Fold the 2026-10-17 DoD first prints: review fy2026 strips, move the anchor

[MaxGhenis]

Why

#361 is the resolve_pending.py append lane. At dd929cba it recorded FY2026 first prints for five USAspending DoD series and regenerated the catalog. The append lane merges on its own gate, and this wave left two catalog tests red on codex/thesis-ledger-facts. Arch checks therefore fails for every PR into this base, including #360, the generator fix Thesis needs before the DoD FY2026 window closes on 2026-10-22.

What

tests/fixtures/series_catalog/reviewed_stripped_segments.json

• test_committed_catalog_is_current_and_valid reported fy2026 as the first strip in each series' source_record_id shape, usaspending.dod.<series>.{P:fy9999}.registered_query_snapshot.
• Each observation declares period fiscal_year 2026 and spells it directly; the builder's kind is derived. The docket entries declare period: FY2026, and the USAspending query is keyed by fiscal_year.
• fy2026 is therefore a period label, not a statute, cohort or edition label. The five strips are recorded as reviewed, under the concept, field and shape the audit names, with a dated fy2026 note.

tests/test_build_series_catalog.py, test_identity_uuid_map_matches_reviewed_anchor

• dd929cba's ten registry events are five retire+succeeds enrichments of geography-free placeholders. Each keeps its UUID and takes the observation's US current geography and economy aggregate entity:
  • new_prime_awards (8146ecb0);
  • prime_award_transactions (82ff2d7f);
  • prime_contract_obligations (11ea0d3d);
  • small_business_contract_obligation_share (91e3b668);
  • unique_prime_contract_recipients (96d17026).
• I recomputed the digest with the test's algorithm. 8c4a0f86 reproduces the old pin (56bb47bc…). The delta to dd929cba is exactly those five tuples, re-keyed with the same UUIDs; the other 216 are unchanged and there are 221 rows.
• The new digest (b06bf665…) is pinned with a dated comment.

No ledger bytes change.

Verification

• uv run --locked pytest tests/test_build_series_catalog.py -q: 222 passed. On the base tip it fails these 2 tests.
• uv run --locked python scripts/build_series_catalog.py --check: catalog current: 221 series.
• uv run --locked ruff check tests/test_build_series_catalog.py: clean.
• The reviewer can regenerate everything from dd929cba's committed catalog. The strip kinds come from build_catalog(...)[1]["strips"] on the committed inputs; the digest delta comes from the test's own algorithm applied at 8c4a0f86 and dd929cba.

axiom: n/a: catalog test fixture and anchor, no policy encoding

🤖 Generated with Claude Code

[MaxGhenis]

Independent review: MERGE (Opus 5.5 on a Subfleet review --tier hard run. The Codex/GPT-6.1 Sol lanes were on operator hold, so the review fell back to Opus per lane policy.)

Reviewed head: 9a2bb393. Report: ~/chief-of-staff/state/chronicle-lineage/dod-second-enrichment/review-363-sol.md.

The reviewer independently confirmed the following:

• Strips. All five fy2026 strips are period labels. They match the builder's derived kinds and the audit's concept/field/shape keys.
• Anchor. The digest reproduces the old pin at 8c4a0f86 and the new one at dd929cba. The delta is exactly the five DoD tuples, re-keyed with the same UUIDs. The ten registry events equal what the builder plans when replayed from the base inputs.
• Nothing else moves. No ledger bytes change, and --check is current.
• Tests. The file passes at the head and fails only the two named tests at dd929cba.
• Scope of the reviews. They admit only later prints of these five series in the same shape with a derived kind. Synthetic sibling, shape, field and mismatched-period scenarios are still caught.

Disposition of the two NITs:

• NIT-1: the comment says "No mint" while the builder reports minted=5, all of which are succeeds events. The facts in the comment are correct, so it is left as is.
• NIT-2: the file isn't ruff format-clean. That predates this PR, and CI runs ruff check only.

Since the review, the only change is 152ea213, a clean merge of 86da0233. That commit is the append lane recording one continued-claims first print, with no registry events. At 152ea213: 222 passed and --check current.