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.