OCN 1.3.0 — the chronicle layer: openings that carry their history

Release notes: docs/release-ocn-1.3.0-notes.md

Artefacts

file	what it is
ocn-1.csv	the canonical catalogue, 14 columns
ocn-1.json	the whole catalogue as one JSON document (schema ocn.catalog.v1), with SAN movetext and the pipe-packed columns pre-split
ocn-1.positions.tsv	one row per concrete slug: fen_key, full FEN, transposition-group size
ocn-1.lichess-xref.tsv	Lichess opening lines mapped to OCN slugs
ocn-1.eco.tsv	scalar join table, one row per (slug, ECO code)
ocn-1.eco-divergence.tsv	rows whose OCN class letter is absent from their ECO codes, with rationale keys into the spec
ocn_chess-*.whl, ocn_chess-*.tar.gz	the ocn-chess Python package, catalogue bundled, zero runtime dependencies
SHA256SUMS	sha256 of every file above

Built from the tag by .github/workflows/release.yml. The
canonical source is catalog/ocn-1.csv; everything else is
derived and regenerable.

OCN 1.2.1 — the launch release: public repo, installable package, honest spec

Release notes: docs/release-ocn-1.2.1-notes.md

Artefacts

file	what it is
ocn-1.csv	the canonical catalogue, 14 columns
ocn-1.json	the whole catalogue as one JSON document (schema ocn.catalog.v1), with SAN movetext and the pipe-packed columns pre-split
ocn-1.positions.tsv	one row per concrete slug: fen_key, full FEN, transposition-group size
ocn-1.lichess-xref.tsv	Lichess opening lines mapped to OCN slugs
ocn-1.eco.tsv	scalar join table, one row per (slug, ECO code)
ocn-1.eco-divergence.tsv	rows whose OCN class letter is absent from their ECO codes, with rationale keys into the spec
ocn_chess-*.whl, ocn_chess-*.tar.gz	the ocn-chess Python package, catalogue bundled, zero runtime dependencies
SHA256SUMS	sha256 of every file above

Built from the tag by .github/workflows/release.yml. The
canonical source is catalog/ocn-1.csv; everything else is
derived and regenerable.

OCN 1.2.0 — diacritic-true names, audited ECO, unconditional gate

Summary

Everything the 360 audit rated P0/P1 on the catalogue itself is closed:

• Diacritic normalization, Tiers 1+2 — 10 + 4 eponym surnames respelled
  to the person's orthography across all naming columns (López, Grünfeld,
  Réti, Sämisch, Maróczy, Göring, Hübner, Löwenthal, Hromádka, Møller;
  Mikėnas, Krejčík, Opočenský, Pelikán). Policy + evidence:
  docs/diacritic-normalization-map.md. Sørensen and Würzburger stay
  parked pending per-row referent evidence.
• ECO corrections — 12 rows double-sourced (audit claim + independent
  longest-SAN-prefix derivation against lichess-org/chess-openings):
  London family A48→D02 (10), Dragon Yugoslav B76→B78, QID Nimzowitsch
  E17→E15. Four audit claims were refuted by the derivation and left
  unchanged (record in docs/eco-corrections-dry-run.md).
• Naming hygiene — 24 identity aliases dropped, 2 whitespace typos
  fixed; the two checks promoted to errors.
• Phantom + duplicate-name decisions executed
  (docs/phantom-and-duplicate-name-decision.md): path-marker children
  spec-blessed and ECO-aligned (2 rows); the four duplicate canonical
  names made unique (5 rows). Both validator allowlists now empty.
• Validator naming + ECO waves — checks 13–20 live: global name
  uniqueness, banned characters (middle dot, invisibles, controls),
  whitespace, identity-alias, diacritic regression guard (20 banned ASCII
  variants), same-moves-ECO consistency, plus --audit-naming /
  --audit-eco sweep heuristics. The default gate runs unconditional —
  zero allowlists, zero standing warnings.
• Engine — third safety mode eco_legacy_only; every change in this
  release went through a manifest (8 in docs/manifests/), each with a
  committed dry-run record.
• Schema unchanged — 14-column catalogue; transposes_to/same_as
  layer untouched (unresolved_groups=0 reconfirmed).

