[MaxGhenis]

Round 8 approved the head, a9aae687, with no high or medium finding, and CI passes there. CE1 of the covered-earnings v2 design (docs/design/covered_earnings_v2.md §2.1): a versioned, cited registry of OASDI coverage rules for income years 1968–2026, so that CE3 can classify a component-year from its facts. Every case decides at least one invented worker, and when a covered case's listed condition fails, the worker reaches the exclusion or residual its source gives. Each family uses one date axis, and where a boundary's clauses mix axes, the later row pins the one that governs. The registry's first-match rule resolves 13 of the 17 published examples, and the two section 218 re-entry and two gross-concept examples compute from parameters.

What it adds

• src/populace_dynamics/income_resources/coverage_rules.py holds the row, case and citation types, schema validation (coverage_rules.ce1.v3), the capture-hash check and the lookup by family and income year.

  • The lookup refuses with a typed error in a gap, and the loader refuses an overlap between rules.
  • A row may carry cases, each with a case_id, a precedence, trigger-fact conditions and a consequence. Precedence runs 1..n and the first matching case decides; the row's residual default applies only when no case matches (CoverageRule.resolve).
  • A deferred consequence names, in defers_to, the families that decide instead. The loader refuses a deferral with no target, to an unknown family, to the row's own family, or around a cycle.
  • An open case (an unsettled legal sub-question or a missing fact) names its readings, and resolve refuses it with OpenRuleError.
  • The module raises separate error types: RegistryValidationError for bad registry data, UnknownIdentifierError for a name the caller got wrong, and the legal refusals for gaps and open questions.
  • The registry and its captures load from separate roots.

• data/external/income_resources_ce1/ holds the generator assemble_registry.py, the fragment files it reads and the registry coverage_rules.json it writes. It lives outside the snapshot tree, which .gitattributes marks binary, so review diffs show it.

• data/external/snapshots/income_resources_ce1_v1/ holds the pinned public captures: statutes, page-range extracts of Statutes at Large, the Federal Register and IRS publications and forms, past U.S. Code section editions, the CFR, POMS, and an Internet Archive copy of SSA's coverage-threshold table. Each has its URL, retrieval time, size and SHA-256 in a manifest, and every file in the folder has a manifest entry.

• The registry holds 119 rules in 58 families, 21 threshold series, 5 table definitions and 17 published worked examples. 90 rules carry 364 cases, 34 of which defer to another family. 38 cases leave a legal sub-question open, with 76 named readings, and the 8 open rows name 16 more (92 in all). One case in each clergy row refuses a worker whose facts don't establish an exemption status.

• Every rule carries identification_class: awaiting_ce2. Each rare exclusion carries both readings of design question 1: strict, and negligible_candidate, which holds null wherever no search found a published bound. No search found one, so every rare exclusion still reads strict: unresolved with a null candidate, and each row keeps its search records. CE1 chooses neither reading.

• 8 open legal rows refuse unsettled historical questions. Each names its readings, has a null consequence and no chosen reading, and the registry's open_questions lists the 8. Where the law leaves only a sub-question unsettled, the row returns the verified rule and an open case carries the sub-question.

• Of 3,422 family-years, 155 refuse (343 before the round-2 fix, 197 before the round-3 fix).

• No family reads 410(a)'s base definition. scope_gaps names the three clauses of 410(a) that reach service abroad and records the service abroad that none of them reaches:

  • (A)(ii): service on or in connection with an American vessel or aircraft, irrespective of citizenship or residence;
  • (B): a citizen's or resident's service for an American employer, or under (B)(ii) for a foreign affiliate while a 3121(l) agreement is in effect;
  • (C): service a section 433 agreement designates, in taxable years beginning on or after 1983-04-20.

  The four federal residuals name the same clauses.

• The earnings-test concept accessor refuses every year with unverified dependencies.

• The birth-evidence seal lists coverage_rules.py as an exclusion, beside master's exclusion of the package's __init__.py, and a reachability check confirms that the birth-evidence reducer reaches neither module.

Two rows follow their captured sources against an earlier review:

• Passive income uses payment dates. The captured 42 U.S.C. 411 notes make both boundary amendments effective on calendar dates and state no taxable-year rule.
• Vocational nonresident aliens have closed rows for 1968–1986. Pub. L. 100-647 sections 1001(d)(2)(E) and 1019(a), with the Tax Reform Act of 1986 section 151(d), make both readings agree before 1987. From 1987 to 1988-11-09, only authorized M-1 service for the admission purpose stays open.

