DEFAULT date taken from a cached index, any absence of a section, part, item or appendix letter (the current edition may be newer). Not covered by this rule: a part-level appendix matching no structure node (manual_review_required) and a title outside 1-50 (not_found)"},"outage":"since 1.11.3/1.11.4: if eCFR's title index is unreachable the default as-of date comes from the newest cached index and is UNCHECKED (the current edition may be newer), or your explicit as_of is accepted unchecked; verification.notes says so and coverage.completeness is 'partial' on served units. At a default date from a cached index every absence is a gap (reporter_not_covered, 1 credit); at YOUR as_of an absence eCFR confirms is an accusation. With no as_of and no index at all the call fails (tool error, refunded) rather than guess a date. A non-calendar as_of is a caller error (400); a CFR title outside 1-50 is a not_found envelope","text_encoding":"UTF-8 by document declaration since 1.11.2 — earlier versions could serve 'Â§' for '§' in regulation text; a non-UTF-8 upstream body is an incident, not a caller error"},"label_metrics":{"rule":"recommended_tier is derivable policy: block only labels whose measured precision supports it; review 0.5-0.9; advise below. Labels without an entry are UNMEASURED on real-corpus prose — treat unmeasured accusation labels as review-tier, never blocking, until an entry appears. Fabrication-class resolution labels run under a standing zero-false-positive release gate instead of a sampled number.","case_name_mismatch":{"precision":0.255,"recall":0.995,"sample":"400 opinions / 31,460 verified-cite contexts, census panel (3 blind judges, controls 7/7), 188 deduped emissions","measured_at":"2026-08-21","note":"measured on the pre-v8 comparator; the v8 fixes cut emissions 201->55 on the same sample (residual dominated by wrong-RESOLUTION and historical-OCR classes); re-measurement rides the quarterly rebuild","recommended_tier":"review"},"citation_year_mismatch":{"precision":0.667,"recall":null,"sample":"21 deduped emissions, same census panel","measured_at":"2026-08-21","recommended_tier":"review"},"case_name_variant":{"precision":null,"recall":null,"note":"advisory BY DESIGN — a spelling nudge, never a wrong-case claim; blocking precision is not applicable","recommended_tier":"advise"},"caption_reversed":{"precision":1.0,"recall":0.968,"sample":"TWO blind 3-judge panels, majority vote, 2026-09-09, 30 DISTINCT emissions between them (panel 2 re-harvested the same opinions, so 9 of panel 1's 10 recur and are counted ONCE). Panel 1, before sovereign captions were read: 10 emissions, 10/10; controls 15/15 reversed and 15/15 correct-order. Panel 2, sovereign captions read: 29 emissions from 2,500 judicial opinions (63,099 verified captioned citations, 1 in 2,176), 21 of them sovereign, 29/29; THREE control classes unanimous across all three judges — 12/12 constructed reversals, 12/12 correct-order, and 12/12 TRAP controls substituting a DIFFERENT sovereign, which a judge must call a different case. The packet carries the two captions and NO citation, because a citation field let a judge tell emissions from controls. Pooled over distinct items 30/30, Wilson 95% [0.89, 1.00]. RECALL 0.968 through the SHIPPED path on a seeded 4,000-caption population (tools/party_panel/recall.py; 0.708 before sovereign captions), ZERO false fires on the correct-order twins. ALIASING TRAP (tools/party_panel/trap.py, committed and seeded): the same sovereign reversed fires 6,712/6,807; a different state 0/6,807; a different designator 0/6,807; a state SHARING a token (North/South Carolina, Virginia/West Virginia, New York/New Mexico) 0/100","measured_at":"2026-09-09","note":"REVIEW class, never the fabrication class: the case exists and the citation resolves. The pooled floor is 0.89 — it does NOT clear the 0.90 a blocking tier would want, which is why this finding demotes and asks for a look rather than accusing. It fires only where the lane reads BOTH party names from the document's own caption; end-to-end recall on running prose is lower than 0.968 and is not claimed","recommended_tier":"review"},"sovereign_mismatch":{"precision":0.71,"recall":1.0,"sample":"REVIEW class (1.20.2): the document names a sovereign of a different FAMILY from the reporter's caption (federal vs state) — or a different state where the old suppression passed it by a generic word — and the other party matches. TRAP (tools/sovereign_panel/trap.py, 400 real sovereign captions, committed and seeded, on the shipped predicate): the caption as printed 400/400 clean; the alias ('State' for a state name