trading: R1 deviation D6 binds the critic's overnight-volume read and its reviewed class - #796

Scope

• What this PR changes. It adds R1 deviation D6 (deviations-20261006-4.json) and its class-review receipt. D6 binds the completeness critic's missed read: the 2026-09-28/29 overnight-volume watch. It also binds the reviewed class of in-window observation receipts.
• Base commit: 0d5e65064. The bound records are read at R1's base commit ecfa11276.
• Lane: lane:trading.
• Owned paths touched:
  • blueprints/us-equities/rnd-wave1-20261006/deviations-20261006-4.json (new);
  • blueprints/us-equities/rnd-wave1-20261006/d6-class-review-20261006.json (new);
  • manifests/evidence.json (two rows, in the last commit).

What D6 binds:

• 100 records reviewed at the base commit: 48 COUNT and 52 OUT by market-data verdict, each with its reason.
• 57 entries in total (49 out_items; 3 OUT-verdict reads bound anyway, D6.3):
  • the critic's read (D6.1);
  • 53 further entries (D6.2). Records with disjoint recorded windows give one entry per window;
  • three in-window reads of non-market data, bound under PREREGISTRATION.md:40-42 and H1.1 (:57-58), which allow no fixture exception (D6.3): SEC filing metadata; issuer and exchange documents; one real 8-K filing kept as a test fixture.

What D6 does not do:

• It declares no eligibility (D6.6). R1's own attempt 1 already reported zero eligible sessions in every v1 segment from the known reads alone, and D1's prospective-holdout clause governs what follows.
• It changes nothing frozen. PREREGISTRATION.md, plan.json and the three earlier deviation files stay byte-identical, with their sha256 in kept_sha256.

Judgment calls (D6.4) are resolved conservatively and proposed for the command center's confirmation at its ACK. Binding more can only mark more cases inspected.

Rework after the command center's adjudication (both P1s upheld; head 773c342):

• P1-1: D6-O-N03 is bound (event, data_read, 2021-03-29).
• P1-2: an endpoint the record does not determine stays unknown (frozen D1, deviations.json rule.unknown); D6-C02 keeps ["unknown", "2026-09-29"], and D6.4 no longer substitutes the window endpoint.
• P2: the class review records the exact-equality comparison with the critic's proposal (path, file sha256, three span line ranges and hashes: all equal).

Round 3 (head b253161; the command center upheld the co-op delta read's P1 and both P2s, and corrected its own earlier narrowing option):

• P1: D6-C10 is an ADDITIONAL binding over 2021-01-04..2026-09-25 with the stronger access kind outcome_selected. It overlaps known_reads[1]'s unchanged D4.1 fixture (2021-01-04..2025-12-31, result_inspected); the overlap is stated in C10, in D6.4 and in the receipt's double-count explanation.
• P2: the old critic limitation stays only as an explicitly dated snapshot, superseded by the closing comparison. C02's discretionary narrowing sentence is gone from both records. The market-data OUT classification is separated from the required inspection bindings, and the fixture exclusion covers synthetic content only.
• Kept: C02's unknown start and all three D6.3 entries. Still 57 entries and 49 out_items; the zero eligible-session totals are unchanged.
• P3: the OUT lists are separated below.

OUT items, by reason

Covered by a known read (R1.known_reads, D4.1 fixtures) (15):