Invariants

• For every family and income year 1968–2026, the lookup returns the rules in force or refuses with a typed error.
• Ranges and axes:
  • Within a family, effective ranges partition the years the family covers, and the loader refuses an overlap.
  • Every row of a family uses one date axis: service_date, taxable_year_start or payment_date. Any axis but the default quotes its effective-date clause from a capture the row cites.
  • At each internal boundary, the effective-date clauses that the two rows' captures print for that date all use the family's axis. Where they mix axes, the later row pins the governing clause, a clause on another axis carries an axis_exception, and the earlier row's cases carry the off-axis date; no code reads the exception.
• Cases:
  • A row's case precedences run 1..n, the first matching case decides, and the residual applies only when none matches.
  • A case ranks above every case it overrides and gives a different consequence.
  • The fixture states each INVENTED sweep worker as the exact condition texts true of that worker, and every case of every row decides at least one of them. So rewording or narrowing a case, re-ranking it below another case its worker also meets, or giving it another consequence fails the worker it decides, and a deferred worker pins its target. Broadening a case fails a test only where some worker just misses the case on the dropped condition. 120 listed conditions each have one, and for 29 of them that worker states the missed fact with a text no case of the row reads. 169 of the 511 conditions of multi-condition cases have none, and a test fails if that count rises. When one of the 25 listed covered statutory conditions fails, the worker reaches the exclusion or residual its source gives by a named path. A case reaches only the people its source reaches by level of government. Some case reads each trigger fact a row lists, or the row names its reason. condition_cases ties each parameter that states a condition with a word the guard lists (WORDED_CONDITION's words, "only" or "limited") to the cases, thresholds or reason that account for it. On every closed row, condition_cases ties each parameter that names a condition or a scope to the cases that carry it, to thresholds, or to a reason it sets no coverage condition.
  • The fact-pattern tests match a condition only when a text the fixture states for the worker equals it.
  • A case-free closed row that names a condition carries case_free_basis: an explanation and a citation quoting a capture the row cites. No parameter record holds a when or if key.
• State and local reach: in the section 218 and election families, every case that excludes, sets a threshold or defers names the level its law reaches: a State (the 50 States, Puerto Rico and the Virgin Islands under 410(h)) or a political subdivision or wholly owned instrumentality, other than the District of Columbia, Guam and American Samoa. The mandatory cases rest on 410(a)(7)(F), agreements on 418(b)(1), and the general exclusion on 410(h) with the family's scope. Each such election case names election work, and a private worker, a District of Columbia employee or other State or local work reaches the residual.
• Federal reach: every federal case that excludes or stays open names service for the United States (410(a)(5)-(6)).
• Deferrals: every deferral reaches a family with a rule in each deferring year. Where that family's rule stays open, the deferral refuses, and the deferring row records the target, years and reason in deferral_exceptions.
• Open questions:
  • An open legal row or open case names two or more readings, carries no chosen reading, and refuses.
  • A closed row's named options, at any depth, say which one governs.
• Bad registry data, a caller's unknown name and a legal refusal raise disjoint error types.
• Citations and captures:
  • Every rule cites a pinned capture, every cited hash matches its file, and every citation resolves inside the capture root.
  • Every supporting passage holds nonempty text, and published amounts sit in their cited passages.
• FIELD_PURPOSES or a listed CE1 extension names every trigger fact.
• awaiting_ce2 never reads as identified.
• A negligible_candidate exists only with a cited source, bound and population.
• Thresholds:
  • Thresholds that the law makes monotone never fall.
  • Quarterly and monthly thresholds aggregate within one employer-employee relationship.
  • Source amounts keep their units and comparisons.
  • CoverageRegistry.threshold checks every supplied index input even when a published value decides.
  • A threshold_id_by_case entry names a case of its row, a known threshold by a string threshold_id, and a first and last income year. Both years fall in 1968–2026 as integers, and the first never follows the last. For every year from the first to the last, the threshold has a published value or an indexing range that covers the year. The loader refuses anything else with a RegistryValidationError.
• The earnings-test accessor refuses unverified years.
• Published examples: CoverageRule.resolve computes 13 of 17; the other four compute from the row's parameters. Each outcome equals the published one and conforms to the lookup that cites its passage.
• Determinism: loading and lookup give the same result on every run and for any row order, and round-trip through JSON. The committed registry equals assemble_registry.build()'s output byte for byte.
• Capture folder:
  • Every file in the folder has a manifest entry whose size and SHA-256 match.
  • Each recorded capture sits in the folder, or its record says why not.
  • Each page extract records its original's file, size, SHA-256, page count and kept pages. It keeps its cover unless a kept page's pinned text states the edition, and a recorded chain of qpdf steps leads from the original to its bytes.

The general checks have Hypothesis properties on invented rows, and the real registry and capture folder get conformance checks over every row; pytest collects each published example as its own test.

Stated limits

Guard gaps (round 7's low 1):

• D3, a reason's pointer word: the reason-pointer guard accepts any pointer word, even "whatever".
• D4 and D5, unread-fact reasons and condition_cases ties: the tests pin only 24 of 365 reasons and 15 ties, so emptying an unpinned reason or retying a parameter to a reason passes.
• D6b's family, parameter texts: the worded-condition guard reads a fixed word list, now with "only" and "limited", and another synonym passes.
• D7, registry prose: the scratch-note guard matches case-sensitively.
• D8, case-threshold years: nothing checks that a threshold_id_by_case entry's income_years match the case's reach.
• D10, threshold indexing: no test reaches an indexed threshold's year bounds.
• D2, federal covered cases' level text: the federal covered-case guard passes any text naming "United States" or "Federal"; other tests catch the mutant.

Guard and fixture gaps (round 8; the reviewer ran the pinned inputs, and none of these changes a row or a published outcome):

• Import-time parses: import_time_registry_parses matches only the spelling json.loads(REGISTRY_ARTIFACT.read_text()), so a second module-level parse written read_text(encoding="utf-8") passes.
• Worded-parameter guard: no test exercises the re.IGNORECASE on PARAMETER_CONDITION_WORDS.
• Threshold ids: removing the loader's isinstance(threshold_id, str) check fails no test. A list or integer id still meets the unknown-threshold refusal, and only a dict id reaches the TypeError the check prevents.
• income_years length: no test gives three income_years, so a loader that accepts three passes.
• Run-together sentences: FUSED_SENTENCE misses a capital before the period, as in "IRS.Publication"; a wider scan of the committed prose finds no such text.
• Scope-gap clauses: SCOPE_GAP_CLAUSES checks substrings, so a scope gap that keeps the clause names and reverses its conclusion passes.
• Just-miss lists: just_miss_required and just_miss_own_text have no count pin, so dropping an entry passes. The 169 bound catches a dropped required condition only once it loses its last just-miss worker, and a dropped own-text entry falls back to the loose test, which accepts an omitted fact.
• overrides: no test checks that a case lists every case it displaces, and resolve never reads the field. se_nonresident_alien_1968_1983 holds the overrides its description claims, with no test to keep them.

