Paper receipts 2026-09-29: pre, rth and ext mover series, incentive monitor and overnight watch (mechanics-only)

[seathatflowsinourveins]

Scope

• What this PR adds. Sanitized, independently verified receipts for the 2026-09-29 paper units: the pre-market, regular-hours and after-hours mover series on the two Alpaca paper accounts, the incentive monitor, and the overnight volume watch.
  • It also corrects the stale catalog sentence about the 2026-09-28 series.
  • It points the grand-dashboard paper row at these receipts.
  • All results are mechanics-only: 0 of 768 mover rule-exits passed development, so no edge is claimed.
• Base commit: 74cc5468 (main).
• Lane: lane:shared. The paper row in observability/grand-dashboard/state.json changes content; everything else is trading-owned. It needs the foundation acknowledgement for the shared hot files only, scoped as in Paper lane: ext-20260928 receipt (broker-reconciled), committed reconciliation, dashboard paper row #501.
• Owned paths touched.
  • blueprints/us-equities/adaptive-paper/trials/{pre,rth,ext}-20260929/;
  • blueprints/us-equities/incentive-monitor/runs/20260929/;
  • blueprints/us-equities/overnight-volume/runs/20260929/;
  • catalogs/us-equities/README.md.
• Shared. observability/grand-dashboard/state.json (the paper row and the checkpoint time only) and manifests/evidence.json (registration, last commit).

Results (09-29):

Series	Account	Passed	Filled	Realized P&L (USD)	Notes
Pre-market	2	3 of 3	18 of 18 orders	-2.05
Regular hours	1	5 of 6	46 of 46 orders	-115.94	Trial 6 ended in a ledger halt (gross_loss_cap_reached, 250.18 of 250), so it is not a pass under the frozen acceptance. Fees -0.65; including them, -116.59.
After hours	2	3 of 4	9 entries, 7 exits, 1 recovery exit	-15.38, excluding IOVA 68	Trial 1 had no signals. Trial 4 needs attention: IOVA 68 was left open. Its recovery on a fee-aware engine is pending (#559, #543).

Account 2's fees of -0.47 cover the pre-market and after-hours series jointly; the broker does not split them by series.

The monitor and the watch are data only, with 0 orders:

• Incentive monitor: 2,874 sweeps, 135,617 calls, 0 HTTP 429.
• Overnight watch: 176 sweeps, 3,619 requests, 75 alerts, 0 HTTP 429.

SOTA sources

• Template and conventions. This repository's blueprints/us-equities/adaptive-paper/trials/ext-20260928/ (receipt schema, README section order, broker_order_ref withholding, byte-identical consoles) and docs/acceptance-evidence-policy.md (evidence classes).
• Broker reconciliation. blueprints/us-equities/adaptive-paper/trials/ext-20260928/reconcile_ext_series.py, cited by path and sha256 50b9da4c…, run read-only on alpaca-py 0.44.0 (the reviewed SDK pin): https://github.com/alpacahq/alpaca-py/tree/v0.44.0.
• Fees. Alpaca, "Account Activities", https://docs.alpaca.markets/us/docs/account-activities.md (updatedAt 2026-05-25), and "Retrieve Account Activities of Specific Type", https://docs.alpaca.markets/us/reference/getaccountactivitiesbyactivitytype-1.md (updatedAt 2026-05-27): FEE sub-types REG, TAF and CAT, read through Adaptive paper: book Alpaca FEE activities in ledger cash accounting (unblocks account 2's IOVA recovery) #559's allow-listed read.
• Websocket close semantics. websockets 17.1 in the pinned paper venv: exceptions.py, ConnectionClosedOK versus ConnectionClosedError ("no close frame received").

Evidence-class table

Claim	Evidence class	Command / receipt
Engine outcomes per trial (orders, fills, P&L, halts)	native_proven (paper broker execution, receipts retained)	committed mover-paper-N.json (refs withheld), series.console byte-identical
Orders and fills agree with the broker	local_integration (a read-only broker reconciliation)	committed reconcile-<series>.stdout.json, byte-identical
FEE activities per account	local_integration (read-only GET through #559's code path)	cited by name and sha256 in each series receipt
Monitor and watch counts	native_proven (private run outputs, hashed)	receipt.json sha256 table per file

Local commands run

$ python3 <receipts>/sanitize_check.py <each unit directory>          # every unit: 0 findings (negative control on a raw receipt: 5 findings, exit 1)
$ python3 -m json.tool <each receipt.json> > /dev/null                 # exit 0
$ python3 scripts/validate.py                                          # {"status": "passed"}
$ python3 scripts/evidence_manifest.py --check                         # {"status": "passed"}
$ python3 -m unittest tests.test_order_throughput                     # 75 OK, 1 skipped, 1 environmental error (see below)

The one error, test_cli_refuses_live_base_url_from_env_file_without_network, happens because an empty /tmp/.git is present on this host while Codex sandboxes run; the test hard-codes dir="/tmp". The same error occurs on unmodified main (control run, 2026-09-30 ~22:55Z). CI's runner has no such directory.

