# ADJUDICATION-LAW — the comparison laws (published; disputable by design) **NORMALIZATION-1 (Tani 2026-09-13T18:41Z):** a method artifact presented as a directory defect is exactly what we refuse to sell. **PARSE-ADDRESS-1 (Tani 2026-09-13T19:07Z):** the reading must be what the authority actually publishes — never a truncated rendering that hides it. These laws are part of the package — a reviewer may dispute any of them, and the dispute is a method ruling, not a data change. The machinery (`scripts/engagement-findings.mjs`) implements exactly these laws; the bundle's `comparison_laws` block restates them verbatim. **NUMERIC-CLAIM-1 (Tani 2026-09-15T05:21Z, standing):** every numeric delta/count claim in this law text is stated in the canonical name-bound form — `computed flips v→v = ` or `computed counts v = VERIFIED / CONTESTED / UNVERIFIABLE / STALE` — and package-check leg [18] verifies each claim against the computed value it names, by name, or the build fails. Prose contradicting computation inside a signed artifact is the A3 class (Astra) — ours to disclose, not theirs to find. **ADJUDICATION-2 (Tani 2026-09-14T06:44Z):** no comparison is asserted before its preconditions hold. A comparison that fails a precondition returns UNVERIFIABLE naming the precondition; both readings are preserved on every row, including precondition failures. ## Comparison preconditions — v4 (standing in v5; ADJUDICATION-2, unchanged by NPI-ROLE-1) **(a) SUBJECT BINDING (blocking).** The kernel's Tier-A rule (CROSSWALK-DETERMINISTIC-1: Tier A asserts on identifier equality; name-match is Tier-B). For this engagement the ONLY Tier-A link between the board record and the NPPES record is **MO-taxonomy-license equality after normalization**. A name match alone — however exact — is REFUSED as subject binding: no engagement-only name-match shortcut. ROW-012 and ROW-030 (names agree on both sides, no identifier link) revert to UNVERIFIABLE with reason "binding not established". Identity-linkage mismatches (two different published persons, no Tier-A link) are CONTESTED with finding_class identity-linkage — the actionable ticket class — and their dependent address adjudication is blocked. **(b) FIELD ROLE.** Address readings carry roles: board = address_of_record, NPPES = practice location (LOCATION) or mailing (MAILING), buyer = published practice location. A NPPES mailing address is never contradiction fodder. Role mismatch is not a contradiction. **(c) TEMPORAL SCOPE.** Every reading carries its as_of; every comparison reason states the corpus windows (board drop 2026-08-10 vs NPPES fetch 2026-09-09 vs buyer publication frozen 2026-09-13 — non-contemporaneous readings; difference interpretation is bounded by that window). **(d) NORMALIZATION.** City-component alias expansion: ST→SAINT applied ONLY in the city field, never in a street suffix. Unit-omission is a wave-off, not a conflict: either side omitting a unit the other carries is named as a wave-off in the reason, never counted as a difference. ## Address law — v3 (superseded by ADJUDICATION-2) 1. Case-fold; strip punctuation; collapse whitespace. 2. Standardize street-suffix, directional, and unit-designator tokens to USPS canonical abbreviations (STREET→ST, DRIVE→DR, NORTH→N, APARTMENT→APT, SUITE→STE, BUILDING→BLDG, …). 3. The compared board line is the **FULL published line** — the second address line the board publishes (units; or the name-shape street) is INCLUDED. A buyer address omitting a unit the board publishes is a real difference, named in the reason. 4. Compare ZIP5 separately — a 9-digit ZIP against a ZIP5 is not a difference. 5. Match if the standardized forms are equal, or equal with all whitespace removed ("LA QUESTA" ≡ "LAQUESTA"). 6. A remaining real difference — different street number, different street, different ZIP — stays CONTESTED. **Name-shape publications (PARSE-ADDRESS-1):** for some registrants the board's `ba_address` field genuinely carries a practice name where a street address is expected — raw DEN.TXT line 1795 (licence 2017017505): `ba_address = "Elevate Dental Studio,304 W Weaver Road"`. **Verdict: case (b)** — the board's own publication carries the business name; our parser preserves both parts (street + unit) and takes no wrong column. 