ruff_python_spo: residual ledger for the plain arm — silent skips become named, addressed rows

[AdaWorldAPI]

Summary

First increment of the self-adaptive drill loop over the plain-Python arm (#95): before any proposer can turn residue into data-as-config rows, the residue has to exist as data. The plain arm's declared gaps — module-level constants invisible, nested classes unwalked, complex annotations unresolved, non-literal assigns dropped, unresolvable bases dropped, unparsable files silent — were all silent skips. This PR makes each one a PlainResidual row: a named reason (closed enum, no catch-all — the intake-arm slag doctrine), a module + class + name address, and a reason-specific detail.

Strictly additive: extract_plain_from_source_with_residuals / extract_plain_with_residuals return (ModelGraph, Vec<PlainResidual>); the existing entry points delegate and drop the ledger. A dedicated test pins that the graph half is identical between the two paths on a fixture exercising every residual reason at once — observation, never behaviour change.

The CURIE channel enters as measurement, not harvest

A module- or class-level string constant whose value is CURIE-shaped (PREFIX:localid) gets a curie_constant row carrying the prefix only. The local id is opaque by design: OBO locals and SNOMED sctids alike carry no structure in the identifier — hierarchy for such vocabularies lives in the ontology's edges (the terminology horseshoe), so drilling past the prefix is the resolution target's job, never string parsing. The shape check is two-sided and tested: URLs (rest starts /), Windows paths (backslash), clock times (digit prefix), and spaced strings all stay module_constant.

Measured — three corpora + one adversarial

corpus	rows	dominant reasons	curie
dismech (84 files)	1,452	non_name_assign_target 493 · non_literal_assign 455 (451 = call) · module_constant 386 · unresolved_annotation 96 (all binop unions)	22 rows, 12 prefixes: KISAO 4, infores 3, MONDO/dismech/linkml/sqlite/xsd 2, Monarch/RO/biolink/rdfs… 1
A2UI sdk (127 files)	299	module_constant 239	0
ruff/scripts (23 files)	67	module_constant 45 · unresolved_annotation 12	0
ruff_linter fixtures (1,586 deliberately-malformed files)	3,786	exit 0; unparsable_source 9	—

Three readings:

1. The histogram concentrates — 2–3 reasons dominate per corpus. There is something to drill toward; the loop deserves its next increment (a proposer that emits candidate config rows from the dominant buckets, gated by an exact-count coverage check).
2. curie_constant fires exactly where ontology work lives (dismech) and is silent on both non-ontology corpora — the can-fire / can-stay-silent pair, measured at corpus scale rather than only in unit fixtures.
3. sqlite: among the prefixes is the design argument in one row: a connection-string URI is CURIE-shaped but not an ontology reference, so prefix→codebook-target binding must be a config decision a measurement gate ratifies, never an auto-lift.

Falsifier discipline

Every new test asserts exact counts (==). Each residual-recording site was verified by a disable run — mutation anchor asserted to exist, test observed red, restored, full suite green: the slash exclusion (URL test goes red), the class-level CURIE recording (class-level curie row panic), and the module-constant recording. One first-attempt mutation failed to compile and thereby verified nothing — caught because the filtered run's output was empty rather than red, redone with a compiling mutation before being counted.

Test plan

• cargo test -p ruff_python_spo — 70/70 (66 pre-existing + 4 new)
• cargo clippy -p ruff_python_spo --all-targets -- -D warnings — clean (one doc_markdown fixed en route)
• cargo fmt + RUSTDOCFLAGS="-D warnings" cargo doc --no-deps -p ruff_python_spo — clean (the R2IL intake arm: plan, implementation spec, and ruff_r2il crate (PR 1) #94 gate holds over the new surface)
• uvx prek run --from-ref main --to-ref HEAD — all hooks pass, git status clean after (no fixer rewrote anything)
• Census reproduced over all four corpora above; adversarial corpus exits 0

---

Generated by Claude Code

[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_7a579ae2-70eb-4ca7-92d5-b9c1e25921e2)

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.