Skip to content

Wave 1 Human Golden E2E closeout evidence

Issue: #73. Date: 2026-09-19. Mechanically derived from the actual operator run below; not a manual proof workflow.

Scope ruling: ingestion/publication is credited, not re-run

An independent review (2026-09-19) requested that preflight and the mutating run (ingestion/publication) legs also be executed and archived alongside verify, reading propositions 1-2 as requiring literal re-execution within this closeout run. Two things make that infeasible and, on the issue's own terms, unnecessary:

  1. run_vertical's own _require_absent_run_targets guard (and preflight's rerun check) correctly refuses to re-ingest a partition that is already durably published -- confirmed directly: preflight on this exact target reports rerun: FAIL -- canonical target natural identity already exists; refusing ambiguous rerun. Forcing a re-run would mean bypassing a real anti-duplicate-ingestion safety mechanism, not producing new evidence.
  2. Issue #73's own "Credited evidence" section states: "CREDIT rather than re-prove: the existing conformity harness already owns operator preflight/run/inspect/verify behavior; the existing golden Bybit fixture already proves source row-count/boundary/catalog expectations for the first vertical." Proposition 2's wording ("durable manifests/evidence and S13/S14 outcome are available") is a state, not an action -- consistent with crediting the already-published first vertical (V1 COMPLETE) rather than re-proving it here.

Durable evidence that ingestion/publication succeeded and remains valid is the inspect output already on record: dataset ff12b1c3-..., partition dt=2024-01-15 state valid, row_count 1105145, with manifest/content SHA-256 hashes and producer=human-e2e-operator-harness-v1.

Ruling (human decision owner, 2026-09-19): the credited-evidence reading is accepted; re-executing ingestion/publication against already-validated, archived data is unnecessary. This closeout is treated as PASSED on the verify-leg evidence below plus the already-durable ingestion/publication evidence cited above, not on a fresh re-run of either.

Prerequisite integration gate

Per issue #73's strict post-merge execution gate, this run is only valid if implement/wave-1 includes #58, #59, #60 and #65. Confirmed by ancestry:

candidate base = 15f6017fd975029eb5ccdb78d71924f984332dd3
                 (merge-base of this branch with origin/implement/wave-1)

Merge pull request #72 from NeoNix-Lab/agent/issue-58-16   -> #58  ancestor: yes
Merge pull request #74 from NeoNix-Lab/agent/issue-59-17   -> #59  ancestor: yes
Merge pull request #76 from NeoNix-Lab/agent/issue-60-19   -> #60  ancestor: yes
Merge pull request #77 from NeoNix-Lab/agent/issue-65-20   -> #65 (== base) ancestor: yes

Exact tested candidate

branch          = agent/issue-73-21
candidate SHA   = d6f9a438310211d4dacc213a3173dd7b76e4bd10

Operator run (verbatim, from the checked-out candidate SHA above)

Run on the Debian catalog host (neonix@homelab.local), where the canonical partition's physical Parquet storage root is a real absolute path; this Windows checkout can reach the PostgreSQL catalog over an SSH tunnel but not the physical file (resolve_partition_path requires an absolute storage root, which a Linux path is not under Windows pathlib). Real Bybit BTCUSDT historical SQLite source and PostgreSQL catalog (market_catalog, container market-catalog, already-published 2024-01-15 canonical partition from prior first-vertical work) -- no ingestion, no catalog/storage writes, tools/conformity_e2e.py verify is read-only per collect_preflight's/verify_vertical's own docstrings.

$ git rev-parse HEAD
d6f9a438310211d4dacc213a3173dd7b76e4bd10

$ python tools/conformity_e2e.py verify
GOLDEN EXPECTATION
  venue=bybit instrument=BTCUSDT
  interval=[2024-01-15T00:00:00Z, 2024-01-16T00:00:00Z)
  row_count=1105145 buy=553875 sell=551270
  first_exchange_ts=2024-01-15T00:00:00.492Z
  last_exchange_ts=2024-01-15T23:59:59.931Z
SCAN OBSERVATION
  row_count=1105145 buy=553875 sell=551270
  other_aggressor_side=0
  first_exchange_ts=2024-01-15T00:00:00.492Z
  last_exchange_ts=2024-01-15T23:59:59.931Z
  batches=17 max_batch_size=65536
LIFECYCLE
  initial=OPEN after_first_batch=READING final=COMPLETED
  completed_metadata_present=True
COVERAGE METADATA
  coverage_complete=True
  coverage_gaps=0
PROVENANCE
  result_identity=574aeab50876514bcfc4e7a6df1c3e4d5b0147b15cc8967e3da55a9723fa164b
  catalog_partition_count=1
BOUNDEDNESS TELEMETRY
  unavailable
CANDLE RESULT
  definition_identity=candle-definition-v1:sha256:94b3a5a99c873f0d2805f4d197b772047f8c13e8753ae8fb4314aafebf1440cc
  candle_count=1440
  first_candle={'bucket_start': '2024-01-15T00:00:00Z', 'bucket_end': '2024-01-15T00:01:00Z', 'open': '41731.1', 'high': '41794.1', 'low': '41723.2', 'close': '41759.8', 'volume': '201.935', 'trade_count': '1733'}
  last_candle={'bucket_start': '2024-01-15T23:59:00Z', 'bucket_end': '2024-01-16T00:00:00Z', 'open': '42500', 'high': '42529.6', 'low': '42494.1', 'close': '42508.1', 'volume': '133.328', 'trade_count': '855'}
  result_identity=historical-candle-result-v1:sha256:998dce0c37898a8d165bcd3090d930bc75bed3a4b8066dd9765535d0ff7425e8
GOLDEN E2E: PASS

$ echo $?
0

Focused test / integrity results (local checkout, same candidate SHA)

$ python -m compileall -q src tests
(clean)

$ python -m unittest discover -s tests
Ran 866 tests in 17.026s
OK (skipped=1)

$ python -m unittest tests.test_package_boundaries_v1
Ran 15 tests in 6.857s
OK

$ git diff --check
(clean, exit 0)

Acceptance mapping (issue #73)

  1. 58/#59/#60/#65 integrated before the run -- confirmed by ancestry above.

  2. Exact tested SHA recorded -- d6f9a438310211d4dacc213a3173dd7b76e4bd10.
  3. Reused existing harness/production seams (tools/conformity_e2e.py, quant_platform.application.conformity/golden_conformity) -- no new framework.
  4. tools/ remains an Application-only client (unchanged from prior review).
  5. Application owns the DataGateway -> D03 adaptation (observe_scan_with_candles, unchanged from prior review).
  6. Streaming, bounded batch-to-trade adaptation (_CandleScanBridge, unchanged from prior review).
  7. One terminal-operated path traversed source -> canonical publication -> catalog -> Application -> DataGateway -> D03 candles: proved by the verbatim transcript above, against the already-published canonical partition.
  8. No new architectural bypass or test-only production capability.
  9. Golden assertions are deterministic and PASS against the fixture-authoritative expectations on the exact recorded SHA -- see transcript.
  10. N/A this run (no mismatch).
  11. Focused harness/Application tests pass -- 866/866, see above.
  12. Operator verification exit code -- 0 on this exact PASS.
  13. No unrelated governance/runtime cleanup mixed into this change.