145 of 4197 corpus rows carry this shape. These rows are rendered and compared AS PUBLISHED, and their findings carry the different sentence — "the board publishes a practice name where NPPES publishes a street address" — never a mere "the two authorities disagree." ## Address law — v2 (superseded by PARSE-ADDRESS-1) USPS standardization and ZIP5 comparison (the NORMALIZATION-1 fix), but the board reading was STREET-ONLY: the second address line the board publishes (units; the name-shape street) was dropped from the rendered reading — rendering "Elevate Dental Studio" as the board's whole address. v3 compares the full line. ## Address law — v1 (superseded by NORMALIZATION-1) Exact match after case-fold and punctuation strip. It counted "2001 Laquesta Dr, Neosho, MO, 64850" vs "2001 LA QUESTA DR, NEOSHO, MO, 64850" — the same address — as CONTESTED. That was the finding. ## Name law (v2, unchanged by PARSE-ADDRESS-1) 1. Case-fold; strip punctuation. 2. Drop generational suffixes and credential tokens (JR SR II III IV DDS DMD MD PC INC LLC PLC PA). 3. Compare FIRST and LAST given-name tokens. Middle names, case, and credentials are registry formatting, not contradictions. 4. A different first or last token is a real difference. **Known boundary case (stated, not resolved):** ROW-023 — board "Jerell James Wilson" vs NPPES "JAMES WILSON". Plausibly the NPPES first name is the board middle name presented as first; not provable from the pinned corpora. It stays CONTESTED with both readings; the reviewer rules. ## Licence law (unchanged) Compare digits with leading zeros stripped ("013129" ≡ "13129"). ## NPI law (unchanged by v5) Compare the 10-digit NPI string exactly. ## NPI entity-role law — v5 (superseded by DETERMINACY-1; NPI-ROLE-1 RULED, Tani 2026-09-14T15:15Z) **The ruled field contract:** the buyer's NPI field denotes the individual practitioner (NPPES entity type 1). An organization NPI (entity type 2) in that field is a **ROLE finding with its own reason** — never a name mismatch, never an identity-linkage ticket — and **every NPI reading cites the entity type** (enumeration_type). A named practitioner and a group can legitimately be linked; what the ruled contract makes erroneous is an organization registration occupying the individual-practitioner field. Resolves ROW-052 and ROW-010's org-NPI question (R5-3 completion). **v4→v5 delta line:** the rule changes reasons and readings, never states — computed flips v4→v5 = 0 (`flips_v4_to_v5_total` in findings.json law_delta, computed by the run, never asserted) and computed counts v5 = 59 VERIFIED / 22 CONTESTED / 18 UNVERIFIABLE / 1 STALE (identical to v4). ## Determinacy — v6 (current; DETERMINACY-1, Tani 2026-09-14T18:08Z — Astra R5 static review R6-4) **One decision procedure, explicit precedence, refusal outcomes.** The v4 preconditions stand unchanged; v6 adds three determinacy rules and the precedence that orders them: - **(e) TWO-PERSONS DETERMINACY.** Different names plus an absent identifier link does **NOT** establish "two different persons" — it establishes only that **the buyer's association is unsupported**. The finding states that narrower proposition (finding_class identity-linkage — the actionable ticket class) and never asserts two persons. A two-persons claim requires identifier-grade evidence from the authorities themselves. - **(f) ADDRESS COMPARABILITY.** `address_of_record` and practice location are comparable **only** when both readings bind the same predicate class, the same authority scope, and the same jurisdiction, with roles and as_of dates stated. When they are not comparable, the emitted result is **UNVERIFIABLE naming the incomparability** — never CONTESTED. - **(g) DATE-DIFFERENCE DETERMINACY.** A date difference supports contradiction **only** when both readings issue from the same authority scope within the same publication cycle. A difference between corpus dates (different fetch windows) emits the **temporal-scope note** — publication lag or a legitimate move remains possible — and NEVER itself a CONTESTED verdict. **Precedence on overlap:** (a) binding failure outranks everything (UNVERIFIABLE, precondition named); (b) incomparability (f) outranks comparison; (c) date-only differences (g) resolve to the temporal note, never a conflict. **Falsifiers (ruled):** a legitimate name change without a Tier-A link, and a legitimate address change between corpus dates, must NOT become a conflict. **v5→v6 delta line:** v6 constrains what a finding may SAY — emission determinacy — never its state; computed flips v5→v6 = 4 (`flips_v5_to_v6_total` in findings.json law_delta, computed by the run, never asserted) — the four flips are ROW-021, ROW-028, ROW-033, ROW-019 (practice_address VERIFIED→CONTESTED): under v6 rules on the parser-fixed corpus "Missouri" vs "MO" compares as a mismatch because STATE-ALIAS-1 is v7 law — a method artifact waved off in v7, never a regression in v6 — and computed counts v6 = 55 VERIFIED / 26 CONTESTED / 18 UNVERIFIABLE / 1 STALE. ## State-component alias law — v7 (current; STATE-ALIAS-1, Tani 2026-09-15T00:33Z — REPRO-DRIFT-1, the empty-environment reproduction gate) The clean-environment reproduction caught the shipped findings drifting from the regenerated output: four practice_address propositions (ROW-021, ROW-028, ROW-033, ROW-019) flipped VERIFIED→CONTESTED solely on the state component — "Missouri" vs "MO" — after the parser fix made the board corpus publish full state names. **Missouri and MO are the same state.** Selling that string difference as a directory defect is exactly the NORMALIZATION-1 method-artifact class (Tani 2026-09-13T18:41Z: "a method artifact presented as a directory defect is exactly what we refuse to sell"). **(h) STATE-COMPONENT ALIAS EXPANSION.** The state component is normalized (USPS full name ≡ two-letter abbreviation) before comparison — applied to the STATE FIELD ONLY, the same rule shape as the city ST→SAINT expansion. A state-component alias difference is a WAVE-OFF named in the reason, never a CONTESTED. A genuine cross-state difference (Kansas vs Missouri) remains a CONTESTED finding with the state named. **CROSS-STATE-1 (R7-8 rerun, Tani 2026-09-15T00:33Z).** The EXTRACT-1 parser fix landed (the parsed board reading now carries the board's true state), and the rerun at OUR expense (SERVICE-DEBT-1 discharged) emits the findings the blocked comparison owed: ROW-034 (buyer publishes the practice in Missouri; the board's address of record is Leawood, KANSAS) and ROW-048 (buyer publishes the practice in Missouri; the board's address of record is Meridian, IDAHO) adjudicate CONTESTED cross-state with both states named — a genuine cross-state difference is a buyer-facing finding, never an extraction defect. **DIFF-SIM-1 (disclosed, 2026-09-15).** R7-3's hand-rolled v5 emission simulation was found producing 56/6/38/0 — 30 spurious state flips against the ruled v5 — hidden by a counts-chain alias. It is withdrawn; the v5 baseline is the ruled v5 state rules, and the v5 emission wording simulation remains outstanding work (a state differ cannot measure emission differences — this artifact does not claim it can). **REPRO-PARITY-1 (standing law, Tani 2026-09-15T00:33Z):** the package check compares the regenerated output against the shipped signed findings and FAILS on any difference — content-identical or it does not ship. A reproduction command that produces different findings than the signed delivery is worse than no reproduction command, and it is exactly Astra's falsifier. **v6→v7 delta line:** the v6 state rules compared the state component as a raw string and blocked the EXTRACT-1 rows; computed counts v6 = 55 VERIFIED / 26 CONTESTED / 18 UNVERIFIABLE / 1 STALE. computed flips v6→v7 = 6, each named in `flips_v6_to_v7` with its reason: four `STATE-ALIAS-WAVEOFF` flips (ROW-021, ROW-028, ROW-033, ROW-019 — "Missouri" vs "MO" is the same state) and two `CROSS-STATE-RERUN` flips (ROW-034, ROW-048 — the rerun emits the cross-state findings). computed counts v7 = 59 VERIFIED / 24 CONTESTED / 16 UNVERIFIABLE / 1 STALE. **v7→v8 (R8-1 LAW-IN-THE-BRANCH-1, Tani 2026-09-15T04:30Z, blocking):** the law is enforced in the EMITTING branch, never in prose. Two forbidden assertion patterns are now unrepresentable at emission and gated at build (package-check leg [16]): (a) "two different published persons" — with no Tier-A identifier link the corpora do not establish two distinct persons (a name-presentation variant remains possible; ROW-023 is the boundary case); the narrower unsupported-association proposition is the only admissible emission; (b) an authority-disagreement assertion on a practice address whose delivered NPPES object carries the board-like address under the MAILING role — role mismatch is not a contradiction (ADJUDICATION-2 (b)); that is the role-comparability REFUSAL case (ROW-001 — a REAL row; the refusal fires on NO synthetic fixture row, re-verified R8: ROW-056's fixture NPPES record carries both a LOCATION and a MAILING address, but its labeled defect is licence_number — the row never reaches the address cross-comparison). v8 changes states where the refusal applies — computed flips v7→v8 = 1 (ROW-001 practice_address CONTESTED→UNVERIFIABLE); computed counts v8 = 59 VERIFIED / 23 CONTESTED / 17 UNVERIFIABLE / 1 STALE; counts are recomputed, never carried over.