The live E2E stand currently drops real settled rows whose cost is estimated or unknown. In devtools/e2e_live/run_live_lanes.py, lane_spend requires cost_final is True before counting either dollars or unknown costs. The product writers record_subscription_session and record_unmetered_external_dispatch produce cost_final=False for those cases. At cc6a886, an unknown subscription row therefore reports (0, 0) while the canonical projection reports unknown_unmetered=1. An estimated $4.10 row also reports (0, 0); after settlement the stand can admit another $5 reservation against a $6 run budget. The control using a confirmed $4.10 payment refuses that reservation correctly. The existing test supplies an unknown row with cost_final=True and misses the real writer shape.
The same landed stand update checks landing skip flags only for commit_reviewed. vcs_commit_reviewed is separately registered with the same parameters, and tools.jsonl preserves the called name. A successful alias call with skip_advisory_review=True therefore produces an empty landing_skip_flags list. The SM1 prompt currently asks for the canonical name, so this is a checker gap rather than evidence that a recorded run used the alias.
Both findings came from independent Astra/Fable review of the already-landed #704 changes and were checked against the actual writers and consumers. They are separate from the #552 reviewer-identity repair. Preserve the complete ledger identity and distinguish known cost, finality, unresolved reservations and unknown cost through the existing accounting reader. Cover real unknown/estimated writer rows and both registered commit-tool names; keep the working confirmed-cost and no-skip paths. No changes to product review enforcement or pricing policy are needed.
Source: 8e847e7
The live E2E stand currently drops real settled rows whose cost is estimated or unknown. In devtools/e2e_live/run_live_lanes.py, lane_spend requires cost_final is True before counting either dollars or unknown costs. The product writers record_subscription_session and record_unmetered_external_dispatch produce cost_final=False for those cases. At cc6a886, an unknown subscription row therefore reports (0, 0) while the canonical projection reports unknown_unmetered=1. An estimated $4.10 row also reports (0, 0); after settlement the stand can admit another $5 reservation against a $6 run budget. The control using a confirmed $4.10 payment refuses that reservation correctly. The existing test supplies an unknown row with cost_final=True and misses the real writer shape.
The same landed stand update checks landing skip flags only for commit_reviewed. vcs_commit_reviewed is separately registered with the same parameters, and tools.jsonl preserves the called name. A successful alias call with skip_advisory_review=True therefore produces an empty landing_skip_flags list. The SM1 prompt currently asks for the canonical name, so this is a checker gap rather than evidence that a recorded run used the alias.
Both findings came from independent Astra/Fable review of the already-landed #704 changes and were checked against the actual writers and consumers. They are separate from the #552 reviewer-identity repair. Preserve the complete ledger identity and distinguish known cost, finality, unresolved reservations and unknown cost through the existing accounting reader. Cover real unknown/estimated writer rows and both registered commit-tool names; keep the working confirmed-cost and no-skip paths. No changes to product review enforcement or pricing policy are needed.
Source: 8e847e7