[MaxGhenis]

Draft: waiting on an independent Sol review and a blind adjudication of the readings.

Part of re-framing Track B's "Needs Max's judgment" items under Max's 2026-10-20 delegation (single readings are the session's call under P1–P4, logged in the source-conflict rulings; patterns go to Max). This PR covers the R3 note and the one R3 reading that changes money. The G and R2 items follow in a separate PR after #563 lands, so as not to collide with it.

What changes

Reading taken: a reduced spouse terminated and reentitled before FRA (us:oasdi-arf-spouse-reentitlement-credits, P1 statute first). RS 00615.201A.1 keeps the original reduction months until an ARF at FRA; RS 00615.491A removes the terminated months at reentitlement. The Act decides it:

• 402(q)(1)(B)(i) reduces each month before retirement age by "the number of months in the reduction period for such benefit (determined under paragraph (6))";
• (ii) substitutes the adjusted period (which (q)(7)(C) shortens by terminated months) only "if such benefit is (I) for the month in which such individual attains age 62, or (II) for the month in which such individual attains retirement age";
• the spouse's reduction period starts with the original certificate ((q)(6)(A)(ii)), which stays effective "for the month in which it is filed and for any month thereafter" ((q)(5)(B)(i)).

So apply_arf now computes pre-FRA months instead of refusing them. On the note's invented case (born 1960-01-02, $650 unreduced, first reduced 2022-01, terminated 2022-02, reentitled 2022-03), it pays $422.50 ($422) for March 2022 through December 2026 and $425.20 ($425) from the January 2027 ARF. The .491A reading would pay $425.20 from March 2022, $2.70 a month more. The reason value arf_spouse_reentitlement_before_fra is withdrawn, together with its duplicate in ARFResult's coverage check.

Refusals kept, re-framed in the note:

• Widow(er) FRA for births 1911-01-02 to 1912-01-01 (us:oasdi-survivor-fra-1911-births). This conflict can't change any amount the module computes. Every reduced entitlement of this cohort predates 1978 (before FRA, under either 62 or 65), and arf_initial_entitlement_before_1978 refuses those independently. The helper must return one month, so it still refuses. If a pre-1978 path is ever built, Pub. L. 92-603 §102(i) ("for months after December 1972") supports POMS's 1/2/1911 boundary. 20 CFR 404.409(b)'s 1/2/1912 boundary would then be a regulation contradicting its statute, which P2 sends to Max.

• Nonbinding widow(er) child-in-care ARF months (us:oasdi-arf-widow-child-in-care-credits). The two sources diverge:

  • 402(q)(7)(D) credits only months "in which the reduction in the amount of such benefit was determined under paragraph (5)(D)". (q)(7)(B) gives wives the plainer child-in-care test.
  • POMS RS 00615.482 agrees with the statute.
  • 20 CFR 404.412(a)(4) copies the wife's wording for widows.

  P2 returns a regulation that contradicts its statute to Max, and whether P2 covers SSA's 20 CFR at all is unruled. So this case and that pattern are queued together for Max, recommending the statute. Until he rules, the configuration stays refused and counted.

• Float-AIME and ties-to-even diagnostics, and the RS 00615.015 date erratum, are labelled as not being source conflicts.

Invariants (executed)

• Pre-FRA invariance: for any terminated history and any month before FRA, the payment equals the never-terminated claim's payment and G's 60-month spouse reduction, at zero tolerance. Hypothesis: test_reentitled_spouse_keeps_original_months_until_fra.
• FRA credit: at FRA the ARF credits exactly the terminated months before FRA, so the benefit never falls at FRA (same property).
• Statutory quotes: the quotes and both amounts are pinned (test_spouse_reentitlement_takes_the_statutory_timing), and the public API too (test_spouse_reentitled_before_fra_keeps_original_months_until_fra). The # Source conflict marker sits at the timing line, and test_source_conflict_markers_are_adjacent_to_runtime_decisions checks it.

Mutation checks

Each check ran in an isolated copy of src/, with its import path verified:

Mutant	Result
.491A timing (adjust at reentitlement)	3 of 3 new tests fail
Drop the terminated-month credit at FRA	3 of 3 fail
Re-add a pre-FRA refusal	3 of 3 fail
Restored	3 pass

Verification

• Full Track B suite: 11,411 passed, 3 skipped (pytest -n 4, staged-data roots set to nonexistent paths).
• Repo-wide black -l 79 --check and ruff check are clean.
• Tier counts are unchanged against master 6967142 (one enum-parametrized case removed, one property added), so tests/tier_counts.json is untouched.

The note quotes and ids are logged in ~/reviews/us-hub/source-conflicts/RULINGS.md. The public register (rulespec-us known-source-conflicts.yaml) is updated after merge, per its protocol.

axiom: n/a: microcosm-dynamics statutory reference, not a PolicyEngine country model.

🤖 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	Oct 23, 2026 5:58pm UTC

[MaxGhenis]

Merged at reviewed head 3e707e5 as 4fc2b9f.

• Reviews: GPT-6.1 Sol r1 APPROVED 9419001 with three low documentation findings, all fixed in 3e707e5. Sol r2 delta APPROVED 3e707e5.
• Blind check: a separate blind Opus adjudication of the R3 readings, run on clean master without this PR's conclusions, reached the same three outcomes.
• CI: all checks green at 3e707e5.
• Tier counts: tests/tier_counts.json unchanged, so no tier slot was needed.
• Logged: the reading is in the source-conflict rulings as us:oasdi-arf-spouse-reentitlement-credits (P1). The nonbinding widow child-in-care item waits on Max's decision on whether SSA's 20 CFR counts as 'regulations are law'.
• Follow-up: the rulespec-us register entry for this id moves from refused to computed once the register session picks it up.