Other limits:

• Registry prose: no test reads descriptions or parameter prose, so the prose-only row mutants for dc674bc0, c55ebd95 and 25b02a6a survive.
• Two registry texts (round 8's L-1 and L-4): both sit in prose, and no worker, example or refusal changes. A text-only follow-up corrects them.
  • nonprofit_1968_1977.cooperative_521_small_remuneration says the 1978 row reads former 410(a)(10)'s small-remuneration test as reaching an organization exempt "under 521". Since 4774f453 that row leaves the question open, because 409(a)(14) names only section 501 and 3121(a)(16) names 521 too.
  • The scope gap and the four federal residuals say Pub. L. 98-21 section 321(b) reaches agreements entered into after April 20, 1983. They leave out section 321(f), under which an American employer can elect the same amendment for an earlier agreement.
• Broadened cases: 169 of 511 conditions of multi-condition cases have no just-miss worker, so dropping one fails no test. 38 of them sit in the 66 cases round 7 changed.
• Seven drop-free conditions can't take a sensible just-miss worker (SSR 95-3P's overlapping facts; three rank-implied conditions), and the fixture's just_miss_stated_limits gives each reason.
• clergy_1982_2026.exemption_status_unknown: no just-miss worker states its fact with another text. It ranks last, above an open residual, and round 7 changed only its rank.
• Order rows (round 7's low 9, second half): a vowed member with no election fact reaches the covered residual, because a case matches only facts the worker states.
• FEPCA reemployed annuitants (round 7's low 11): the open case still reaches service after the waiver law's effective date, where both readings give covered, because no capture dates the law.
• Open rows refuse some workers their captures settle: community income 1980–2004, sect 1986–89, 911, health insurance, fishing 1968 and the territories.
• Base definition: no family reads 410(a)'s base definition; scope_gaps records the service abroad that (A)(ii), (B) and (C) don't reach.
• Election rows: the consumer supplies the agreement amount and its period.
• Sibling cases: no guard compares a fact one case reads and its sibling lacks.
• Army corps: the closing rests on the undated 2024 POMS RS 01901.340 C.5.
• Test fixtures: the in-code fixtures stay in code.

Captures

This PR adds the capture folder to master. It holds 30.0 MB in 249 files, and the largest, the 97 Stat. 65 extract, takes 3.4 MB.

Seven whole-act or whole-issue PDFs became page-range extracts. qpdf --deterministic-id --empty --pages made each from its hashed original:

Source	Original	Extract
Pub. L. 92-603, 86 Stat. 1329	31.3 MB, 165 pages	1.90 MB, 10 pages
Pub. L. 90-248, 81 Stat. 821	21.0 MB, 115 pages	0.98 MB, 5 pages
Pub. L. 98-21, 97 Stat. 65	19.0 MB, 108 pages	3.45 MB, 20 pages
Pub. L. 89-97, 79 Stat. 286	24.8 MB, 138 pages	0.76 MB, 4 pages
Federal Register, 10 July 1975	29.4 MB, 172 pages	0.63 MB, 3 pages
Pub. L. 100-203, 101 Stat. 1330	76.4 MB, 473 pages	0.97 MB, 6 pages
Pub. L. 100-647, 102 Stat. 3342	75.7 MB, 471 pages	1.12 MB, 7 pages

The round-2 fix captured the last two from govinfo, and the folder keeps them only as extracts.

Rows cite 38 IRS publications and forms for one value or page, and the folder keeps those as extracts too: 36 editions of Publications 15 and 926 (45.8 MB), the 1985 Form 1040 instructions (4.3 MB) and Publication 51 for 2023 (1.3 MB) became 38 extracts of 3.7 MB in all. The 28 that carried one shared 873,242-byte cover image (Publication 926 for 2008–2025, Publication 15 for 2017–2025 and Publication 51 for 2023) drop it, because a kept page's footer states the edition; the other ten keep their cover.

• What each extract keeps. Each act keeps:

  • its first page;
  • every page of each cited section, from the section's heading to the next section's heading;
  • the page with its approval date.

  The Federal Register keeps its cover and the two pages that hold FR Doc. 75-17887, from its heading to its filing line. Each IRS extract keeps the pages that state the cited value, and its cover unless a kept page states the edition.

• Checks on the extracts. Every kept page's pdftotext output equals the same page of the original, and the text derivatives regenerate byte for byte from the extracts. The scanned 1984 and 1985 Schedule SE pages keep their OCR texts, since pdftotext returns nothing for them.

• The originals. The inventory and manifests record each original's file, size, SHA-256 and page count. Citations that named an original PDF page now give the extract page as well.

The branch also removed these, because nothing cites them:

• IRS Publication 17 for 1999–2007 (20.4 MB). Their inventory rows keep the URL, size and SHA-256, with status not_retained.
• 34 text renderings that no manifest listed, including whole-act text of the three 1968–1983 acts.
• In the round-1 fix, 29 more captures that no rule cites, under one retention rule for every capture: the folder keeps a capture only if a rule cites it or a cited derivative depends on it.
• In the round-2 fix, the 2025 Publication 517 PDF, a duplicate of a cited capture.

This branch's history holds the originals, so this PR must land by squash, and master then never holds them.

Review

• Round 1 (Opus 5.5, at 39b52208): REVISE, with two high, twelve medium and 28 low findings. The two highs found that tribal-council service defaulted to covered and that the registry covered pre-2008 conservation reserve payments with no source.

• Fix (28becb82): the schema's cases and residual defaults and the open rows' named alternatives (M7, M12); the tribal and conservation reserve rows; each row the review checked against its source (M1–M6); employer-employee threshold scopes (M8); outcome, path-escape and byte-equality tests (M9–M11); and the low findings and notes. The commit message lists each.

• Corrections to earlier messages (round 1 Low 28): 1be4b753 added 13 worked examples (5 to 18), not nine; the manifest listed the nine Publication 17 PDFs at 1be4b753, and a later commit removed them as redundant and uncited; and the two PDFs 1be4b753 called needed whole kept 3 and 4 pages in the next commit, one of them an act rather than an issue.

• Round 2 (delta, at 28becb82): REVISE, with two high, two medium and 14 low findings and two notes. The two highs found that conditional exclusions with no cases applied the exclusion whatever the facts, and that the registry covered pre-1995-08-07 cash-equivalent farm transfers against the cited POMS. The two mediums found that nothing defined case precedence and that three families refused rules their captures verify.

• Fix (nine commits, 94219ede to df1dc160):

  • 94219ede ranks cases, first match wins (MED-1);
  • 8476dcf7 makes every exclusion condition a case, with deferrals and case_free_basis (HIGH-1);
  • a6fa1f07 makes pre-ruling cash-equivalent farm transfers noncovered (HIGH-2);
  • c0ea2b24 returns verified rules and keeps only unsettled sub-questions open, with evidence-only boundaries recorded as such (MED-2);
  • 5bac444f tests every loader guard and separates data, caller and legal errors (lows 1–4, 14);
  • 01e1e857 names open readings once and settles the vocational years (lows 8–11, 14);
  • 2dad26e6 keeps IRS page extracts and the OCR text of scanned pages (lows 5–7, 13);
  • 38cd1968 states the capture-folder invariants;
  • df1dc160 turns the low 8, 9 and 14 checks into checker functions with Hypothesis properties.

