feat: Grammar unlocks — verb table, Disambiguator trait, Quantum mode, Animal Farm harness

[AdaWorldAPI]

Summary

Implements epiphanies E3, E4, E5, E7, E8 from PR #279 outlook review. Stacks on claude/grammar-fixes-r2-2026-04-29.

• E3 — 144-cell verb-role table. VerbFamily (12 families: Becomes..Dissolves) × Tense (12) = 144 cells. Each cell holds SlotPrior (TEKAMOLO slot expectations). default_table() seeds Becomes → temporal+modal high, Causes → kausal high. Parsing reduces to table lookup + truth aggregation.
• E4 — Trajectory-as-audit-hash bridge. TrajectoryHash = [u64; 256] (16384-bit semantic hash). Trajectory::binarize(), audit_hash_u64(), trajectory_distance(). Cross-PR bridge: DeepNSM trajectory replaces statement_hash: u64 in audit log.
• E5 — Generalized Disambiguator trait. Disambiguatable trait + disambiguate_general<T>() free function. Domain-agnostic: same machinery for coreference, RLS predicate selection, triplet completion, style dispatch.
• E7 — Animal Farm forward-validation scaffold. EpiphanyPrediction, GroundTruthBeat, HarnessMetrics (precision, recall, arc-shift F1, direction accuracy). evaluate() function. 4 tests including perfect-confirmation and no-match cases.
• E8 — Quantum mode. PhaseTag(u128) with from_angle/to_angle/distance (Hamming on 128 bits). HolographicMode { SinglePhase, PerRole }. Crystal vs Quantum is a knob, not a separate stack.
• YAML robustness. Flow-map parsing (nars: { primary: X }), whitespace-aware # comment stripping, TekamoloSlot::Instrument variant (no longer silently mapped to Modal).
• Awareness persistence. AwarenessSnapshot + snapshot()/restore() round-trip for cross-session persistence.
• top_nars_inference fix. Bootstrap now keeps prior's primary (was falling through to fallback at frequency == 0.5).

Stats

• 9 files changed, +845 / -22 lines
• 3 commits (2 workers + 1 meta integration)
• 302 contract tests + 76 deepnsm tests + 4 animal_farm tests = 382 tests pass

Test plan

• cargo check -p lance-graph-contract — green
• cargo check --manifest-path crates/deepnsm/Cargo.toml — green
• cargo test -p lance-graph-contract --lib — 302 passed
• cargo test --manifest-path crates/deepnsm/Cargo.toml --lib — 76 passed
• cargo test --manifest-path crates/deepnsm/Cargo.toml --test animal_farm_harness — 4 passed
• PhaseTag from_angle round-trips within 0.001 tolerance
• 144-cell table has all 144 cells, lookup returns uniform for unset
• Disambiguator picks closest-to-focal on scalar items
• Snapshot → restore round-trips all truth values

https://claude.ai/code/session_01SbYsmmbPf9YQuYbHZN52Zh

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5f1745969f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Encode π using the same low-64-bit phase representation

PhaseTag::pi() currently sets the full tag to u128::MAX / 2, but to_angle() only decodes the low 64 bits; that low half is all 1s for u128::MAX / 2, so PhaseTag::pi().to_angle() yields approximately 2π instead of π. Any caller using pi() as a canonical opposite-phase constant will get the wrong angle and phase comparisons will be skewed. This should be encoded via the same low-half scheme as from_angle (e.g., half of u64::MAX in the low 64 bits).

Useful? React with 👍 / 👎.