Registration 5: the anchor context report (forecast ledger entry 10)

[MaxGhenis]

Registration 5 of the entry-10 anchor-context report ceremony, under the ratified design (docs/design/anchor_context_extraction.md, 1ad337d) and the §5.3 execution law. Registrations 1–4 were consumed (#317, #319, #322, #324 — incidents 1–5 all committed, latest records merge 809eeeb).

What changed since registration 4

Incident 4's diagnosis (#324) found an implementation defect: the input loader asserted inputs are compact-canonical JSON, which the pretty-printed published first-estimates artifact can never satisfy. The fix — strict parse fidelity (duplicate-key, non-finite-literal, and float-overflow rejection) in place of byte-form identity, with the registered-sha256 gate unchanged — merged as PR #327 after an adversarial referee round, with regression fixtures byte-shaped like production.

Fresh-registration adjudication (stale durable claims)

Run 4 issued its initial attempt before aborting, leaving the two durable claim files at their canonical paths. Registration 4 is dead (incident 4 is not retry-eligible: reason not external_*), so under the coordinator law ("fresh-registration adjudication is required") the coordinator clears both before this registration's launch:

• runs/anchor_context_report_attempt.claim — 24,556 bytes, sha256 7a247fd6c4148b09ad5750372cbcffbed9132feb974409aa358db61c71028bd4. Its prelaunch evidence duplicates what incident 4's committed record and the registration-4 document already carry.
• runs/anchor_context_report_retry_authority.claim — 0 bytes (the empty reservation is the designed persistent state for a non-retryable incident; no retry authority was ever sealed), sha256 e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855.

Both are untracked working files; the committed incident record is the published trace. This mirrors the entry-8 first-estimates precedent.

The registered configuration

• registration_reference: populace-dynamics#328
• implementation_commit: 2500adfdbad8038c18e34fadc6ccbd15d3056e90 (the loader-fix merge; the identity law admits records-only descendants)
• Design binding: revision 4 at 1ad337d3a3eaeba3369a3405469b1e74335e156a (unchanged)
• Inputs (unchanged, immutable): runs/first_estimates_v1.json sha256 719604ca4364e7cdef2293329ed0beb0e011e5d4d1c34f0e508c8f2fd9932977; data/external/ssa_level_anchors_supplement2025_trustees2026_vintage1.json sha256 adc782a1a11c50969103c125a82b1539a7017241662d545d86bc6fc9227730c1 (vintage ssa_level_anchors.supplement2025_trustees2026.vintage1)
• Registered isolated invocation (8 elements, byte-exact):

/Users/maxghenis/PolicyEngine/social-security-model-worktrees/sol-c3-runner/.venv/bin/python
-I
-B
-X
pycache_prefix=/tmp/e10-run5-pyc
scripts/run_anchor_context_report.py
--registration
docs/registrations/anchor_context_registration_5_configuration.json

• Document: docs/registrations/anchor_context_registration_5_configuration.json, canonical bytes sha256 32e9821724e3f610415a62d532f84cb9253934b0f3390f60349ac1b677a759f7 (21,136 bytes), committed via the records PR referenced below.

Execution acknowledgement

One registered run; publishes_regardless; no_self_rescue; retry only for a §5.2 retry-eligible (external_*, no-yield) incident, after its incident publishes. A report pair or a second failure of any kind ends this registration. Entry 10 (p50 2026-07-31, p80 2026-08-05) resolves only at the merge of a publication PR carrying the complete validator-passing report pair.

Pre-launch: the seven-point checklist (tree pristine at the records merge; attempt claim absent; retry marker absent; report pair absent; registered document present and tracked; no bytecode under src/ or scripts/; launcher invocation equal to the registered array element-wise) runs after adjudication, and the sentinel /tmp/e10-run5-pyc is recreated fresh-empty immediately before launch.

[MaxGhenis]

Launch, run, and publication

Fresh-registration adjudication completed as disclosed: both stale claims cleared (attempt claim sha256 7a247fd6…, 24,556 bytes; empty retry-authority reservation) after the ceremony worktree moved to the records merge 5c9ac263.

Seven-point checklist: 7/7 PASS (tree pristine at 5c9ac263; attempt claim absent; retry marker absent; report pair absent; registered document present and tracked; no bytecode under src/ or scripts/; launcher invocation equal to the registered 8-element array element-wise). Sentinel /tmp/e10-run5-pyc recreated fresh-empty immediately before launch.

Sealed launch via one-shot LaunchAgent (fresh label, then retired): pid 20903, ppid 1, nice 0, sys.orig_argv = the registered array byte-for-byte.

Outcome: status: "published". The complete report pair exists; no incident 6.

• runs/anchor_context_report_v1.json — 47,935 bytes, sha256 1ff3f0a04d9a511079e7b1773e01de48a0f4a9b1ec405d4601d410eb88c2cb34
• runs/anchor_context_report_v1.json.env.json — 490 bytes, sha256 9e594df76cbc34a6eb8dc58842d37b528f0ede1e0c104469f558a57cb9fd3b67

Exact-complete results: all nine frozen comparison specs are present — seven evaluated, two unavailable with their registered reasons (oact_annual_award_average_not_registered_in_v1, retired_worker_december_total_monthly_benefit_not_registered_in_vintage1). Frame-relative labels, certifies_nothing, mismatch codes, and the frozen birth-timing status (baseline-only, unresolved) are carried in the artifact.

Weight-scale-invariant comparison means (model intensity / official intensity), 2015 → 2022:

comparison	2015	2022
reported taxable earnings per worker	1.222	1.241
adjusted taxable payroll per covered worker	1.226	1.248
gross contributions per worker	1.222	1.241
net payroll-tax contributions per covered worker	1.234	1.278
retired-worker beneficiaries per worker	0.758	0.871
retired-worker awards per worker	0.558	0.637
retired-worker benefits per reported taxable earnings	0.597	0.683

These are context ratios under the registered mismatch codes — not calibration, not certification of levels.

The publication PR with the exact pair follows; under §5.3, forecast-ledger entry 10 resolves at its merge.

[MaxGhenis]

Completion record

The publication PR #330 merged at 97ddeacf7229698c0bff3dfa7534b6f4ef4bb7bb (2026-07-28T18:16:37Z) with the complete primary report pair (anchor_context_report_v1.json sha256 1ff3f0a0…, sidecar 9e594df7…). Under design §5.3, forecast-ledger entry 10 is resolved by this merge: p50 was 2026-07-31, so error_days_p50 = −3; within p50 and p80. The ledger record follows in a records PR.

Ceremony summary: five registrations (#317, #319, #322, #324, this one), five committed incidents — two launch-layer laws (invocation byte-equality, code-tree identity), one placeholder drift, one real implementation defect (the input canonical-form assertion, fixed in #327 through three adversarial referee rounds), one diagnostic-discipline breach — every one disclosed before the next registration, no input opened outside the sealed runs, and the fifth registration published on its one shot.