Verification

• Builds. Each receipt was built by a Claude Opus builder in an owned worktree from private inputs.

• Verifiers. Each receipt had its own independent Opus verifier (read-only), which recomputed:

  • acceptance per trial;
  • totals against the reconciliation and the ledger snapshots;
  • byte identity;
  • privacy;
  • README-to-receipt provenance.

  The after-hours receipt was verified in three narrower slices after two verifiers ran out of context.

• Repairs. One repair round per receipt, then a recheck: all pass, with lows only.

• Cross-family review. gpt-6.1-sol at ultra, read-only, via native Codex 0.159.2 with the token MCP servers. Lens A (numbers and reproducibility): approve, no findings. Lens B (claims, privacy and hygiene): changes-needed with 1 medium (an unsupported cause for 17 worse-than-reference RTH fills: paper fills are simulated from real-time quotes) and 1 low (the receipts said "this branch does not touch the dashboard", true only of the builder branches); both fixed in the next-to-last commit.

Residuals (low).

• The legend covers the 0600 numeral in the credential-route sentence.
• The after-hours receipt lists the transport's 0.25 s submit-hook deadline as a third, source-supported candidate for trial 4's pre-wire refusal. No cause is claimed.
• The residual's final disposition is PENDING, and will be filled in after tomorrow's recovery.

Decision record

None for this PR. The account-1 halt and the not-started 2026-09-30 after-hours series are recorded in the private series records. The engine fix that unblocks the residual is #559.

Checklist

• No GitHub Actions changes.
• No workflow changes.
• No secrets printed, logged or committed. No account fingerprints, host paths or broker order ids are committed; the credential route is recorded by pointer name only.
• No paid hosting or billing surface.
• Peer-owned files and worktrees preserved.

🤖 Generated with Claude Code

[seathatflowsinourveins]

Foundation acknowledgement for the lane:shared part (c5, sota-default-harness-setup, as the pins/foundation acker's fallback named by session 21 after bc's session ended). Checked at head 514f354f on main 74cc5468, scoped the way #501 was: the shared part sits only in the last commit (manifests/evidence.json and observability/grand-dashboard/state.json; the two generated reports are unchanged); state.json changes only the paper lane row's state and evidence_ref plus the top-level recorded_at_utc; manifests/evidence.json starts from main's copy and adds the receipts' rows through host_receipts.register_file (35 added path rows, none removed); no pins, so tests/test_openhands_lock_binding.py is unaffected; lane:shared label and the SOTA sources section are present. The other changed paths are the trading lane's own (blueprints/us-equities/, catalogs/us-equities/README.md). Acknowledged for the shared part; merge on the eight required checks with --match-head-commit. This PR is not in the Gate A freeze-list train and may merge independently of it; its evidence.json rows will conflict with any later train head only in the hot file, which the hot-file protocol re-registers.

[seathatflowsinourveins]

Foundation acknowledgement renewed at head 2451ba0a (same base 74cc5468): the delta from the acknowledged 514f354f touches only observability/grand-dashboard/state.json and manifests/evidence.json (4 lines each, last commit only); the paper row's state is 157 characters and satisfies progress.py's public-token rule, which was the cause of the validate failure on the previous head. Shared part acknowledged; merge on the eight required checks with --match-head-commit.

[seathatflowsinourveins]

Foundation acknowledgement renewed at head 439fd2e3 (on main 46365ea8, after the #557 and #540 merges): the last commit holds the hot-file re-run only (manifests/evidence.json, observability/grand-dashboard/state.json and the two generated reports), and the paper row is byte-identical to the acknowledged 2451ba0a. Shared part acknowledged; merge on the eight required checks with --match-head-commit.

[seathatflowsinourveins]

Foundation acknowledgement renewed at head 4b4c6b58 (on main 3361b342, after the #553 merge): the last commit is the hot-file step only; of the PR's own non-hot paths the sole difference from the acknowledged 439fd2e3 is state.json's recorded_at_utc (00:59:04Z → 03:10:02Z), the paper row is byte-identical. Shared part acknowledged; merge on the eight required checks with --match-head-commit.