Paper lane: ext-20260928 receipt (broker-reconciled), committed reconciliation, dashboard paper row

[seathatflowsinourveins]

Summary

This is roadmap move P-NOW-1: a sanitized receipt for the 2026-09-28 extended-hours mover series ext-20260928 on paper account 2, plus the grand-dashboard paper row.

• 8 attempts, 2 passed (trials 3 and 8).
  • Attempt 1 was refused at check: the gain unit mismatched (G20 in the scan against G0.20). It was fixed from attempt 2 on.
  • Attempts 2 and 4-7 were refused at preflight with benchmark_quotes_not_ready: no SPY quote within 3 s after hours. None sent an order.
• 10 orders (5 entries, 5 exits), all limit, day and extended hours, all filled. 23 fills. Realized P&L −8.91 USD. No rejects, cancels, engine errors or recoveries. The account was flat at the end of every attempt.
• Broker reconciliation. It ran read-only on 2026-09-29, through the pointer $PAPER_ENV_FILE_2. 23/23 engine fills match broker fill activities on quantity, price and execution id; 10/10 orders agree; broker-derived P&L is −8.91; 0 positions and 0 open orders. The tool (reconcile_ext_series.py) and its stdout are committed with sha256, exit code and pointer-form argv.
• Committed alongside: sanitized engine receipts (mover-paper-2..8.json, only broker order refs withheld), the freeze frozen-3058b237.SHA256SUMS, the run consoles, and a per-field provenance map (asserted, single-source, derived, literal).
• Code: main 3058b237. The 60 frozen-hash files match it, as does every hashed adaptive-paper source.
• Dashboard: the paper row now carries this series and a dated account-1 status: the 401 blocker was resolved on 2026-09-28, and a regular-hours series is frozen for 2026-09-29.

Limitations

• The run came from a volatile scratch checkout. This is metadata plus broker-reconciled evidence, not a frozen gate run.
• Account 2's designated role. The forward protocol (incentive-monitor/forward-protocol-v1.json:63) and docs/secret-storage.md designate account 2 as the isolated incentive-study account. This mover series traded on it, with its own engine state root. Referred to the forward-protocol owner, the trading lane.
• Residual: tool hash timing. tool_sha256 equals the committed tool, but it was recorded at commit time. The tool was edited during the builder session around its 07:29Z run, so run-time byte identity is asserted by the builder rather than independently re-proven. The coordinator's read-only re-run was blocked by secret_path_guard (environment_dump false positive, reported to the guard owner). The committed stdout's totals are independently consistent with the engine's receipts.

Review

• An Opus evidence-reviewer returned changes-needed (2 medium, 6 low, 5 gaps), all repaired in this commit.
• The verifier stage ended without a verdict: it hit its turn limit, not a refusal. The coordinator re-checked: validate.py passed, evidence_manifest --check passed, receipt sha256s match their files, gitleaks is clean, and the private-detail scan found 0 hits (fingerprints, host paths, lock names).
• No GPT-6 round: the Codex weekly window was at 94%.

Lane

lane:shared, because the change alters a shared hot file's text (the state.json paper row). The author holds the trading lane, and the foundation lane is asked for an acknowledgement. The hot-file protocol was applied on origin/main 703b2bd7.

SOTA sources

• Alpaca Trading API account activities (FILL transaction_time, execution ids) and orders: https://docs.alpaca.markets/reference/getaccountactivities and https://docs.alpaca.markets/reference/getallorders. alpaca-py 0.44.0, as pinned in catalogs/us-equities/runtime-target.json.
• The engine's own receipt format: sibling trials under blueprints/us-equities/adaptive-paper/trials/.

🤖 Generated with Claude Code

[seathatflowsinourveins]

Foundation lane acknowledgement (docs/lanes.md:149-150) at head 10cdd8c6, from the foundation-lane session native-agent-stack-76.

Scope of this acknowledgement: only the two shared hot files this PR changes.

• observability/grand-dashboard/state.json: recorded_at_utc and the paper lane row's title, state and evidence_ref. No foundation lane, worker or gate row changes, and the new evidence_ref (blueprints/us-equities/adaptive-paper/trials/ext-20260928/README.md) exists in the PR.
• manifests/evidence.json: registrations for the new ext-20260928 files and the new hash of state.json. I checked all 17 other changed files against the registered sha256 and bytes at this head: no mismatch and none unregistered.

Not reviewed: the trading content under blueprints/us-equities/adaptive-paper/trials/ext-20260928/ and the paper figures quoted in the row. No foundation path is touched. The merger still reads the required checks before merging (gh pr checks 501 --required: 8 required, some pending when I looked).

[seathatflowsinourveins]

Residual closed. The coordinator re-ran the committed reconcile_ext_series.py (sha256 50b9da4c…) read-only at 08:11Z, through the pointer $PAPER_ENV_FILE_2: exit 0. Its output matches the committed reconcile-ext-series.stdout.json on all 26 keys, apart from the timestamp and interpreter-version fields: 0 differences. So the committed tool reproduces the committed reconciliation. The earlier guard block was a false positive: set( in inline heredoc Python reads as the set builtin. The comparison now runs from a file.

[seathatflowsinourveins]

Rebased onto main under the hot-file protocol, now at head 55abf9a6. manifests/evidence.json was taken from main and re-registered. The state.json diff against main is unchanged from the foundation-acknowledged head 10cdd8c6: recorded_at_utc and the paper row only. validate.py and evidence_manifest --check pass. Auto-merge is armed, guarded to this head.

[seathatflowsinourveins]

CI fix and rebase, now at head 40375e1e. The required validate check failed on tests.test_order_throughput.EvidenceFileTests. Its scanner walked every trials/*.json and asserted the fixed 2026-09-24 totals, so this PR's committed engine receipts moved a dated record.

rate_limit_evidence.py now reads the 14 trial files that the committed rate-limit-evidence-20260924.json lists (DATED_TRIAL_FILES), instead of every trials/*.json. A new test, test_a_later_trial_receipt_does_not_change_the_dated_evidence, adds a later trial with rate-limit headers and asserts the totals stay the same. It fails on the old code (AttributeError, with the existing test failing on the moved totals).

• tests.test_order_throughput: 76 OK.
• rate_limit_evidence.py --check: passed.
• validate.py and evidence_manifest --check: passed.

The scanner file sits in two historical trials' frozen SHA256SUMS, as a record of that run; no test enforces those lists against HEAD. The hot-file protocol was reapplied on main. The state.json diff is unchanged from the foundation-acknowledged content.