DisMech compact evidence vocabulary + typed citation sidecar

[AdaWorldAPI]

Lands only what measurement has closed. The endpoint codebook and DisMech-local identity scheme are deliberately left open — see "Not frozen" below.

Scope note: this branch previously carried an S3.0 CausalLiteral address type. It was withdrawn (e4a3cba) after the audit found no falsifier requiring a new address type; the #973 retraction was kept. This PR body describes what the branch actually carries now.

Measured first, then written

All numbers from the upstream corpus (monarch-initiative/dismech, 2,100 disorder files, fetched ephemeral to /tmp, never committed — the /tmp-fixture rule). Full census: E-DISMECH-CORPUS-CENSUS-1.

87.2 MB of strings — and the entire causal semantics is bits + ordinals.

causal_link_type has exactly the four CausalTopology states, so the CE64 bits 59+60 mapping is source-authoritative and never inferred from confidence, edge count, or predicate name:

state	count
DIRECT	9,073
INDIRECT_UNKNOWN_INTERMEDIATES	4,539
INDIRECT_KNOWN_INTERMEDIATES	3,978
UNKNOWN	408
total causal edges	17,998

Two experimental populations, not one

source_knows_intermediates() / mediator_unresolved() separate them:

• 3,978 IndirectKnownIntermediates — the oracle population; the source names the mediators, so they can be hidden and recovery measured.
• 4,539 IndirectUnknownIntermediates — the restraint control; the source itself does not know, so a reasoner that "recovers" a mediator there is hallucinating closure.

Success is two-sided: recovery sensitivity and epistemic restraint.

Source-side only — CE64 stays out of the bake

This module deliberately does not reference CausalEdge64. The durable causal overlay must not become a pile of hot reasoning registers; DismechTopology → CE64 bits 59..60 happens at hydration, in the consumer, as a 1:1 read of to_bits_2().

Every parse fails closed

UNKNOWN is a value the corpus asserts 408 times. Minting it from a parse failure would forge an assertion the source never made — pinned by unrecognised_topology_fails_closed_and_never_becomes_unknown, with the paired silence half proving the real token still parses.

Citation identity is never the title

The corpus is LLM-generated, so wording drifts between regenerations while the citation does not. Measured: 131,904 title occurrences over 31,361 distinct (4.21× reuse), 12.12 MB inline vs 3.58 MB keyed. A stable key already exists on ~104,700 occurrences (PMID dominant, plus DOI/ORPHA/CGGV/NCT/URL), so CitationKey is (namespace, id). Where no stable identifier exists, ContentAddressed says so explicitly rather than synthesising a bibliographic identity.

Falsifier: reference_identity_survives_a_title_rewording — same PMID, reworded title ⇒ same key, different ContentId; different PMID, identical title ⇒ different key, same ContentId.

Not frozen by this PR (measurement says don't)

• Endpoint codebook / DisMech-local identity — 63.6% of unprefixed endpoints are still unresolved, and the tail is three different kinds: provenance leaking into the endpoint slot (Orphanet 120, OMIM 32, ClinGen 23 — database names, not concepts), qualified mechanism propositions (Impaired Neurodevelopment 20) that must stay DisMech-local, and lexical variants.
• "MONDO-derived" applies to the disorders, not the edges — measured endpoint prefixes: no prefix 25.2%, HP 24.9%, GO 11.6%, NCIT 10.5%, hgnc 7.6%, CL 7.3%, MONDO 4.4%. OBO_CORE's five namespaces cover 32.7%.
• HP needs its own resolution domain — 1,169 labels exist in >1 namespace, 1,129 of them HP+MONDO. Of 19,738 resolved phenotype labels, 26.2% landed in MONDO not HP and 23.7% are ambiguous. Collapsing them misroutes phenotype edges onto same-named disease nodes — harder to detect than a missing edge. UBERON, by contrast, is near-collision-free (10 / 14,975 = 0.07%) and is the safe anchor.
• phenotypes[].category is 261 distinct — over the 255 Codebook cap, with Neurologic and Neurological among its top values. Needs normalization or a deliberate split; Codebook::intern returning None is the split signal, never a silent widening.

Deterministic resolution reached 80.7% on phenotype labels (55.3% exact + 23.2% case/punct + 2.1% singular) with no model invoked. The next rung is the HPO synonym table — Sensorineural Hearing Loss and Sensorineural hearing loss both fail, so that tail is a synonym miss, not a case miss.

Gates

• lance-graph-contract 1178/1178 (7 new)
• cargo fmt clean · cargo clippy --all-targets --no-deps -- -D warnings clean

Board

EPIPHANIES ×3 (I-STRINGS-ARE-CAM-INDEX-ONLY-1 operator ruling, E-DISMECH-CORPUS-CENSUS-1, plus the retained #973 retraction pair) · LATEST_STATE contract inventory.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 2481777a-6cae-4f35-a874-b027acea7a28

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_e7ad219e-3456-4047-bbcc-661830c44029)

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_89020a00-36c6-4f55-abd2-5ad859bf6b60)