• D6-O-K01: blueprints/us-equities/broad-universe/README.md. Covered by known_reads[0] (broad-universe reserved segment evaluation) and its D4.1 fixtures; residuals are D6-C05, D6-C06 and D6-C07.
• D6-O-K02: blueprints/us-equities/mover-early-entry/README.md. Covered by known_reads[1] (mover-early-entry validation read) and its D4.1 fixture 2021-01-04..2025-12-31; the holdout-window collection is D6-C10.
• D6-O-K03: blueprints/us-equities/extreme-gainer-audit/README.md. Covered by known_reads[2] (extreme-gainer price audit, 957 events; D4.1 event fixture 2021-01-06..2026-09-22).
• D6-O-K04: blueprints/us-equities/mover-coverage-asof/evidence/summary-asof-20260929.json. Covered by known_reads[3] (E2, 715 events 2021-01-04..2025-12-31). E2 also read 979 lookback rows before 2021-01-04 (outside the window). The positive
• D6-O-K05: blueprints/us-equities/research-evaluation/README.md. Covered by known_reads[4]: the inspected 2021Q1 ETF control (D4.1 fixture 2021-01-01..2021-03-31). Earlier data (2014-2020) is outside the window.
• D6-O-K06: blueprints/us-equities/simulation-research/README.md. Same 2021Q1 ETF control as D6-O-K05; covered by known_reads[4].
• D6-O-K07: blueprints/us-equities/overfitting-controls/README.md. Reads the research-evaluation ledger: development data before 2021 is outside the window and the reserved 2021Q1 segment is excluded; nothing is left
• D6-O-K08: evidence/artifacts/gap-wave2-20260923/us-equities__portfolio-risk/README.md. Re-runs of the 2021Q1 ETF control episodes (dates 2021-01-04..2021-03-31 in raw/11 fills); covered by known_reads[4].
• D6-O-K09: evidence/artifacts/gap-wave2-20260923/us-equities__research-factors-ml/README.md. Frozen LEAN ETF inputs 2014-01-02..2021-03-31 (raw/r4/r4__frozen_ohlcv.json); the in-window part is the 2021Q1 ETF control, covered by known_reads[4].
• D6-O-K10: evidence/artifacts/gap-wave2-20260923/us-equities__evaluation-experiments/README.md. LEAN ETF data with in-window dates 2021-01-04..2021-03-31 only; covered by known_reads[4].
• D6-O-K11: blueprints/us-equities/identity-readiness/README.md. The 2022 identity sample named in known_reads[4]; covered by name.
• D6-O-K12: blueprints/us-equities/security-identity/README.md. Same 2022 identity sample as D6-O-K11; covered by known_reads[4].
• D6-O-K13: evidence/artifacts/gap-wave2-20260923/us-equities__identity-provenance/README.md. Re-run of the same 2022 identity sample (raw/dvc ledgers, 2022-06-08..2022-06-10); covered by known_reads[4].
• D6-O-K14: blueprints/us-equities/research-efficiency/README.md. Model tasks over existing records of the 2022 identity sample; no new data read.
• D6-O-K15: blueprints/us-equities/catalyst-experiment/protocol.json. Source of known_reads[4]; the blueprint's own receipt is a synthetic contract exercise (catalyst-experiment/receipt.json).

Outside the window, synthetic, not run, or a local re-read inside a bound item (21):

• D6-O-W01: blueprints/us-equities/authenticated-data/README.md. Dated outside the window (2020).
• D6-O-W02: blueprints/us-equities/engine-nautilus/equity-replay/README.md. Replays the 2020 bars of D6-O-W01: outside the window.
• D6-O-W03: blueprints/us-equities/corporate-action-readiness/README.md. LEAN bundled data dated 2020: outside the window.
• D6-O-W04: blueprints/us-equities/engine/README.md. LEAN bundled 2013 sample data: outside the window.
• D6-O-W05: blueprints/us-equities/historical-simulation/README.md. LEAN bundled 2019-2020 data: outside the window.
• D6-O-W06: blueprints/us-equities/engine-trials/spy-one-zero-20260926/lumibot/probes/hook_probe.py. Synthetic only: the 2021 dates in the probes are made up.
• D6-O-W07: blueprints/us-equities/alpaca-history-fixture/README.md. Not run: no observation.
• D6-O-W08: blueprints/us-equities/data-lake/README.md. Not run: no observation.
• D6-O-W09: blueprints/us-equities/data-lake/README.md. Local re-read of the broad-universe daily dataset: same scope as D6-C05, adds no case.
• D6-O-W10: blueprints/us-equities/sim-engine-crosscheck/README.md. Local re-read of D6-S01's retained cache with a narrower scope; adds no case.
• D6-O-W11: blueprints/us-equities/sim-crosscheck-hftbacktest/README.md. Synthetic only.
• D6-O-W12: blueprints/us-equities/nanosecond-replay/README.md. Synthetic only.
• D6-O-W13: blueprints/us-equities/mover-v3/README.md. Plan and protocol only; data/ holds calendars and fee schedules, not observations.
• D6-O-W14: blueprints/us-equities/adaptive-paper/evidence/alpaca-account-schema-20260923.json. No market data observed: account field names only.
• D6-O-W15: blueprints/us-equities/broad-universe/README.md. No market data observed: account state read only.
• D6-O-W16: blueprints/us-equities/adaptive-paper/native-faults/README.md. Quote and broker cases dated 2026-10-01: outside the window.
• D6-O-W17: blueprints/us-equities/adaptive-paper/receipts/xnys-primary-20261005/qualification.json. Offline fixture qualification dated 2026-10-05; no market data.
• D6-O-W18: blueprints/us-equities/engine-nautilus/ibkr-acceptance/evidence/ibapi-readonly-20261005.json. Bars dated 2026-10-02..05: outside the window.
• D6-O-W19: blueprints/us-equities/engine-nautilus/ibkr-paper-orders/evidence/receipt-20261005-passed.json. Quotes and orders dated 2026-10-05: outside the window.
• D6-O-W20: catalogs/us-equities/research-20260928-movers.md. Research synthesis of documents; its one stated observed move (line 105) is the positive-control selection bound in D6-C09 and lies inside D6-T15's sc
• D6-O-W21: evidence/artifacts/gap-wave2-20260923/us-equities__market-data-reference/5-parameterized-comparator.json. Local comparison of two retained datasets: the vendor daily parquet of D6-C05 and the LEAN bundled bars of D6-G02. Its securities and dates lie inside