Catalogue diff vs ocn-1.1.0

metric	value
rows	5,899 → 5,899 (0 added, 0 removed, 0 slug changes)
distinct rows changed	782
canonical_name	683
notes	416
aliases	113
historical_notes / attributed_to / attribution_source	20 / 19 / 19
eco_legacy	14
sha256 (release head)	255ab28006ed6ce5bdf713936298cba1cb067eaa859e5fe167cdea84acf34b9a
validator	5,899 entries, 0 warnings, --strict-chess green
tool tests	165

Applied lots, in order (manifest → dry-run record, all in docs/):

lot	mode	rows
naming-error-corrections (P0)	naming_strings_only	16
attribution-polish (P0)	attribution_fields_only	3
diacritic-tier1-normalization	naming_strings_only	663
diacritic-tier2-normalization	naming_strings_only	50
naming-hygiene	naming_strings_only	26
eco-corrections	eco_legacy_only	12
phantom-eco-align	eco_legacy_only	2
duplicate-name-renames	naming_strings_only	5

Downstream impact (chess-parquet)

• Parquet regen is mandatory: canonical_name feeds the parquet's
  name column; 683 values changed.
• Join keys are safe: zero ocn1 changes — anything keyed by slug,
  FEN or zobrist survives untouched. Only name-string joins break (and
  they were joining misspellings).
• Downstream gate, mirroring the 1.1.0 verification doc: regen
  openings.parquet, re-run the efcdb-openings join, confirm row count
  5,899 and zero unmatched slugs, then publish both artefacts together.

Release runbook (each step its own GO)

1. Freeze + gate — clean worktree at the release head; run
   validate.py --strict-chess (5,899/0), full test suite (165),
   audit_transpositions.py --summary (unresolved_groups=0); verify
   account escacsfigueres and tags ocn-1.0.2/1.0.3/1.1.0 intact.
2. GO tag — annotated tag ocn-1.2.0 at the head; tags are
   immutable once pushed.
3. GO push — push main + the tag.
4. GO regen — coordinate chess-parquet: regenerate, run the
   downstream gate above, record checksums in a
   release-ocn-1.2.0-downstream-verification.md.
5. Record — mark this document released (replace the draft header),
   update post-1.1-roadmap.md.

OCN 1.1.0 — fully resolved transposition catalogue

OCN 1.1.0 — fully resolved transposition catalogue

Release notes (draft — tag not yet created). Docs-only record;
no tag/release/upload accompanies this document.

• Proposed release: OCN 1.1.0 — fully resolved transposition catalogue
• Proposed tag: ocn-1.1.0 (minor bump — first slug-rename; slug
  identity is not guaranteed stable across this boundary)
• Target commit: current main (the release-cycle head; this
  notes commit or its successor)
• Previous release: ocn-1.0.3 (dd2abd3)

Summary

The transposition layer is fully resolved: every duplicate-FEN
group in the catalogue is now classified.

• 0 unresolved duplicate-FEN groups (was 116 at the start of the
  post-0.2 cleanup; 31 at the start of this push series).
• QID Miles/Petrosian slug-migration applied — the mislabelled
  a3/Nc3 Kasparov-Petrosian subtree was re-homed under
  E.QID.Pet.KPe (OCN's first slug-rename).
• Nimzo Bot/Kmo naming fixed — E.Nim.Sml.Kmo (spurious
  "Kmoch") → transposes_to E.Nim.Sml.Bot, relabelled "a3 Move
  Order"; "Kmoch" stays at 4.f3 (E.Nim.Fou), per Lichess E20.
• same_as / transposes_to contract unchanged — same two
  resolution channels, same semantics.
• Schema unchanged — 14-column catalogue, 13-column
  openings.parquet (incl. canonical_ocn1, zobrist).

Metrics (at the release head)

metric	value	vs ocn-1.0.3
catalogue rows	5,899	(1.0.3 baseline)
duplicate_groups	124	—
resolved_groups	124	—
unresolved_groups	0	down from the long-tail
multiple_canonical_groups	17	grew via the same_as programme
schema columns	14 (catalogue) / 13 (parquet)	unchanged

Compatibility / breaking note

⚠️ This is OCN's first slug-rename. Slug identity is not
guaranteed stable across release tags — ocn-1.1.0 establishes that
precedent.

