[MaxGhenis]

Problem

test_committed_catalog_is_current_and_valid pinned the exact list of stripped period spellings in the committed catalog. Every Thesis resolver append that records the next weekly claims print adds a spelling (week_2026-08-29, week_2026-09-05, …), so every append failed the required "Arch checks" status. The resolver merges through the API with an admin token (it waits only for "Append gate"), so the red check was ignored:

• Record 5 first-print observation(s) via resolve_pending.py #240 merged 2026-09-03 with Arch checks failing.
• 598f394 merged 2026-09-29 with Arch checks failing.
• Hand re-pins (Re-pin the series-catalog tests to the 2026-08-23 first-print wave #244, and the list hunk in Resolve catalog supersede links against every row's effective assertion id #306) only reset the clock until the next week.

What the list protected

The builder can't tell a real period label from a statute, cohort, or edition label that happens to spell the row's own period. stripped_segments exists so a curator can catch that (build_series_catalog.py module docstring; classify_segments). The exact list forced a human look at every new spelling. It also failed when a spelling disappeared.

It had a gap: only spellings were pinned, not concepts. A new concept starting to strip an already-listed spelling (for example 2026_07) passed unseen.

The invariant now checked

build_series_catalog.stripped_segment_review_problems(stripped_segments, strips, reviewed) audits every strip. A strip is keyed by (canonical concept, identifier field, strip shape, spelling):

• the identifier field is measure.concept or source_record_id;
• the strip shape (strip_shape) is that field's family pattern with each {P} written {P:<template>}, for example abs.cpi.all_groups.yoy.{P:9999_99}.first_print. period_spelling_template turns digits into 9 and month words into <month>, and keeps separators and literal words. So a shape fixes every kept segment and where each stripped segment sits. Two identifiers share a shape exactly when they have the same kept segments, stripped slots in the same positions, and the same template in each slot. They can differ only in what the template normalizes: digits, including those inside an after_<qualifier>_ slot, and month words, with full and abbreviated names counting as one spelling.

The reviewed map is tests/fixtures/series_catalog/reviewed_stripped_segments.json, keyed concept → field → shape → spelling → classify_segments kinds. It holds the catalog at 55bbf3d that #244 reviewed, plus the 2026_08 ABS CPI decision (see notes).

A strip that is not in the reviewed map passes only when both of these hold:

1. its concept already has a reviewed strip in the same field and the same shape, so it is the next week or month written the same way, in the same place;
2. every current observation behind it spells its own declared period directly (kind derived, never overlap).

Everything else is still reported:

Event	Old exact list	Now
next weekly/monthly print, same concept, field and shape	fail (every append)	pass
new concept strips for the first time (new series, or a label that spells its row's period)	fail	fail
existing concept strips a new template (e.g. week_ending_… where it used week_…)	fail	fail
reviewed template in a new field or position, or beside a new kept segment (e.g. abs.cpi.2026_09.all_groups.yoy, ….revised)	fail only if the spelling was new	fail
overlap strip (identifier finer than the declared period, the June-13 defect)	fail only if the spelling was new	fail
reviewed strip starts being reached as an overlap	pass	fail
new concept strips an already-listed spelling	pass	fail
reviewed strip disappears from its field and shape	fail (spelling only)	fail
non-period spelling, spelling that fills no slot of its shape, catalog–plan disagreement	n/a	fail

build_catalog exposes the strips as plan["strips"]. The catalog bytes do not change: --check is green, the byte-regeneration assertion passes, and no catalog keys are added, so Thesis's frozen CATALOG_TOP_LEVEL_KEYS/CATALOG_ROW_KEYS are unaffected. Literal {P:...} segments are now reserved alongside {P} (no current ledger or docket-seed identifier uses one), so a shape can never mistake a kept segment for a stripped one.

