README_READ is not a depth. It is the verdict "not verified" wearing a depth's name, and it is false in 100% of the cases it is assigned.
Where
src/lib/sources/code-inspection.ts:
depth: verified ? (hasEvalArtifacts ? "BENCHMARK_FOUND" : "CODE_INSPECTED") : "README_READ",
That README_READ branch is reachable only when examined.length > 0 — an inspection that opened no files returns INSPECTION_FAILED instead. So README_READ always means "we opened and read source files and found no qualifying match", and never means "we only read the README".
Why it matters
depth is the single field every downstream consumer keys on:
claim-ledger.ts — inspectionDepthByRepo()source-coverage.ts — DEPTH_EVIDENCE_CEILING, and now the check that a required CODE_VERIFIED was actually obtainedscripts/audit-run.ts
Observed
Run cmtg7821u inspected three repositories and read 36 source files between them — Iron-mem opened 12 of 48 including src/bench.rs, synapse read src/synapse/encoding.py. All three stored "depth": "README_READ".
Everything downstream, every human reading the record, and the report itself then treated those as repositories whose code had not been read. This misled the maintainer directly: the state was reported to a user as "all three stopped at README_READ, none reached CODE_INSPECTED", which is the opposite of what happened.
It is this project's recurring defect in its purest form — a search that ran and returned nothing, stored as a search that never ran.
Suggested fix
Split the field:
depthReached — derived from what was actually opened. examined.length > 0 means at least code-inspection-level looking; eval artifacts mean benchmark level.verdict — CODE_VERIFIED | CODE_NOT_VERIFIED.
Promotion in maxEvidenceClassForSource then requires both (depthReached >= CODE_INSPECTED && verdict === "CODE_VERIFIED"), which preserves today's ceiling exactly while making the depth field true. Rename README_READ to something it can honestly mean, or drop it from the ladder.
Good first issue?
Yes, with care. The change is small and local, but every consumer of depth must be updated in the same commit, and each has a test.
README_READis not a depth. It is the verdict "not verified" wearing a depth's name, and it is false in 100% of the cases it is assigned.Where
src/lib/sources/code-inspection.ts:That
README_READbranch is reachable only whenexamined.length > 0— an inspection that opened no files returnsINSPECTION_FAILEDinstead. SoREADME_READalways means "we opened and read source files and found no qualifying match", and never means "we only read the README".Why it matters
depthis the single field every downstream consumer keys on:claim-ledger.ts—inspectionDepthByRepo()source-coverage.ts—DEPTH_EVIDENCE_CEILING, and now the check that a requiredCODE_VERIFIEDwas actually obtainedscripts/audit-run.tsObserved
Run
cmtg7821uinspected three repositories and read 36 source files between them —Iron-memopened 12 of 48 includingsrc/bench.rs,synapsereadsrc/synapse/encoding.py. All three stored"depth": "README_READ".Everything downstream, every human reading the record, and the report itself then treated those as repositories whose code had not been read. This misled the maintainer directly: the state was reported to a user as "all three stopped at README_READ, none reached CODE_INSPECTED", which is the opposite of what happened.
It is this project's recurring defect in its purest form — a search that ran and returned nothing, stored as a search that never ran.
Suggested fix
Split the field:
depthReached— derived from what was actually opened.examined.length > 0means at least code-inspection-level looking; eval artifacts mean benchmark level.verdict—CODE_VERIFIED|CODE_NOT_VERIFIED.Promotion in
maxEvidenceClassForSourcethen requires both (depthReached >= CODE_INSPECTED && verdict === "CODE_VERIFIED"), which preserves today's ceiling exactly while making the depth field true. RenameREADME_READto something it can honestly mean, or drop it from the ladder.Good first issue?
Yes, with care. The change is small and local, but every consumer of
depthmust be updated in the same commit, and each has a test.