• Correction (round 2 Low 12): 28becb82's message says it removed unsupported monotone flags from 13 constant series; it removed them from 14.

• Merge (ea599b92): master 172f690f. Git auto-merged the two birth-evidence files, which both sides changed; against each parent the merged files only add lines.

• Round 3 (delta, at ea599b92): REVISE, with two high, six medium and ten low findings and four notes. The two highs found that an unlisted existing employee under a nonprofit waiver fell to a covered residual and that the public-office exclusion could never match.

• Fix (nine commits, a645009c to 2472668e):

  • a645009c excludes unlisted existing employees under an actual waiver; a deemed waiver covers everyone (H1).
  • 35713143 conditions the public-office exclusion on a public-office fact (H2).
  • 819efef9 puts every statutory condition in a case's when on the review's rows, ties the condition parameters of 32 rows with cases to their cases, and extends the detector to those rows (M1).
  • f042289b gives the community-income and 1984–1985 units rows cases, makes case_free_basis a quoted record, and refuses when records in parameters (M2).
  • 598b2462 moves five families to payment dates and eleven to taxable years, and adds a boundary-clause check (M3).
  • 9eb28f89 answers both CRP windows with one open case, narrows the vocational dispute to M-1 admission-purpose service before November 10, 1988, and adds section 931 to the clergy gross concept through the 1987 amendment (M4–M6).
  • 46f3af78 takes lows 1–5 and 10.
  • 45ef91d3 drops the unused alternative mechanism (note 3) and takes lows 7 and 9.
  • 2472668e makes Pub 51 (2023) a page extract (low 8) and records one retrieval of 97 Stat. 65 (note 2).

• Where the fix follows the sources over the review:

  • Nonprofit: deemed waivers cover every employee (RS 01901.540 D.7), so only an actual certificate leaves an unlisted employee excluded.
  • Clergy abroad: the case keys on nonresident-alien status (411(b), Publication 517), not citizenship.

• Refusals: 155 of 3,422 family-years refuse, down from 197; CRP's 40 and the vocational 2 now return rules with open cases.

• Corrections to earlier claims:

  • "Every deferral reaches a family that decides" overstated the registry: three deferrals reach refusing families, and the deferring rows now record each one in deferral_exceptions.
  • "Every mutant … fails a named test" overstated the tests for the review's o1, o3, o4 and o15, which the new tests and guards now target.

• Round 3, rest of the fix (three commits, fc1d7e04 to 195be7b8):

  • fc1d7e04 takes low 6's eleven narrow omissions:
    • a limited partner's 707(c) payment not for services stays excluded;
    • the registry excludes section 7873(c)(2)'s de minimis amount;
    • 410(a)(4)'s clause (A) tests the employment, as POMS RS 01901.170 B reads it;
    • the federal rows carry Pub. L. 98-369 section 2601(c), (d)(3) and (e);
    • the registry also covers a nonmember election worker under a lower agreement amount at the statutory annual amount;
    • the foreign-agricultural row cites Publication 51 (2023) section 3;
    • the 1968 tribal row no longer lists the posting fact;
    • five rows' case wording, locators or fact keys now follow their sources.
  • 16b208fa drops the shared cover from 28 IRS extracts whose kept pages state their editions (note 1). The capture folder falls from 59.5 MB to 29.7 MB.
  • 195be7b8 corrects two of fc1d7e04's descriptions: the 1989 redesignation of 409(j), and what the 1954 note on 410(a)(4) says.
  • The review's mutants o1, o3 and o4 now fail the tests that round 3's low 1 added.

• Where the fix follows the sources over the review:

  • Foreign agricultural workers: 410(a)(1) and POMS key on the worker's admission, while Publication 51 (2023) names only agricultural labor in connection with the visa. An open case with both readings takes the worker's other service.
  • Tribal councils: POMS RS 01901.700 C.1 says council service is not employment, and its posting note sets a crediting rule. So round 3's tie stands, and the row no longer lists the posting fact.
  • Federal, from 1987: a section 2601(d)(3) delay ends before 1987 unless Federal employment elsewhere continued after July 18, 1984; the 1987 row records that narrow case without a case.

• Correction: fc1d7e04's message counts eight outcome and record tests; it adds seven, besides the Hypothesis property.

• Round 3, sibling rows (seven commits, f2409b26 to fd5b7b2a):

  • f2409b26 splits election_work_1995's lower-amount case by membership, as the 1991 row's: POMS SL 30001.357's NOTE states no end date.
  • 07cc9cfc places the 1978 home-worker cash test in former 409(j) until the 1989 redesignation, and words the case with 410(j)(3)'s contract and transportation-facility clauses.
  • 45f22320 drops the posting fact from tribal_council_service_2018_0920; the posting rule reaches tax years through 2005 only.
  • c9885558 and fd5b7b2a add former 410(a)(1)(A), the title V contract-worker clause, to foreign_agricultural_service_1968 for service before July 18, 1984. The row records that no capture says whether any title V contract remained in force after 1967.
  • e51e708b applies POMS RS 01901.720 C.2–C.3 (TN 62 (08-24)) to student service from 2024-08-15, keyed to the service date, with the determination-date reading recorded.
  • 6e33faed has both federal rows quote Pub. L. 100-647 section 8015(c)(3) as the 2024 Code note under 26 U.S.C. 3121 sets it out, beside LII's paraphrase.

• Where the fix follows the source over the brief: the captured POMS edition starts the student criteria on 2024-08-15, so service earlier in 2024 keeps the statutory test.

• Round 4 (delta, at fd5b7b2a): REVISE, with three high and five medium findings, ten lows and three notes. All three highs share one shape: an exclusion no real fact value selects, so the worker fell to a covered residual. They concern the nonprofit waiver period, the 3121(r)(3)(A) member condition and the 2018 tribal row's federal recognition.

