gate_m6 candidate-3 verdict: PASS (registration 12, one-shot artifact)

[MaxGhenis]

gate_m6 candidate-3 verdict: PASS — the campaign's first

Publishes the one-shot write_new output of registration 12
(#42 comment 5064153427),
invocation #3, under publishes_regardless. Two frozen files, byte-exact
as written by the runner at 06:49 ET 2026-07-24:

• runs/gate_m6_candidate3_v1.json — sha256
  caf254925f44b27c1bd1131336055e27fbc311daec00b0050b2f9293a74e82cf
• runs/gate_m6_candidate3_v1.json.env.json — sha256
  6a2559d053ef9da9abbcfc710f9d45b91b749c918b4e48a0fbee3c54003cb38f

Verdict (from the artifact, not the stdout line)

verdict.status = PASS, valid = true. Acceptance conjunction:
gate_contract_result.pass = true (4/5 seeds; seed_pass
{0:T, 1:T, 2:F, 3:T, 4:T}) AND must_not_regress_result.pass = true.
Registration binding echoed in-artifact: registration_id "5064153427",
one_registered_run, no_self_rescue, publishes_regardless,
fresh_registration_required: true.

Scored target cells (score vs tolerance, per seed 0-4, % of tolerance):

cell	tol	s0	s1	s2	s3	s4	seeds
earn_autocorr_lag2 (F2, routed target)	0.087	75.3%	52.6%	71.5%	93.6%	38.6%	PASS 5/5
earn_dlog_mean.prime (F1)	0.043	85.8%	77.9%	120.6%	70.6%	63.4%	PASS 4/5
remarriage.18-64 (law unchanged)	0.403	59.0%	73.2%	102.7%	73.8%	72.3%	PASS 4/5

• F2, the amendment-6 routed target, cleared in all five seeds —
  candidate 2 breached it at 109% (seed 3). The train-side evidence
  (−0.109 gap ≈ 3.9 floor-SDs) transferred.
• F1 improved in every seed vs candidate 2's worst (c2 seed-2:
  129.9% → c3 seed-2: 120.6%; all other seeds 63-86%); the registered
  forecast made no F1 claim (unresolved-by-design, M6 c3 §5.1 diagnostic: F1 mechanism split (train-only, boundary 2010) #271 conditioning
  caveat) and the ≥4/5 contract rule absorbs the residual seed-2 breach.
• Remarriage reproduced the candidate-2 scored band (c2: 59-103%,
  median 73.2%) — exactly what registration 12 forecast with the law
  unchanged, and further confirmation of the §2.4 transport-calibration
  datum (the old boundary-transport forecast would have predicted doom).
• Seed 2 fails the same two cells it failed under candidate 2; the other
  four seeds are 11/11.
• Verdict scope, from the artifact: earnings certification =
  "M6-first-certified forward earnings law; no gate_1 backward-law
  certificate transfers"; certifies_nothing_about_mortality_drift;
  family B/C remain gated off; report-only surfaces unchanged.

Bindings (sidecar)

contract.blob_sha b0c39af1e13a705f90b85d3e6b9a91e1d3c5485c (the live
CONTRACT_BLOB pin from the #280 flip), head_sha f10cca5457b16d12b9284d00628e7331871f23e7
(the ratified #281 squash; this branch is cut from that exact commit),
pinned populace fitting stack ee8f7fc1… with OMP_WAIT_POLICY=ACTIVE
— the #273 lock-addendum runtime, enforced fail-closed by the runner.

Execution record

Registration 12's invocation chain, all disclosed on #42: invocation #1
(guard refusal, thread environment unset — refusal-to-start, comment
5064171894); invocation #2 (18:29 ET, killed by a host power-management
clamp after ~2h at full speed — incident 5065343857; the disclosed
re-execution allowance was consumed, zero leakage evidenced in the
report); invocation #3 (21:49:20 ET via the hardened launchd-domain
topology, verification comment 5065367143) ran uninterrupted and wrote
the artifact at 06:49 ET 2026-07-24 — ~9h00m wall from the disclosed
launch time. (An earlier revision of this body quoted an average-core
figure from live coordinator observations; it is removed because no
retained evidence reproduces it.)

Mechanical note: this commit was made with --no-verify because the
repo's beads pre-commit hook (a .beads/issues.jsonl flush, unrelated
to the ceremony) errors in this worktree; the commit message and file
bytes are verified in the PR as shown above.

What ratification means

Merging this PR ratifies the campaign's first PASS: the correlated-
refresh forward earnings law (q*=0.55, ρ*=−0.60) is first-certified on
the registered 2016/2018 gate_m6 surface. No gate_1 certificate
transfers; nothing is certified about mortality drift, the 2020-2022
shock window, or any report-only path. Per the standing ceremony:
adversarial referee round → ratify-by-merge. Forecast ledger entry 9
(certified engine p50 2026-07-26) is satisfied two days early at
ratification.

🤖 Generated with Claude Code

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
social-security-model	Ready	Preview, Comment	Jul 24, 2026 10:54am

[MaxGhenis]

Adversarial referee round (sol, delegated) — VERDICT: FIX-FIRST → fixes applied → converts to SHIP by its own terms.

The reviewer completed all ten checklist items. Its conclusion, verbatim from the session record (the review file could not be written from the read-only audit sandbox; the delegate output stream carried the verdict):

All ten checks are complete. The frozen artifact's PASS recomputes exactly; the only blocking completeness issue is an unreproducible PR-body runtime statistic, plus a cosmetic table-header defect.

The artifact itself survives the adversarial recomputation: its published PASS is numerically exact. I'm concluding FIX-FIRST, narrowly, because the PR body asserts an average-core statistic that has no retained evidence and therefore cannot satisfy the requested 'every number recomputes' standard; the fix is body-only and does not alter the frozen files.

Findings and disposition (both body-only; the two frozen files are untouched, shas unchanged):

1. Unreproducible runtime statistic (blocking) — the body's '~6.7 cores average' came from live coordinator observations with no retained evidence. FIXED: removed, with an explicit note; the execution record now carries only evidence-backed values (launch time from the Candidate 2 design: latent-permanent conditioned chained QRF (research memo) #42 disclosure, artifact write time, wall duration derived from those).
2. Table-header defect (cosmetic) — the per-seed table's verdict column had no header. FIXED: 'seeds' header added.

Referee session record: ~/m6-sol-lanes/review-283-out(.err.log) (coordinator-side). Ratification proceeds under the standing merge gate (fresh CI checks + MERGEABLE).