[onexbot-occ-writer]

Summary

Binds each acceptance criterion of OMN-19970 to the omnimarket tests that decide it. The autobind companion bound all three to one check that counts def _publish in cli_seed.py, which proves a function exists and no criterion's behaviour, so the sweep could never read any of them as receipt-proven. Same shape as OMN-19997's onex_change_control#11828.

What changed

• contracts/OMN-19970.yaml: three appended dod_evidence items, source: manual, each with binds_ac, an ac_bindings entry and a supersedes_ac_binding retiring the grep binding on dod-OmniNode-ai-omnimarket-pr-3082 for that label. No existing entry is edited.
  • ac1-seeding-twice-leaves-the-same-rows → AC1: tests/test_omn19970_dev_seed.py::test_seeding_twice_leaves_the_same_rows_all_labelled_fixture.
  • ac2-measured-savings-exclude-fixtures → AC2: test_metering_excludes_fixture_rows_unless_asked (SQLite) and test_summary_view_keeps_fixture_savings_out_of_the_measured_total (real PostgreSQL).
  • ac3-fixture-label-reaches-the-rows → AC3: test_seed_wire_messages_land_labelled_through_the_runner_seam and test_runner_labels_from_the_envelope_tag_and_defaults_to_real (real PostgreSQL).
• drift/dod_receipts/OMN-19970/<item>/test_passes.yaml: one PASS receipt per item.
• Not done: the grep bindings on the omnimarket#3103 and #3104 items are left alone; those PRs are open and not this change's.

How it was verified

• Criterion hashes recomputed with onex_change_control.validation.ac_criteria.criterion_hash from the ticket's live criteria (read 2026-09-30 09:4xZ): all three match the hashes carried here.
• Each item's exact check_value run on omnimarket dev 532ec1f53 (on .201, PostgreSQL 16 binaries on PATH, nothing skipped): 1 passed (AC1), 2 passed (AC2), 2 passed (AC3). Negative control: the same selection against 43cbc9035^ (the commit before omnimarket#3082 merged) fails to collect, 2 errors, pytest exit 4.
• check_ac_binding_acceptance.py --ticket-bodies <live body>: exit 0. Control: binds_ac: AC9 is refused with ac_binding_unknown_criterion, exit 1.
• check_receipt_hardening.py --paths-file0 <the 4 changed files>: exit 0. Control: a zeroed contract_entry_sha256 is refused as a mismatch, exit 1; restored byte-identical, exit 0.
• check_contract_change_rebinds_receipts.py: self-test exit 0; gate "3 touched dod_evidence entr(ies) … every receipt bound to one of them re-verifies".
• validator_occ_append_only --ticket-id OMN-19970: "No append-only violations".
• AC3's exposure half: the lab readback on the lakshman lane (Runs exposure, 20 of 20 rows data_source: fixture) is posted on omnimarket#3082 and cited in the AC3 receipt. It is not a check here, because the hosted sweep cannot reach the lab lane.

Failure paths

n/a: this change adds evidence items and receipts; it writes no runtime receipt, verdict or report of its own.

Open defects

none

Local gates

• No gate was bypassed — no --no-verify, no SKIP=, no --no-gpg-sign, no core.hooksPath override. Every pre-commit and pre-push hook ran.
• Any hook that failed was fixed at the input, not worked around.

The one hook that failed on the first commit attempt, reject-deploy-gate-skip-token, failed on the laptop's Bash 3.2 ("requires Bash 4+"); it passed once Homebrew Bash 5.3 was first on PATH.

Not in this change

• A readback check for AC3's exposure rows: the sweep cannot reach .201, so it would fail by construction.
• Any edit to the autobind-generated items.

Evidence-Ticket: OMN-19970

[Patel230]

Merged as f4b841d9d3. OMN-19970's three acceptance criteria are now each bound to the tests that decide them: seeding twice leaves the same rows, measured savings exclude fixtures, and the fixture label reaches the rows.

Thank you!