Why field and shape, not just concept and template (review of 0852fcd)

The first version of this PR keyed decisions on (concept, spelling template). The independent review executed a copy of the 2026-08 ABS CPI first print (ledger line 231), moved to 2026-09, with its period also spelled inside measure.concept:

row["measure"]["concept"] = "abs.cpi.2026_09.all_groups.yoy"
row["source_record_id"] = "abs.cpi.all_groups.yoy.2026_09.first_print"

The builder folds it into the reviewed abs.cpi.all_groups.yoy identity with no mint, under a new family pattern abs.cpi.{P}.all_groups.yoy. The template-only audit passed it, so an unreviewed cohort or edition slot inside the concept borrowed the approval of the source_record_id period suffix. Now the source_record_id strip passes (reviewed shape), and the measure.concept strip is reported:

'2026_09' on 'abs.cpi.all_groups.yoy': first strip in measure.concept shape 'abs.cpi.{P:9999_99}.all_groups.yoy' for this concept — confirm it is a period label, not a statute, cohort, or edition label, and add it to the reviewed map

test_a_reviewed_template_in_a_new_concept_position_needs_review is that exact regression on the real ledger.

week_2026-06-13 vs week_2026_06_13

Not a naming bug. Both come from the one 2026-06-13 initial-claims row (ledger line 44):

• its source_record_id is us.dol.initial_claims.sa.week_2026-06-13 (ISO hyphens);
• its measure.concept is us.dol.initial_claims.sa.week_2026_06_13 (underscores).

The builder strips period tokens from both fields and records each distinct spelling. Both strip as overlap because the row declares month 2026-06, the cadence defect already in docs/catalog-curation-backlog.md ("Initial claims cadence metadata"). They are now two reviewed strips, one in each field's shape. The fixture notes and the backlog document this, and test_reviewed_stripped_segments_fixture_is_wellformed asserts it.

Evidence on real data

The fixture was regenerated from 55bbf3d's inputs with this builder. It covers exactly the 176 (spelling, concept) pairs the previous fixture listed, with the same kinds, now as 354 strips in 297 (concept, field, shape) decisions across 120 concepts.

Each merged append since 55bbf3d and each refused proposal head, built with this branch's builder (before the #344 merge, which changes no strip) and audited against the #244 map (the fixture without the 2026-10-12 ABS CPI decision):

Head	Unreviewed strips	Problems
#283 78308f8 (refused)	10	0
#297 7917dbe (refused)	10	0
#298 7bc1595 (refused)	10	0
598f394 (merged, 8 prints)	11	0
#289 177c461 (refused)	11	1: '2026_08' on 'abs.labour.unemployment_rate': first strip in source_record_id shape 'abs.labour.unemployment_rate.{P:9999_99}.first_print'…
2ed9bae (merged)	13	1: '2026_08' on 'abs.cpi.all_groups.yoy': first strip in source_record_id shape 'abs.cpi.all_groups.yoy.{P:9999_99}.first_print'…
d7e45bc (merged; the base before #344)	15	1: the same ABS CPI strip

Scoping by field and shape added no false alarms on any routine append. The two remaining reports are real curation events:

• Record 8 first-print observation(s) via resolve_pending.py #289: the docket placeholder abs.labour.unemployment_rate took its first print beside an already-observed abs.labour.unemployment_rate.australia, which looks like a concept-spelling duplicate.
• 2ed9bae: the ABS CPI series switched from 2026-06 to 2026_08 in its source_record_id. 0852fcd reviewed that strip (a period label) and recorded it with a note. Against the full fixture, 2ed9bae and d7e45bc have no problems.