• 10 canonical OCN slugs changed: E.QID.Mil.MLn.* → E.QID.Pet.KPe.* (the QID Kasparov-Petrosian subtree), and the
  duplicate E.QID.Mil.MLn was deleted.
• No FEN / moves_uci / zobrist / schema change for those
  positions — only the slug identity (and canonical_name
  "Miles" → "Kasparov-Petrosian") moved.
• Consumers pinning E.QID.Mil.MLn* slugs must update to the
  E.QID.Pet.KPe.* names. Position-lookup consumers joining on
  FEN/zobrist are unaffected.

Assets (candidate — sha256)

Regenerated against the migrated catalogue; local candidates pending
upload at tag time.

asset	size	sha256
ocn-1.positions.tsv	1,475,677 B	0e78c315c62b1c37a7267b5df3c82f98e1af4e1ae0dbe8e9d47fc68211d89d7d
openings.parquet	369,663 B	a9180bfcd4fee272f0a9a76e866ffbe4ca2311ea68f373b435a5aff75209a5c2
_efcdb_manifest.json	328 B	a8aa50e904328e240290c1f17c56578a3c71875629f953b901ff551ca4ede9f6

_efcdb_manifest.json: efcdb_version 1.3, rows 5899,
zobrist_variant polyglot-v1.0, source ocn-1.1.0-candidate-<commit>. Regenerate fresh at tag time if
the head moves; do not reuse ocn-1.0.3 checksums.

Downstream (chess-parquet)

The efcdb-cli openings producer was run against the migrated
catalogue and smoke-tested:

• openings.parquet rows 5,899; schema unchanged (13 cols);
  canonical_ocn1 rule 0 violations; transposes_to 112,
  same_as 34; distinct zobrist 5,765, 124 multi-row zobrist
  groups (= duplicate_groups).
• QID migration absorbed cleanly: 0 E.QID.Mil.MLn*, 10
  E.QID.Pet.KPe.* in the parquet — no producer code change.
• cargo test -p efcdb-openings -p efcdb-core → 15 + 10 passed.

Upgrade guidance (from ocn-1.0.3)

• If you pinned ocn-1.0.3, expect the QID slug names to change
  on upgrade (E.QID.Mil.MLn.* → E.QID.Pet.KPe.*); update any
  hard-coded slug references.
• canonical_ocn1 is still computed by the transposes_to rule
  (canonical_ocn1 = transposes_to if set, else ocn1).
• same_as semantics are unchanged — symmetric co-canonical
  pairs (17 groups), still joined many-to-one on zobrist downstream.
• Nimzo: the depth-3 Sämisch E.Nim.Sml.Kmo is now "a3 Move Order"
  (transposes to E.Nim.Sml.Bot); "Kmoch" is E.Nim.Fou (4.f3).

Open follow-ups (NOT in this release)

• E.Nim.Rub.Kmo's "Kmoch" is the same artifact as the fixed Nimzo
  case — a cosmetic relabel for a future pass (does not affect
  resolution counts).
• E.Nim.Sml.Kmo.MLn parent-chain quirk (cosmetic).

Both are documented in
nimzo-botvinnik-kmoch-apply-preflight.md;
neither blocks the release.

OCN 0.2 post-cleanup release

OCN 0.2 post-cleanup release.

Source:

• Tag: ocn-1.0.3
• Commit: dd2abd3

Catalogue:

• Rows: 5,905
• Schema: 14 columns with transposes_to and same_as
• transposes_to rows: 91
• same_as rows: 12
• Duplicate FEN groups: 130
• Resolved groups: 93
• Unresolved groups: 37
• Multiple-canonical groups: 6

Assets:

• ocn-1.positions.tsv
  sha256: 7b1af5a9748fe7c8a5159eb20ccf2ece94e1b39c763d5cb8f67af02b8f0f8933
• openings.parquet
  sha256: 19df2a4def9646b32a27307716e738de4bf9620b87d9928177dc90728bcd54be
• _efcdb_manifest.json
  sha256: f310558842e240beb7fc57cfcd0942102386de4cafe311edccd1687b4713580e

Downstream contract:

• Consumers must not assume one OCN row per zobrist.
• transposes_to maps non-canonical rows to canonical_ocn1.
• same_as preserves co-canonical literary identities.
• openings.parquet was generated with chess-parquet producer compatible with OCN 0.2.