Deferred from PR #26 (round-4 review). Currently _RunCompatibilitySignature computes harness-kind from ALL task runs in a run, so a run mixing a minimal provider and a command-template provider carries a run-wide harness-kind list. When aggregating, a mixed-harness run's minimal results can be wrongly blocked from combining with another run's minimal results, even though the selected results are the same kind.
The core guarantee (minimal never aggregates with command-template/droid) IS enforced and tested. This is a refinement for the mixed-harness-in-one-run case, which does NOT occur in the official v1.0 path (parity runs are minimal-only). Needs a per-harness result-selection refactor of the exporter, not a run-wide signature tweak — deferred to avoid a risky change under the review cap.
Fix: scope harness-kind compatibility to the harness kind(s) of the SELECTED task results being aggregated, not the whole run. Add an exporter test: a mixed-harness run's minimal results aggregate with another run's minimal results. Refs #12 #10.
Deferred from PR #26 (round-4 review). Currently
_RunCompatibilitySignaturecomputes harness-kind from ALL task runs in a run, so a run mixing a minimal provider and a command-template provider carries a run-wide harness-kind list. When aggregating, a mixed-harness run's minimal results can be wrongly blocked from combining with another run's minimal results, even though the selected results are the same kind.The core guarantee (minimal never aggregates with command-template/droid) IS enforced and tested. This is a refinement for the mixed-harness-in-one-run case, which does NOT occur in the official v1.0 path (parity runs are minimal-only). Needs a per-harness result-selection refactor of the exporter, not a run-wide signature tweak — deferred to avoid a risky change under the review cap.
Fix: scope harness-kind compatibility to the harness kind(s) of the SELECTED task results being aggregated, not the whole run. Add an exporter test: a mixed-harness run's minimal results aggregate with another run's minimal results. Refs #12 #10.