Invariants (tested)

Property tests use Hypothesis, derandomized so this required check stays deterministic:

• T1: a spelling's template depends only on how it is written, never on the period it names.
• T2: different spelling families get different templates. Full and abbreviated month names share one template on purpose, because may is both.
• T3: every generated family spells its own period directly.
• T4: an identifier's strip shape depends only on its kept segments, where its period sits, and how it is written. It never depends on the period named, and a different position or template gives a different shape.
• A1, routine continuation passes: end to end through the real builder, appending further periods of observed series written as they already are never produces a problem. A non-vacuity assertion checks that the new strips really are unreviewed.
• A2, soundness: end to end, the audit always names, by spelling and concept:
  • a new concept;
  • a new template;
  • a new shape: the period spelled inside measure.concept, which folds into the reviewed identity, or beside a new kept segment of source_record_id;
  • an overlap strip, whether it is a fresh spelling or a reviewed one that gains a kind.
• A3, persistence: a reviewed strip missing from the plan is reported as gone.
• A4, monotone in review: reviewing more live strips never adds problems, and reviewing every live strip leaves none.
• A5: the output is sorted, has no duplicates, and does not depend on input order.
• A6, exactness: a reference model read straight off the rule names the same (spelling, concept) pairs as the audit, in both directions. Its reviewed maps include reduced kinds and stray reviews.

Other tests:

• Differential: test_strips_match_an_independent_replay re-derives every strip from the ledger with classify_segments and period_spelling_template alone: its field, its shape, and its kinds. It resolves each observation to its catalog row through that row's concept and aliases, and requires plan["strips"] and the catalog's stripped_segments to agree exactly.
• Real ledger, next period: for every eligible current observation (223: its source_record_id strips a period and all its strips are direct spellings), append the next period written the same way, month words kept full or abbreviated as written. The test asserts all 223 are appended (the first version skipped the 66 month-name prints). They land on the existing identities with no mints, supersedes or drops. They add ≥50 new strips, all in shapes their concepts already strip in, and none needs review.
• Examples: test_strip_audit_flags_a_reviewed_template_in_a_new_shape covers four cases, each folding into the reviewed identity:
  • the period inside measure.concept;
  • the very shape string of the reviewed source_record_id strips, but in measure.concept;
  • a new kept segment (.revised);
  • a new source_record_id position.
• Corruption self-test: test_committed_audit_fails_on_corrupted_catalogs derives its unprinted week from initial-claims prints and asserts only that the initial-claims pair is absent (the review's minor finding). It runs once with a valid continued-claims print of that same week, the case that broke the old global-absence assertion.
• Mutation check (run at ab6d9b9): 14 mutants of the new code. Each one is killed by at least one test:
  • keying review by concept and template only (the old rule);
  • ignoring the field;
  • shapes without templates;
  • dropping the slot check;
  • reporting "gone" by pair;
  • stripping source_record_id only;
  • reserving {P} only;
  • dropping each of the gained-kind, overlap, catalog/plan and non-period checks;
  • unsorted output;
  • swapped field labels;
  • kinds merged across shapes.

Scope and coordination

• Fold seven entity-drift lineages into their economy/aggregate siblings #344 (entity folds) merged first (65822f4); this branch merges it in f2a6d76. The Hypothesis pyproject.toml/uv.lock hunks were identical. build_catalog keeps both plan keys (entity_folds_applied, strips). The committed test keeps this audit in place of Fold seven entity-drift lineages into their economy/aggregate siblings #344's one-time re-pin of the exact list. _committed_build(), the ABS CPI regression and the next-period test pass bsc.ENTITY_FOLDS. The replay resolves a row that an active fold absorbs at the fold's survivor entity: geography and absorbed entity match, and the row's concept is one of the declared absorbed_names, read straight from the seed. The folds change no strip, so the reviewed fixture is unchanged.

• Unchanged pins: the row-count pin (== 221 since Fold seven entity-drift lineages into their economy/aggregate siblings #344) and test_identity_uuid_map_matches_reviewed_anchor fire on identity events (mints, enrichments, supersedes). suspect_segments == ["LNU02374597"] fires on those and also on any new date-like kept segment (for example a source_record_id ending .edition_2026), with no identity event. The repo documents all of these as needing curation.

