Observed evidence
During the evidence-only import PR #36, CI #123 attempt 1, job 101967716754, failed the existing C09 real-CLI loopback test.
The same report contains:
- three completed command items, exact first/second/finish ordering, every exit code 0;
- two real automatic compactions, recovery contexts, and a completed turn;
- loopback peer received successful receipts for all three phases;
- structural JSONL observer recorded
c09_first.c09_receipt_ok=false, while second and finish were true.
The checker correctly reported BLOCKED. It was not bypassed. Product code, runtime helpers and this loopback test were unchanged by the import. Subsequent CI #124's CLI job passed with the same CLI/runtime/test implementation. An unchanged-head diagnostic repeat of #123 was also requested; original attempts remain in Actions.
Scope and uncertainty
The exact cause is not established: bounded structural diagnostics do not retain the underlying aggregated output. Do not assume the command failed, that the product regressed, or that passing once fixes the observer. The discrepancy is between the model-facing tool result and the recorded CLI-event receipt in a hosted loopback test.
Owner-initiated full live #27 (34140846679) independently passed all C01-C16, including exact C09 receipts and finite C08 completion, before this import. Its immutable archive must not be rewritten or relabeled. This ticket tracks conformance-test reliability, not a missing #27 capability.
Follow-up acceptance
Capture only bounded structural facts needed to distinguish a missing/truncated/fragmented aggregated-output receipt from a parser defect; compare against the pinned upstream 0.153.4 execution/event path. Reproduce with the actual CLI and add an executable regression before changing behavior. Preserve exact receipts, complete-process assertions, sandbox, authentication isolation, and privacy. Do not turn retries, substring-only peer observations, or partial completion into passing qualification evidence.
Observed evidence
During the evidence-only import PR #36, CI #123 attempt 1, job
101967716754, failed the existing C09 real-CLI loopback test.The same report contains:
c09_first.c09_receipt_ok=false, while second and finish were true.The checker correctly reported BLOCKED. It was not bypassed. Product code, runtime helpers and this loopback test were unchanged by the import. Subsequent CI #124's CLI job passed with the same CLI/runtime/test implementation. An unchanged-head diagnostic repeat of #123 was also requested; original attempts remain in Actions.
Scope and uncertainty
The exact cause is not established: bounded structural diagnostics do not retain the underlying aggregated output. Do not assume the command failed, that the product regressed, or that passing once fixes the observer. The discrepancy is between the model-facing tool result and the recorded CLI-event receipt in a hosted loopback test.
Owner-initiated full live #27 (
34140846679) independently passed all C01-C16, including exact C09 receipts and finite C08 completion, before this import. Its immutable archive must not be rewritten or relabeled. This ticket tracks conformance-test reliability, not a missing #27 capability.Follow-up acceptance
Capture only bounded structural facts needed to distinguish a missing/truncated/fragmented aggregated-output receipt from a parser defect; compare against the pinned upstream 0.153.4 execution/event path. Reproduce with the actual CLI and add an executable regression before changing behavior. Preserve exact receipts, complete-process assertions, sandbox, authentication isolation, and privacy. Do not turn retries, substring-only peer observations, or partial completion into passing qualification evidence.