• Fix (nineteen commits, 00df716a to 210a5c85):

  • 00df716a holds nonprofit waivers to the certificate's period and makes the last exclusion reach every 501(c)(3) employer with no certificate in effect (H-A). It splits the actual-certificate cases by hire date and opens the retroactive-period hire on which RS 01901.540 C.4 and NL 00902.500 NPO505/NPO506 disagree (M-2).
  • da09aab8 gives a member who fails 3121(r)(3)(A) a noncovered path (H-B).
  • 64255e0e requires federal recognition for the 2018 tribal exclusion (418a(b)(2); POMS RS 01901.700 F) (H-C).
  • 8f4ab88f words both fishing rows' crew test as "normally made up of fewer than 10 individuals" for each boat, with the average as the deeming rule, and the pre-1996 bar as cash only (M-3).
  • 8ec7f152 puts 411(b)'s nonresident-alien exclusion in the family self_employment_nonresident_alien. Its territorial and section 433 exceptions each run from their date, and the clergy rows agree (M-4).
  • 8bc0bab9 makes the axis checker require a pin wherever the printed clauses mix axes, pins all 17 such boundaries with four exceptions, gives domestic 1995's under-18 case its service date (M-1), and corrects 18 axis attributions (low 8).
  • a431176b refuses parameter records keyed if (M-5) and fixes a stale docstring (low 1).
  • ff7ceef3 tests that failing each covered required condition reaches its exclusion, for 22 entries (M-5).
  • a9fccaee, d8a44dc6, 1b2f58b5 and 92d2f409 fix 33 defects the exclusion sweep found across 107 rows (item 9; table in the lane's exclusion-sweep.md).
  • 200f24d4 ties all 149 scope parameters in 80 rows and adds the seasonal hand-harvest exclusion (M-5).
  • a0e2e23f takes lows 2–7 and five low-10 items; 8968b124 restates one fixture phrase.
  • 94a9af84 holds 20 State and local cases to State and local service: state_local_1991_2026's three mandatory-rule cases, the election family's 16 threshold and noncovered cases, and the Pub. L. 90-248 section 122(c)(2) election. Each now names the level 410(a)(7)(F), 418(b)(1) or 411(c)(1) prints, and each election row ties it to a new scope parameter, so a private worker or a District of Columbia employee reaches the residual.
  • 7d6f7a72 holds the 16 election cases to "election official or election worker" (418(c)(8)(A); 410(a)(7)(F)(iv)), so other State or local work no longer meets the election threshold.
  • e54479dd opens a qualifying insurance termination payment before 1998 (low 10). 411(j)'s heading calls it a "Codification of treatment" and its text construes 411(a), but the 411 notes date it "applicable to payments after Dec. 31, 1997", and no capture records the earlier treatment. The open case names an effective-date reading and a codification reading.
  • 210a5c85 adds a Hypothesis property: for each of the 22 cases, a worker with the case's own conditions meets it only at the State or local level.

• The sweep: 543 exclusions across 102 swept rows, plus the five H rows. 46 defects: the fixes resolve 44, 1 becomes an open case and 1 stands on an axis exception. 94a9af84 fixes the tie drafts' late finding, the government level that state_local_1991_2026's mandatory-rule cases and the election family's cases lacked.

• Mutants: the follow-up's tests kill all 15 of its mutants, among them one that drops the level from mandatory_rule_exclusions and one from each row's outside_agreement_member.

• Where the fix follows the sources over the review:

  • Axis count: 17 mixed-axis boundaries, not 18; vocational 1987 already had its pin.
  • Overlap rows: an open case carries M-2's overlap, not a ruling, because no capture holds former 410(a)(8)(B)'s text.
  • Election work: the election cases lacked an election-work condition as well as the level; 7d6f7a72 adds it from 418(c)(8)(A) and 410(a)(7)(F)(iv).
  • Low 10: the heading becomes an open case, not a ruling, because no capture records how 411(a) treated a qualifying payment before 1998.

• Corrections to earlier messages:

  • The capture folder held 29,667,828 bytes at 16b208fa, not the 29,665,513 its message gives.
  • The checker of 598b2462's time flagged 24 boundaries, not the 21 its message gives.
  • 45ef91d3's "passes both … cases" matches one case.
  • 200f24d4's ties split 68, 18 and 63, not 69, 18 and 62.

• Round 5 (delta, at 210a5c85): REVISE.

  • Highs (7): the clergy deferral left out inactive vowed members (H-1); the nonprofit rows missed supplemental lists, group joiners and 3121(k)(4)(C) (H-2 to H-4); the base definition missed guaranteed payments (H-5); federal cases lacked the federal level (H-6); and the 2601(d)(3) delay stopped at 1986 (H-7).
  • Mediums (7): the pins route nothing; the test oracle (5 of 13 mutants survived); CNMI; agreements outside the clergy rows; Army Corps aides; 105(b); the Invariants.
  • Also: ten lows and three notes.

• Fix (eight commits, ad404489 to 486453fc):

  • ad404489 excludes every vowed member's order duties and keeps 3121(r)(2) only in the election's covered case (H-1).
  • c85029d6 holds supplemental lists to their filing quarter, covers group joiners and opens 3121(k)(4)(C)'s window with a covered 312(c) request case (H-2 to H-4).
  • 623c7d5e keeps 707(c) guaranteed payments and former 411(a)(5) income in the base definition (H-5).
  • a01f60c2 holds 32 federal cases to service for the United States (H-6), and the sweep brings that to 36.
  • 6787b329 carries the 2601(d)(3) delay past 1986, and moves the three off-axis pins' dates into cases (H-7, M-1).
  • 4435b801 makes the oracle exact and every fixture name its deciding case (M-2).
  • 6392e710 adds two guards (unread trigger facts; parameters that state conditions) and a coverage test. It sweeps all 119 rows in both directions with 1,331 INVENTED workers and fixes the 40 defects that sweep found, with M-3 to M-6, lows 1, 2, 4, 5, 7–9 and note 3.
  • 486453fc takes lows 3 and 10.