• CI steps run locally (receipt 0.6.2, Python 3.14, Hypothesis 6.169, ruff 0.14.10):

  • build_series_catalog.py --check;
  • pytest tests/test_build_series_catalog.py (222 passed at ab6d9b9);
  • the module's doctests;
  • --verify-registry-append-only against the base registry;
  • ruff check on both changed Python files.

  I did not run the Arch-surface pytest list, the DB build or the wheel steps locally. No module outside this test imports the builder or reads the plan, and CI runs them.

• ledger/ is untouched.

axiom: n/a: catalog test and tooling change, no policy logic.

🤖 Generated with Claude Code

[MaxGhenis]

Adopting this PR to land it. Its authoring session can't be found in the session index, and the PR has had no activity since 2026-09-30.

Why it matters now: the ledger branch's required "Arch checks" job has been red since 598f394 (the stale stripped-spelling list in test_committed_catalog_is_current_and_valid), and branch protection requires it, strict. That blocks the Chronicle entity-fold PR for Max's ruling d353 (fold seven entity-drift lineage pairs; thesis side is ThesisInstitute/thesis#342). This PR is the durable fix.

Plan: bring the branch up to date with codex/thesis-ledger-facts (two resolver appends landed since, 2ed9bae and d7e45bc), let CI rerun, get an independent GPT-6.1 Sol review at the updated head, and merge only if CI is green and the review agrees. #306 also re-pins the same list and will need a rebase after this lands; I am not touching it.

Session: Resolve duplicate Chronicle lineages for three BLS concepts (local_c7ccacf7-d314-449b-b837-d0cd5835d3e0).

[MaxGhenis]

Update from the adopting session. I am not landing this PR.

The independent GPT-6.1 Sol review at 0852fcd requested changes (full report: ~/chief-of-staff/state/chronicle-lineage/reviews/pr309-review.md on Max's machine). Summary:

1. Major. Strips from measure.concept and source_record_id are merged without field or position, and continuation needs only the canonical concept and spelling template. The reviewer executed a crafted row (concept abs.cpi.2026_09.all_groups.yoy, record id abs.cpi.all_groups.yoy.2026_09.first_print) that minted a new family pattern abs.cpi.{P}.all_groups.yoy and passed the audit. The exact list would have rejected it. Fix: scope reviewed decisions to identifier field and full family pattern.
2. Minor. The corruption self-test's derived unprinted week asserts global absence but only looks at initial-claims weeks, so another concept's valid continuation could fail it. Assert the (spelling, us.dol.initial_claims.sa) pair instead.
3. CI at this head was not verified by the reviewer (no network).

#344 (Max's ruling d353) no longer depends on this PR. It re-pins the existing list as routine maintenance, the way #244 did, and carries the same Hypothesis dev-dependency hunks as this PR, so the two merge cleanly in either order. A follow-up session to address the findings above is proposed to Max.

[MaxGhenis]

Merged at head ab6d9b9f (merge commit 8c4a0f86) on these gates:

• gh pr checks exited 0 on the latest runs: Append gate, Arch checks and Trusted base append gate all passed, and so did Arch bundle drift.
• mergeable was MERGEABLE (merge state CLEAN), the PR was not a draft, and no changes were requested.
• Independent reviews by GPT-6.1 Sol (Subfleet review --tier hard):
  • 0852fcd6: REQUEST_CHANGES. One MAJOR: a reviewed template could authorize a new identifier shape. One MINOR: the corruption self-test's unprinted week.
  • 5fa0c34a: APPROVE. One MINOR (month-name coverage in the next-period test) and three documentation NITs.
  • ab6d9b9f (final head, after merging Fold seven entity-drift lineages into their economy/aggregate siblings #344): APPROVE. One PR-body NIT, now fixed.
• Locally at ab6d9b9f:
  • tests/test_build_series_catalog.py: 222 passed;
  • --check: current, 221 series;
  • doctests and ruff check clean;
  • registry append-only against the base;
  • all 14 mutants killed.
• ledger/ is byte-identical to the base.