Problem and observable outcome
T-09.1's source-bound benchmark evidence has an incomplete ingress validator. xtask/src/benchmark_baseline/artifact.rs::validate searches for expected metadata lines and counts scenario/profile rows; it does not admit the complete canonical report grammar. A report containing both the expected git-commit and a conflicting git-commit is accepted. Existing unit fixtures also demonstrate that bare scenario<TAB>index and profile<TAB>index rows pass admission without headers or metrics.
Outcome: only one complete, unambiguous keep.streaming-cas-baseline/v1 report bound to the captured environment can reach artifact publication.
Reproduced RED
On original main-equivalent source in copy-isolated Docker, Rust 1.96.0, dedicated target: append metadata<TAB>git-commit<TAB>ffffffffffffffffffffffffffffffffffffffff<LF> to the accepted existing report fixture. validate returns success. The law conflicting_duplicate_source_metadata_refuses_before_publication fails at conflicting source coordinates were admitted. The probe was restored after execution; no production change or bad artifact was published.
Scope
Implement bounded canonical report admission, validated state before publication, precise typed admission failures, and permanent corruption/fuzz regression witnesses. Validate schema, complete unique metadata, exact headers, ordered unique scenario/profile catalogs, field widths, canonical numeric fields, configured sample/warmup coordinates, threshold posture, and required counter relationships. Bind captured source/compiler/host coordinates without accepting conflicting duplicates. Reject unknown/trailing rows and noncanonical framing.
Acceptance checks
Prerequisites and safe merge boundary
No prerequisite on #71/#74, GC, migration or performance optimization: the current valid report format can be admitted strictly on main. PR #140 fixes unrelated hardware-dependent test setup; it is needed to clear that full-workspace Docker limitation, not to make report grammar correct. Shared files do not create semantic dependency edges. The fix must preserve the currently valid report format and leave main working independently.
Exclusions
No new timing measurements, thresholds, CI performance gates, benchmark scenarios, wire format/version change, forged environment evidence, or native test execution. T-09.2 and T-09.3 were originally unfinished and are excluded from this completed-task audit.
Refs #131 and #132. This is one executable issue for one coherent admission boundary, not a tracking container. The fixing PR must include regression evidence, issue linkage and resulting integration traceability.
Problem and observable outcome
T-09.1's source-bound benchmark evidence has an incomplete ingress validator.
xtask/src/benchmark_baseline/artifact.rs::validatesearches for expected metadata lines and counts scenario/profile rows; it does not admit the complete canonical report grammar. A report containing both the expected git-commit and a conflicting git-commit is accepted. Existing unit fixtures also demonstrate that barescenario<TAB>indexandprofile<TAB>indexrows pass admission without headers or metrics.Outcome: only one complete, unambiguous
keep.streaming-cas-baseline/v1report bound to the captured environment can reach artifact publication.Reproduced RED
On original main-equivalent source in copy-isolated Docker, Rust 1.96.0, dedicated target: append
metadata<TAB>git-commit<TAB>ffffffffffffffffffffffffffffffffffffffff<LF>to the accepted existing report fixture.validatereturns success. The lawconflicting_duplicate_source_metadata_refuses_before_publicationfails atconflicting source coordinates were admitted. The probe was restored after execution; no production change or bad artifact was published.Scope
Implement bounded canonical report admission, validated state before publication, precise typed admission failures, and permanent corruption/fuzz regression witnesses. Validate schema, complete unique metadata, exact headers, ordered unique scenario/profile catalogs, field widths, canonical numeric fields, configured sample/warmup coordinates, threshold posture, and required counter relationships. Bind captured source/compiler/host coordinates without accepting conflicting duplicates. Reject unknown/trailing rows and noncanonical framing.
Acceptance checks
Prerequisites and safe merge boundary
No prerequisite on #71/#74, GC, migration or performance optimization: the current valid report format can be admitted strictly on main. PR #140 fixes unrelated hardware-dependent test setup; it is needed to clear that full-workspace Docker limitation, not to make report grammar correct. Shared files do not create semantic dependency edges. The fix must preserve the currently valid report format and leave main working independently.
Exclusions
No new timing measurements, thresholds, CI performance gates, benchmark scenarios, wire format/version change, forged environment evidence, or native test execution. T-09.2 and T-09.3 were originally unfinished and are excluded from this completed-task audit.
Refs #131 and #132. This is one executable issue for one coherent admission boundary, not a tracking container. The fixing PR must include regression evidence, issue linkage and resulting integration traceability.