• Where the fix follows the sources over the review:

  • H-4 stays an open case, because no capture holds (k)(4)(C)'s terms.
  • H-7's staffer served in Federal employ on July 18, 1984, so 2601(d)(1) governs that service, not (d)(2), and the February 1987 service stays covered. The fix still carries the clause, which excludes a (d)(2) staffer through the delay.
  • M-4 sits in the nonresident-alien family with no start date, because no capture gives one.
  • Stands: the rule on aliens working abroad before 1984 comes from 410(a)'s preamble and reaches every employer, so it falls outside the federal family; no capture dates an agreement before 1982, so the earlier clergy rows carry no agreement exclusion (low 6); and the 1973 gross-concept row cites 2025 guidance.

• Corrections to earlier messages:

  • 200f24d4's message says the 1984 axis_exception "routes" fiscal-year church service. No code reads it; since 6787b329 the earlier rows' cases carry the date.
  • a01f60c2's 58 government_level cases, 93 tied rows and 32 federal limited cases, and 6787b329's 59 and 94, held when written. The final figures read 67, 100 and 36.

• Mutants: at 486453fc, from the host, the tests kill all 37 mutants in the lane's set, among them the review's five survivors, its District and uniformed mutants, a mutant restoring substring matching and one restoring all 64 changed rows.

• Round 6 (delta, at 486453fc): REVISE.

  • Highs (3): H-1, a section 306 certificate excluded a minister's United States ministry; H-2, the "employing" membership text made a statewide-system member a nonmember (both from round 5's sweep); H-3, the Communist-organization row applied the post-1977 year test from 1968.
  • Mediums (14): the oracle could not see a broadened case (nine of ten mutants survived); open cases refused workers on whom their readings agree; the units person test; the order control-abroad case; 410(a)'s base definition; the unreachable casual 1977 case; the worded-condition guard and the chaplain limit; silent nonprofit choices; the undated agreement exclusion; two federal readings; closed and over-wide readings; the election agreement amount; "another individual"; the Invariants.
  • Also: thirteen lows and three notes.

• Fix (seventeen commits, 740138a7 to f926fe30):

  • 740138a7 holds the section 306 certificate to its Canadian service (H-1).
  • ba1b4a13 counts a statewide system that covers the position as membership (H-2).
  • 16fb9b75 tests the quarter before 1978 (H-3).
  • 15c09d93 narrows or settles the open cases (M-2).
  • 56c40b2f gives the units church cases the service test, drops the control-abroad case and records 410(a)'s base definition (M-3 to M-5, low 3).
  • 0b3deb74 keys the 1977 casual cash test on the payment date and checks case thresholds in the loader (M-6).
  • 4a545434 adds just-miss workers and their guards, pins deferrals, reasons and ties, and deletes the vacuous level properties (M-1).
  • 80cc8e2c carries the government-chaplain limit and widens the worded-condition guard (M-7).
  • 60200c1e opens or limits the silent nonprofit, agreement and farm-arrangement choices (M-8 to M-10, M-12, M-13).
  • edef65df and a174270c remove scratch-note citations and tidy the module (low 2, code quality).
  • 4a4f0579, fbfc0672, c3c0b4e2, 852da875 and 132efcef open the sect, gross-concept, 401(a)-trust, pre-1983 agreement and minor direct-seller readings, and carry (G)'s definition of "subject to" (M-8 to M-11).
  • f926fe30 corrects the State and local residuals' wording (low 5).

• Where the fix follows the sources over the review: drop_order_control_election has no case to drop after M-4, so its successor drops the abroad case's place condition.

• Left for round 7: the over-wide refusals of the open rows (stated limits) and lows 1, 4 and 6 to 13.

• Corrections to earlier messages:

  • 6392e710 says a case "reworded, narrowed or broadened" fails a worker; broadening fails one only where a just-miss worker exists (M-1).
  • 6787b329 and the round-5 fix report say the review's $80 units worker reaches church_combined_below_100; it reaches the residual, whose $400 floor gives the same outcome (low 3).
  • The round-5 fix report's "Choices for review" 6 (the units person test) and 7 ("employing") did not hold (M-3, H-2), and round 6 closed its H-4 open case (M-2).
  • Round 6's messages carry interim counts that held at their own commits. Round 6's final figures read 39 open cases, 94 readings, 360 cases, 106 tied rows, 369 unread facts and 1,432 workers.
  • edef65df left one test failing; a174270c fixes it.

• Mutants: at f926fe30, from the host, the tests kill all 29 mutants in the lane's set, among them the review's nine survivors and one restoring every row this job changed, and 36 of round 5's 37. The other, r5_units_church_cases_dropped_1968, no longer fails the one test the set names for it, because low 3's fix took that outcome off the row; it fails 6 sweep tests.

• Round 7 (delta, at f926fe30): REVISE. Worker numbers in this entry hold at f926fe30; at a9aae687 each falls by one, because 43deeb60 removed worker 248.

  • Highs (2):
    • H-1: round 6's membership text made a member of their employer's qualified system a nonmember for a second, excluded position. Round 6 regressed here, and two of the review's repros came out covered.
    • H-2: scope_gaps and the four federal residuals misstated 410(a)'s preamble against their own capture.
  • Mediums (10):
    • the section 306 certificate read contributions per service;
    • the registry closed the agreement exemption from 1983-04-20 with no capture;
    • the pre-1983 agreement open cases refused workers whom both readings exclude;
    • the registry left open a 1973–1987 church TSA question that the captures settle;
    • electing-church pay from an employer paying under $100 fell to the units residual;
    • the casual rows closed §351(d) with false support;
    • the minor direct-seller case had no newspaper limit;
    • four open cases refused workers whom both readings decide alike, and 521 cooperatives followed 3121(a)(16) silently;
    • the (k)(4)(C) window missed an employer that resumed paying;
    • round 6's test docstring, 4a545434 and an Invariant overstated what the tests check.
  • Also: eleven lows and five notes.