No market data and not bound (dated outside the window, or no payload) (4):

• D6-O-N04: blueprints/us-equities/financial-data/README.md. No data observed: HTTP 403, no payload.
• D6-O-N05: blueprints/us-equities/catalyst-dataset/README.md. Dated outside the window (2020) and no market data.
• D6-O-N06: blueprints/us-equities/catalyst-provenance/README.md. SEC index acquisitions (2026-09-24 new host, 2026-10-04 pin move) fetch the 2020-03-02 daily index: dated outside the window, no market data.
• D6-O-N07: blueprints/us-equities/delisting-coverage/README.md. Dated outside the window: Form 25 filings 2016-2019 and bar checks ending by about February 2020; current ticker mappings are not market data.

Wrapper records of another record (9):

• D6-O-X01: evidence/receipts/broad-universe-research-20260921.json. Same observation as D6-C05, D6-C06 and known_reads[0]: a wrapper or summary record, not a separate read.
• D6-O-X02: evidence/receipts/alpaca-paper-e2e-20260921.json. Same observation as D6-P02: a wrapper or summary record, not a separate read.
• D6-O-X03: evidence/receipts/adaptive-paper-practice-20260921.json. Same observation as D6-C14: a wrapper or summary record, not a separate read.
• D6-O-X04: evidence/receipts/ibkr-readonly-acceptance-20260923.json. Same observation as D6-I01 and D6-I02: a wrapper or summary record, not a separate read.
• D6-O-X05: evidence/receipts/ibkr-readonly-acceptance-20261005.json. Same observation as D6-I03 and D6-O-W18: a wrapper or summary record, not a separate read.
• D6-O-X06: evidence/receipts/pit-availability.json. Same observation as D6-C11, D6-C12 and D6-O-N01: a wrapper or summary record, not a separate read.
• D6-O-X07: evidence/hosts/macos-m5pro-20260924/macos-m5pro-20260924--alpaca-py--use--20260924.json. Same observation as D6-T10 and D6-T11: a wrapper or summary record, not a separate read.
• D6-O-X08: evidence/hosts/macos-m5pro-20260924/macos-m5pro-20260924--nautilus-trader--use--20260924.json. Same observation as D6-T10 and D6-T11: a wrapper or summary record, not a separate read.
• D6-O-X09: catalogs/us-equities/paper-practice-20260921.md. Same observation as D6-P02 (readiness read): a wrapper or summary record, not a separate read.

Bound in D6.3 although their market-data verdict is OUT (3): D6-O-N01, D6-O-N02, D6-O-N03

SOTA sources

• The frozen protocol this deviation follows: blueprints/us-equities/rnd-wave1-20261006/PREREGISTRATION.md at frozen commit d4601bfcb:
  • R1 Method steps 1-3 (:40-52), including the conservative rule;
  • H1.1 (:57-58): a miss found after the freeze is added as a dated deviation, never silently;
  • D1 (:63-66).
• R1's entry fields come from plan.json (R1.entry_fields, R1.conservative_rule). The scope-kind spellings come from D4.1 in deviations-20261006-2.json.
• Deviation practice already used in this repository:
  • deviations.json, -2 and -3 (landed in b0df1fb04, trading: preregister north-star R&D wave 1 (inspection registry, E2 replay on NativeStack2604) #778): dated files beside byte-identical frozen files, each with kept_sha256;
  • extreme-gainer-audit/plan.json with its deviations.json.
• Hash verification uses git's own object store: git show ecfa11276:<path> for every bound file and line span. The span convention is lines joined with \n plus a trailing \n, the same bytes as sed -n 'a,bp'.

Evidence-class table

Local commands run

$ python3 scripts/validate.py
{"components": 69, "hashed_files": 10338, "profiles": 4, "receipts": 214, "status": "passed"}  (exit 0)
$ python3 scripts/evidence_manifest.py --check
{"files": 10338, "status": "passed"}

Decision record

D6 itself is the dated record (recorded_at_utc inside). The gate is the command center's item task-native-agent-stack-5f-20261006T143150Z: a GPT micro-check, then the command center's ACK, then R1's rerun loads this file.

🤖 Generated with Claude Code