The normalizer and its original 44 tests were ported from the hardened Cuemon multi-dimensional coverage experiment:
- Repository: codebeltnet/cuemon.
- Branch:
experiment/sonar-normalized-coverage-20261005. - Verified implementation:
cc19d02643829c198e576887f230bc1053135605. - Successful Sonar acceptance run: 37308393043.
This is historical validation evidence, not a runtime dependency. The shared implementation contains no repository-specific expected counters, project names, released revision or experiment exclusions. Source layout and partition identity are explicit contracts; any identity-spread acceptance bound belongs to the caller.
Compiler-divergent methods can make ordinal/path correlation ambiguous. The experiment's ExceptionConverter.ParseJsonReader demonstrated that matching ordinal/path pairs can be attributed to different decision locations across compilation variants. Comparing structural signatures and source-line spread exposes this ambiguity but cannot resolve compiler decision ownership from OpenCover evidence alone.
Normalization provides deterministic logical aggregation based on the available OpenCover evidence, not mathematically perfect compiler-independent branch identity. Do not assume all extra decisions are suffixes, that the union always equals the richest variant, or that a spread guard bounds every semantic mismerge. A caller that requires stronger identity proof needs additional source/compiler evidence or an explicit acceptance policy.
No experiment-only coverage exclusions belong in production. Sonar retains its normal source universe and may count report-absent coverable lines as uncovered. The shared action verifies persisted report/summary consistency; it does not require production coverage to equal an experimental percentage.
Codecov continues consuming raw TestResults* execution reports through jobs-codecov@v1 and codecov-scan@v1. Sonar consumes one derived generic report only when the caller opts in. These independent representations are deliberate.
Validate static metadata, the normalizer tests, action/workflow contracts, unchanged default OpenCover arguments and a controlled normalized fixture before publishing the approved backward-compatible additions through sonarcloud-scan@v2 and jobs-sonarcloud@v3. Tag updates and publication require the repositories' normal approval step.
Only after both existing major tags expose the tested inputs, make a normal PR branch from Cuemon main with this addition to the Sonar job in .github/workflows/verify.yml:
coverage-mode: normalized
coverage-source-root: src
# Cuemon acceptance guard from reviewed evidence; never a universal default.
coverage-max-identity-spread: '26'Do not merge the disposable experiment branch. Keep the Linux/Windows matrices, Codecov, CodeQL, source universe and release/deploy behavior unchanged. PR scanner logs must show exactly one generic coverage parse and zero raw OpenCover parses. A successful full PR verification is required before proposing one v10.8.0 assurance replay to repair canonical Sonar main. The replay requires separate approval; no release publication is needed to test it.
Move the normalizer model/parser/writer, tests, summary schema and provenance with their semantics intact. Rework hosting and invocation when DLCGEAR has a defined caller-managed runtime and artifact contract. Preserve raw evidence ownership, explicit source roots and compilation partitions, exclusive Sonar coverage configuration, persisted reconciliation and independent Codecov ingestion. Compiler-divergent identity correlation needs additional evidence rather than a new abstraction alone. No DLCGEAR runtime or architecture is introduced here.