• Fix (25 commits, e83ac7cd to 22edbd38):

  • e83ac7cd and 1986c1b6 capture 26 C.F.R. 31.3121(b)(7)-2 and decide membership entity by entity (H-1).
  • fb0fd700 names 410(a)'s (A)(ii), (B)(ii) condition and (C) in the scope gap and residuals (H-2, low 9).
  • 764b730a reads section 306(a)(3) as who may file (M-1).
  • f07e2bee captures 1401 and 433 and closes the pre-1983 agreement cases (M-2, M-3).
  • 1f5d0ef2 closes the church TSA question (M-4).
  • 239ef16f keeps under-$100 church pay out of the units rows (M-5, low 7).
  • 95cfe0a2 records the §351(d) payment-date reading as CE1's choice (M-6).
  • f57f380a limits the minor direct-seller cases to 410(a)(14) work (M-7).
  • 4774f453 limits four open cases and opens 521 (M-8).
  • e5129c32 rereads the (k)(4)(C) window (M-9).
  • 95fa4faa adds government-level just-miss workers, narrows the docstring and widens the parameter guard (M-10).
  • 43deeb60 dates the army-corps case from 10 U.S.C. 2360 (lows 5, 6).
  • 0fe75bde limits the FEPCA case to 410(a)(5)(A) service (low 11).
  • dc674bc0, c55ebd95, 25b02a6a and 8fc6cf34 fix labels, quotes, the agreement amount's period and stale text, and add the run-together-sentence guard (lows 2, 3, 4, 8).
  • 3b884119, 0b93343f and 6d7cf040 fix the threshold-year check, the fixture-mutating test and the import-time parses (low 10).
  • c9b32a42 fixes the rest of low 5's fixture defects.
  • 423b050a and 22edbd38 add just-miss workers for the two units $400 church cases and the gross concept's pension case. a36f0734 gives eight just-miss workers a text of their own for the fact they miss.

• Where the fix follows the sources over the review or brief:

  • The captured regulation settles worker 1348 as covered (Example 2; (e)(2)(v)); the review found no capture that settles it.
  • The 1401 capture closes M-3's open cases instead of conditioning them.
  • se_nonresident_alien_1968_1983 gains the overrides its description claimed.
  • The conformance module parsed the registry four times at import, where the review counted two.
  • Worker 1388 and the 7873 case text both carried the defect.

• Open questions the rows name:

  • 10 U.S.C. 2360: the registry reads the army-corps case as reaching only a student whom the federal government otherwise employs. One reading of POMS RS 01901.340 C.5 excludes all 2360 aide service, and the federal family can't express that reading. No published example depends on the difference.
  • FEPCA waivers: no capture dates the waiver law, so the open case still reaches later service.
  • Section 351(d): CE1 chooses the payment date for the casual cash test and names the conjunctive reading, the one nonprofit_1968_1977 applies to 409(p).
  • Pub. L. 99-514 section 1882(a): no capture dates it, so the sect case stays open.
  • (G)'s window: its end, 1988-11-09, and worker 1421's new date rest on the review; no capture checks them.
  • Requalified staffers (workers 333, 334 and 336): the after-official override treats them consistently, but no capture settles their treatment.

• Corrections to earlier messages:

  • 4a545434 claims just-miss workers for "the level … conditions of this round's changed cases". fers_or_fsps_election, after_official_service_from_1988_11_10 and the army-corps cases had none until 95fa4faa (M-10(b)).
  • The PR's Invariant "ties every parameter that states a condition in words" overstated a guard that reads a fixed word list (M-10(c)).
  • 56c40b2f, round 6's low-3 docstring and round 6's PR-body line say the review's $80 units worker reaches the residual and its $400 floor. Former 411(a)(13)(C) keeps that pay out, and church_employer_below_100 now decides it (M-5).
  • f926fe30's message and residual text call 410(a)(7)(C)–(E) the District clauses. Only (D) names the District of Columbia (low 2).
  • The PR's scope line said service abroad counts as employment "only under 410(a)(B)". (A)(ii) and (C) also reach it, and (B)(ii) needs a 3121(l) agreement in effect (H-2).
  • Round 6's test docstring claimed a just-miss worker for every condition of a changed case; 95fa4faa narrows it (M-10(a)).
  • e83ac7cd says "seven workers' prose now cites the rule". Six cite it; the seventh, worker 1387, only rewords.
  • e83ac7cd says the two CFR editions "differ only in (d)(2)(i)'s cash-out amount and apostrophes". Their edition headers and source notes differ too: the 2025 edition adds T.D. 8794 (63 FR 70338, Dec. 21, 1998) and T.D. 8891 (65 FR 44682, July 19, 2000).
  • 43deeb60 says "Captures 246 -> 247 files". The folder went from 248 to 249.
  • 3b884119's message leaves the four import-time parses as stated limits; 6d7cf040 cuts them to one.
  • An automatic commit step created 95fa4faa with the subject "fixup! CE1: date the army-corps case …", no body and no trailer. It holds M-10's government-level just-miss workers and test edits.
  • e83ac7cd through 95cfe0a2 left test_every_covered_required_condition_names_its_failure failing; 1986c1b6 fixes it.
  • The other interim figures in round 7's messages match their own commits.

• Left open: round 6's lows 1, 4, 6, 7 and 9–13.

• Merge turn (4b484f91, a9aae687): 4b484f91 merges master 97003ac2, and with it Federal SSI interface and source-to-input contract (SSI0) #573 (SSI0) and Aggregate-table captures and readers, part 1 (D4) #574 (D4). Each added a block to the birth-evidence exclusions, their expected tuple and the reachability test, the three places where CE1 adds its own. The merge keeps master's blocks and puts CE1's after them in all three places. Both sides add src/populace_dynamics/income_resources/__init__.py as the same blob, and D4's block already lists it, so CE1's exclusion blocks keep only coverage_rules.py. a9aae687 recounts the tiers on master's files: CE1 adds 83 unit and 2,199 artifact tests. From 22edbd38 to a9aae687, 311 paths changed: the two tier files, the two birth-evidence files that combine both sides, and 307 paths that now match master's blobs. None of CE1's other paths moved.

