[MaxGhenis]

Implements Max's ruling d353 (2026-09-27): seven series whose hand-recorded May 2026 rows carry a bespoke entity, while every resolver print carries economy/aggregate, fold into their economy/aggregate lineage. Companion PRs: ThesisInstitute/thesis#342 (merged: stages the fold seed in the resolver's catalog regeneration, accepts the new catalog key) and ThesisInstitute/thesis#346 (re-stamps three docket pins and adds their BLS API templates; merges after this PR and the next resolver run).

The defect

Catalog identity is (concept, geography, entity). Alias inheritance never crosses an entity slice, and the curated-merge recipe restores the split on regeneration (reproduced 2026-09-22). So seven series held two observed lineages each (the "entity-drift families" in docs/catalog-curation-backlog.md). In Thesis, a docket entry that declares a source binding must find its source id in exactly one candidate lineage, which held back BLS API templates for the unemployment rate and CPI headline and core.

The fix

scripts/build_series_catalog.py reads a curated seed, ledger/seeds/entity_folds.json (chronicle_entity_folds_v1). Each fold names a concept, geography, absorbed entity, survivor entity, absorbed_names (the absorbed row's concept and aliases) and a note.

• Before identities are grouped, current rows of the fold's concept and geography that carry the absorbed entity and are spelled by one of the absorbed names are read (as copies) as carrying the survivor entity. The observation file is never changed; the immutable prefix is untouched.
• The absorbed row's names join the survivor's aliases; its binding retires with the --remint-note; the survivor keeps its UUID. The catalog commits to the seed as entity_folds_sha256 (absent without a seed; generator version stays 3).
• Every seed edit (new, retargeted or undone fold) is an identity change that needs --allow-remint --remint-note. Deleting the seed under a catalog that declares it is refused.
• Undo: mark a fold "undone": true. The absorbed identity revives with its original UUID, spellings it had absorbed earlier rejoin it, and exactly its declared names move back off the survivor. Deleting a fold entry is refused while git HEAD holds it.
• Refusals: absorbed_names must equal the absorbed row's names when the fold is made; a name already held in the survivor's slice is refused at fold time; while a fold stands the survivor must hold every absorbed name alone; a fold must name an identity the registry has bound and a survivor with a committed observed row; a folded row's own entity is still validated. A fold matching no current row is inert.

Regenerated artifacts

	Before (d7e45bc)	After
Catalog series	228	221
Observed series	184	177
Sum of observation_count	233	233
Registry lines	276	283 (seven retired, append-only)
Mints, supersedes		0, 0
stripped_segments, suspect_segments		unchanged

Survivors keep their UUIDs: db2f3857, 70ca2ecf, e5e402d5, 1d271205, 8c999902, 9668ecdb, dbfd5050. Retired: 8dbbd54f, 3e796803, d85c3c7f, e8e6e43e, d0212a1e, 978252a9, d5634fc4. Replaying the fold from d7e45bc's catalog and registry with the committed seed reproduces both files byte for byte. Marking all seven folds undone reproduces the pre-fold catalog exactly, apart from the seed and registry digests.

Invariants (stated and tested)

• Conservation: the observed series' observation_count sums to current_assertion_rows before and after any fold (Hypothesis property over any split of one series across two entities).
• No fold mints or supersedes; survivors keep their UUIDs; each absorbed binding retires exactly once.
• Determinism: regeneration with the same inputs is byte-identical.
• Exact reversibility: marking a fold undone restores the pre-fold catalog apart from the two digests.
• The observation journal is never written.
• Appends after a fold (new prints under either entity, superseding corrections of versioned absorbed rows) regenerate with the registry unchanged.

Also in this PR

• test_committed_catalog_is_current_and_valid's exact stripped-spelling list was stale since 598f394 (the branch head's required "Arch checks" has been red). This PR re-pins it to the current catalog; the fold changes no strip. Audit period strips by reviewed template, not an exact spelling list #309 proposes replacing the list with a reviewed audit.
• Hypothesis joins the dev dependencies (same pyproject.toml/uv.lock hunks as Audit period strips by reviewed template, not an exact spelling list #309). The runtime set the trusted append gate installs (--no-dev) is unchanged.
• Pre-existing and out of scope: the builder refuses a correction that supersedes a pre-versioning row (every immutable-prefix row), folds or not; Resolve catalog supersede links against every row's effective assertion id #306 addresses it. Documented in the backlog doc.
• If Put every file that decides a gate verdict on the gate surface #308 lands first, this PR's mix of ledger/** with the generator and pyproject.toml/uv.lock would become a mixed proposal for the append gate and would have to be split.

Review

Independent GPT-6.1 Sol reviews, rounds 3 to 7 (round 1 and 2 reviewed the proposal). Rounds 3 to 5 requested changes, each answered in a commit: legacy-correction claim narrowed, retarget ceremony, exact undo, entity validation, undo scoped to the explicit marker, absorbed names declared. Round 6 approved at 38716ee with minors; 0e1dc63 answers them and the round-7 delta review approved it. Reports and responses: ~/chief-of-staff/state/chronicle-lineage/reviews/ on Max's machine.

Locally at 0e1dc63: 198 tests pass, ruff 0.14.10 clean, --check green, doctests pass.

🤖 Generated with Claude Code

[MaxGhenis]

Merging at 0e1dc63 under Max's ruling d353 (2026-09-27: economy/aggregate survives, all seven entity-drift pairs, generator fold).

• CI: all required checks pass at this head (Append gate, Trusted base append gate, Arch checks; Arch bundle drift also passes).
• Independent review (GPT-6.1 Sol, subfleet run --task review --tier hard): rounds 3 to 5 requested changes, each answered in a commit; round 6 approved at 38716ee with minors; 0e1dc63 answers them and the round-7 delta review approved it (no blocker or major).
• Verified locally: 198 tests, ruff 0.14.10, --check green; a fresh fold from d7e45bc reproduces the committed catalog and registry byte for byte; marking all seven folds undone reproduces the pre-fold catalog apart from the two digests.
• Next: the first Thesis resolver run after this merge walks the site's ledger pin; then Give the three folded BLS series their templates and re-stamped pins ThesisInstitute/thesis#346 (re-stamped pins, BLS API templates) is refreshed, re-tested and merged.