Resolve copula predicate to subject's licensed type (issue #128 R12)

[konard]

Issue #128 R12 — resolve "Hawaii is a state." to the most-correct sense

This follow-up to PR #129 (which delivered R1–R11) addresses the deeper
requirement raised in review: the predicate "state" in "Hawaii is a state."
must resolve to the most correct sense in context —
Q35657 U.S. state
(Russian: Штат США) — not the
generic federated-state concept Q7275.

Root cause

A bare noun like "state" cannot be disambiguated on its own. But in a copula
sentence the subject already carries an asserted type: Hawaii instance of
(P31) Q35657 "U.S. state". The formalizer never used that licensed type to
constrain the predicate, so "state" stayed generic and the translation leaned
on a hardcoded English→Russian rule.

Fix — language-neutral copula type resolution at formalization time

A new pass promotes the copula predicate to the subject's asserted
instance-of / subclass-of type when their head words agree:

• Hawaii is a state. → "state" resolves to Q35657 → en.wikipedia.org/wiki/U.S._state
• Гавайи это штат. resolves identically via the same logic — no per-language rule

The bespoke english-us-state-predicate-to-russian-shtat translation rule is
removed; targetLabelFor now prefers the interlingua concept form. The Rust
core, its curated reference, and the committed WASM build are flipped to
Q35657 for JS↔Rust parity (rule_count 3→2; Q7275 retained as a documented
generic reference).

How to reproduce / verify

node examples/copula-type-resolution.js

Output (fully offline):

Sentence: "Hawaii is a state."
  "state" resolved to: Q35657 (U.S. state)
  Wikipedia: https://en.wikipedia.org/wiki/U.S._state
  English -> Russian: Гавайи это штат.
  transformations: english-article-omission, english-copula-to-russian-eto

Architecture

• js/src/copula-types.js — resolveCopulaTypes pass (dependency-injected to avoid a circular import).
• js/src/wikidata-entity-context.js — shared Wikidata-entity helpers used by both formalize.js and the new pass; also keeps formalize.js within the 1500-line budget.

Tests

• New offline example examples/copula-type-resolution.js exercises the full path.
• Updated JS integration/e2e tests (issue-16, -35, -37, -56, -61-wasm, -74) and Rust tests (issue35_translation, semantic_relations) assert the Q35657 outcome.
• Full suite green locally: JS unit 63 / integration 295 / e2e 82, cargo 70, and npm run check (lint, format, duplication, docs, parity, cache) all pass.

Docs

• docs/REQUIREMENTS.md and the issue-128 / issue-35 case studies updated (R12, RC7, solution plan, timeline, README).
• Changeset: .changeset/issue-128-copula-type-resolution.md (meta-expression: minor).

Closes #128.

[konard]

Working session summary

Everything is complete and green. Final summary:

Done — Issue #128 R12 fully addressed in PR #130

The deep fix: "Hawaii is a state." now resolves the predicate "state" to its most-correct contextual sense — Q35657 U.S. state (Russian Штат США) — instead of the generic federated-state concept Q7275, language-neutrally and without any per-language translation rule.

How it works: a new formalization pass promotes the copula predicate to the subject's asserted type (Hawaii instance of Q35657) when their head words agree. The same logic resolves "Гавайи это штат" identically, so the bespoke english-us-state-predicate-to-russian-shtat rule was removed.

Architecture changes:

• js/src/copula-types.js — the dependency-injected resolveCopulaTypes pass.
• js/src/wikidata-entity-context.js — shared Wikidata-entity helpers used by both formalize.js and the new pass (also brought formalize.js from 1711 → 1463 lines, under the 1500-line budget).
• Rust core / curated reference / committed WASM flipped to Q35657 for JS↔Rust parity.

Verification (all green):

• npm run check — lint, format, duplication, docs, parity, cache ✓
• JS tests: unit 63, integration 295, e2e 82, 0 failures ✓
• cargo test: 70 passed ✓
• Offline example examples/copula-type-resolution.js prints the expected Q35657 / U.S._state / "Гавайи это штат." output ✓
• CI on commit 8269caa: Broken Link Checker, JS Checks, Rust Checks — all success ✓

Docs: docs/REQUIREMENTS.md, issue-128 case study (R12, RC7, plan, timeline, README), issue-35 supersession notes, and changeset .changeset/issue-128-copula-type-resolution.md (meta-expression: minor).

PR #130 is updated, marked ready for review (no longer draft), working tree clean: #130

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $24.015280

📊 Context and tokens usage:

Claude Opus 4.8: (6 sub-sessions)

1. 116.7K / 1M (12%) input tokens, 30.0K / 128K (23%) output tokens
2. 113.5K / 1M (11%) input tokens, 40.5K / 128K (32%) output tokens
3. 111.4K / 1M (11%) input tokens, 39.8K / 128K (31%) output tokens
4. 117.0K / 1M (12%) input tokens, 36.9K / 128K (29%) output tokens
5. 116.5K / 1M (12%) input tokens, 33.8K / 128K (26%) output tokens
6. 78.9K / 1M (8%) input tokens, 21.5K / 128K (17%) output tokens

Total: (45.2K new + 617.9K cache writes + 27.1M cache reads) input tokens, 242.7K output tokens, $23.729123 cost

Claude Haiku 4.5:

• 96.4K / 200K (48%) input tokens, 7.6K / 64K (12%) output tokens

Total: (2.8K new + 93.7K cache writes + 1.3M cache reads) input tokens, 7.6K output tokens, $0.286157 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Main model: Claude Opus 4.8 (claude-opus-4-8)
• Additional models:
  • Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (11626KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.