• Mutants: at a9aae687, from the host, the tests or the loader kill 39 of r7_mutants.py's 51, among them the review's D1 and D6b and one that restores every row round 7 changed. Nine survive as stated limits: D3, D4, D5, D7, D8, D10 and the three prose-only row mutants. Two more survive their named tests. Undoing 95cfe0a2's row changes fails test_closed_rows_name_their_operative_reading when the whole conformance module runs, though the mutant's named tests leave that test out. Dropping Mae, the (k)(4)(C) just-miss worker that e5129c32 added, passes the whole conformance module, because 25 other INVENTED workers just miss the (k)(4)(C) case on nonprofit_coverage_waiver. r7_unseen_limit_lowered expects a limit of 171, and 423b050a lowered it to 169, so the mutant stops at its precondition; at 169, the same mutant fails test_unseen_conditions_are_a_stated_limit. Of round 6's 29, the tests or the loader kill 27. r6_membership_employing_restored and r6_unseen_limit_raised stop at their preconditions, because round 7 rewrote the membership text and lowered the limit.

• Round 8 (delta, at a9aae687): approved, with no high or medium finding and no regression against the parent; four low findings, eight stated limits and seven notes. The reviewer read all 31 changed rows against their captures and found all 115 new quotes in them. The reviewer also decided 42 sampled workers blind: the registry and fixture agree on all 42, and each of the reviewer's seven substantive differences settles for the registry.

  • L-1: nonprofit_1968_1977's 521 case still credits the 1978 row with a reading that 4774f453 opened.
  • L-2: the Invariant on overrides stated the loader's rule backwards.
  • L-3: the Invariant on re-ranking claimed more than the tests check. Moving se_units_1984_1985.church_employer_below_100 to last passes every test, because no worker can tell the ranks apart.
  • L-4: the scope gap and the four federal residuals describe section 321(b)'s amendment without section 321(f)'s election.
  • Mutants: of the reviewer's 14, 3 fail a test and 11 survive. The eight stated limits cover nine survivors, L-3 covers one, and the last raises UNSEEN_CONDITION_LIMIT, an upper bound by design.
  • Notes: the round-7 entry numbers workers as f926fe30 did. At a9aae687, workers 820, 839 and 1123 carry a bare "(NOTE)", worker 391 quotes a text the residual no longer prints, and worker 523's prose ends "The registry refuses. M-6." though it decides section_105b_departed_employee_request. A test comment keeps round 6's 24 of 374 as history. Section 351(d) stays a disclosed choice.

• Body update (no commit; the head stays a9aae687): this body corrects L-2 and L-3, adds the round-7 entry's numbering clause, and states L-1, L-4 and the eight gaps under Stated limits.

Plan

1. The coordinator merges this PR in the tier-count slot, by squash. If another PR lands first, this branch merges master and recounts the tiers.
2. A text-only follow-up corrects the two registry texts that Stated limits names.

Tests

• The two CE1 modules: 2,282 tests, 83 unit and 2,199 artifact: the same unit count and 2,096 artifact more than at ea599b92.
• With the birth-evidence seal tests: 2,294 passed at 4b484f91, from the host, and a9aae687 changes only the two tier files. The one failure, the reachability test's Track U assertion, needs modules this sparse clone leaves out; the CE1 block before it passes, and CI runs the whole test.
• Mutants: each guard that round 3 added fails its tests when a mutant removes its clause, and each new data test fails on ea599b92's registry. 37 runs, 37 killed. At 210a5c85, from the host, the tests kill the follow-up's 15 mutants and 12 of round 4's 13, among them the review's three gap probes and one mutant per new guard and data test. The thirteenth, r4_old_rows, fails 30 tests across six of its seven expected tests; the seventh passes because none of its covered entries changes outcome on those rows. Round 5: at 486453fc, from the host, the tests kill all 37 mutants in the lane's set, among them the review's five survivors, its District and uniformed mutants, a mutant restoring substring matching and one restoring all 64 changed rows. Round 6: at f926fe30, from the host, 65 of the 66 mutants in the two sets fail their named tests, and the sixty-sixth fails the sweep tests (see the Review section). Round 7: at a9aae687, from the host, the tests or the loader kill 39 of the 51 round-7 mutants and 27 of round 6's 29. Nine survive as stated limits, and the Review section accounts for the other five.
• Rebuild: the registry rebuilds byte for byte from its fragments, and each cover-less IRS extract has the same bytes as qpdf's selection of the same pages from its original.
• tests/track_u2/: CI runs it; this sparse clone leaves out files its protected-bytes test hashes.
• Lint: black -l 79 and ruff check pass CE1's four Python files and the two merged birth-evidence files.
• Tier policy: a9aae687 recounts the tiers on master 97003ac2's files: unit 23,736, artifact 6,726 and 32,790 in all. CI's tier policy test checks that count on the full collection.
• CI: at a9aae687, CI passes, the tier policy test included: run 41640791929's four pytest shards report 31,689 passed, 1,116 skipped and 1 xfailed.

axiom: n/a: a cited coverage-rule registry for the microsim's covered-earnings classifier; no PolicyEngine country-model or RuleSpec encoding changed

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
social-security-model	Ready	Preview	Nov 3, 2026 1:03pm UTC

[MaxGhenis]

The Microcosm Dynamics coordinator merged this PR at a9aae687a15bfd4df82304fa63fec161400b901f with --squash --match-head-commit, as 02897edecda1ab3c377cd5aaaa994fe4efb75444 (10:25Z, 2026-11-06). The squash commit's tree matches the head's.

• Review. The reviewer in Subfleet job 20261103-083456-ce1-review-r8 approved a9aae687, leaving no finding of medium or higher severity open. Of its four low findings, the income-and-resources lane corrected one and narrowed another in this PR's description. The other two concern registry texts that the description names under "Two registry texts", and the lane's text-only follow-up PR fixes them.
• After the review. The lane pushed no commit and changed only this PR's description.
• The squash. Earlier commits on this branch hold original source PDFs that the head no longer carries, so the coordinator squashed the PR and kept them out of master's history.
• CI. gh pr checks 583 exited 0 at the head, where 11 checks passed and deploy skipped (Tests run 41640791929). In that run, tests/test_tier_policy.py compared the committed counts with the collected suite and passed. GitHub reported the PR mergeable.
• Tier counts. Before the next PR that changes the counts lands, its owner merges master 02897ede, or later, into it and recounts the test tiers.