and the reverse) 242/242 clean; 'Government' for the United States 158/158 clean; a state or 'State' written for the United States 316/316 flagged; the United States written for a state 242/242 flagged; a different state name 28/28 still the accusation; a wrong OTHER party beside the sovereign 197/197 still the accusation; 'United States' for 'United States of America' 54/54 clean. EMISSION RATE on real prose (the shipped path over 3,000 judicial opinions): 75,598 verified captioned citations, 27,051 with a sovereign party, 7 findings (0.03%; 1 in 3,864), every one read and real. BLIND PANEL, three judges on three models, no label and no citation field, the legal question (same / compatible / different sovereign): 12/12 captions as printed SAME, 12/12 aliases COMPATIBLE, 12/12 constructed swaps DIFFERENT — all unanimous; findings 5/7 unanimous, Wilson floor 0.36. The two disputes are one shape — a state written where the reporter has a corporation while the United States side matches — where the judges said the sovereign slot matched: the findings stand and the evidence now diagnoses them as a DISPLACED party rather than a federal-vs-state swap. Precision 0.71 is the panel's figure on the question as asked, published with the disagreement named","measured_at":"2026-09-09","note":"REVIEW class, decided by the panel: the floor is 0.36 at n=7, nowhere near an accusation tier, and the class stays review until a larger harvest and a second panel say otherwise; the marker and the evidence do not change with the label","recommended_tier":"review"},"rule_absent":{"precision":0.89,"recall":1.0,"sample":"court_rules (1.20.0): manual_review_required on a federal rule citation whose number the named ruleset does not hold. THREE blind 3-judge panels on three models, the legal question, no label and no detector answer; the last over ALL 9 findings a 12,000-opinion harvest produced (9,328 named federal rule citations, 1 in 1,036): 12/12 correct-order controls and 12/12 constructed absents unanimous, findings 8/9 unanimous, Wilson 95% floor 0.56. The ninth (Bankruptcy Rule 7011) all three judges called held by inference from the 70xx numbering; the Office of the Law Revision Counsel's published Part VII holds no 7011, and that publication is the lane's store. The disagreement is recorded, not resolved in the lane's favour. RECALL 60/60 constructed absent numbers through the shipped path (tools/rules_panel/recall.py, committed, seeded), 0/60 false fires on the originals; a decimal under a held integer is a chosen non-fire (0/60 by design: the integer rule is verified, the suffix unread)","measured_at":"2026-09-09","note":"REVIEW class by the floor AND by construction: the store is the current ruleset with no renumbering history, so a correct 1982 citation to Bankruptcy Rule 752 is absent here too, and one signal never accuses. The evidence names a sibling ruleset holding the number (`elsewhere`), which is the commonest real error","recommended_tier":"review"},"supports":{"precision":0.98,"recall":null,"sample":"two blind 3-judge panels, majority vote, 52 emissions pooled: (1) 15 over a 409-record template corpus — 14/15 (the miss a form fragment, closed); false controls 15/15 per judge, true controls weak (syllabus coverage, raw figure); (2) 37 over 52 judicial opinions (CAP full text, real court prose) — 37/37; false controls 15/15 and paraphrase-pair true controls 14/15 per judge, so the panel was calibrated both ways and the corrected figure equals the raw. Pooled 51/52 = 0.98, Wilson 95% [0.90, 1.00]; by overlap band 0.6-0.8 23/24, >=0.8 28/28","measured_at":"2026-09-08","note":"a lexical match of the sentence's proposition to a court-authored description of the case; never a legal conclusion; advisory by construction","recommended_tier":"advise"},"related_insufficient":{"precision":0.4,"recall":null,"sample":"120 lane non-matches over two panels: judges found a paraphrase of the proposition among the descriptions in 68 (template corpus 41/60; judicial opinions 27/60) — as a 'no support found' signal the non-match was wrong more often than right","measured_at":"2026-09-08","note":"WITHDRAWN as a label since 1.16.3 (the measurement is kept here so a pre-1.16.3 stored record's entry can be read); the block verdict remains with the closest description quoted, for the reader — a block verdict is never a label","recommended_tier":null},"not_found_in_covered_volume":{"gate":"zero-false-positive release gate: every release re-proves zero false accusations over a fabricated/real probe corpus; field FP classes are fixed same-day and disclosed","recommended_tier":"block"}}}