evidence(trading): retain accepted E2 replay totals and provenance

[seathatflowsinourveins]

Scope

Publish the accepted NativeStack2604 E2 offline replay as a totals-only receipt. D2 reports REPRODUCED, exit0: four native commands exited0, ArmA matches the pinned hashes, and both arms' results/labels/summary are byte-identical. Counts are715events, N2=594, and507/261/97 at1.5x/2x/3x.

• Base commit:9f3af37dd4659d8686c454d8ac727a3de73452a4, current origin/main and descendant of c1300c1.
• Lane:lane:trading.
• Only new receipt and manifests/evidence.json; registry is the LAST commit.
• Placement follows the existing E2 convention mover-coverage-asof/evidence/summary-asof-20260929.json.
• 5f accepted the private outputs cross-family. This stays draft for the co-op's GPT first-pass read, CC ACK and5f-controlled landing.

SOTA sources

• Frozen R&D Wave1 plan atc5834a1ba, R2a; D2/D3 deviations.
• Unchanged ArmA refetch.py at99dfec7c, SHAfc6d5415550198601d1089f92d722026622e41ff2b12758e3a2c0b9e1f99d3e4.
• Unchanged ArmB refetch.py atecfa1127, SHAcdac99732694dca24467af3dfb4dc8dc8c80be485c8ef27c995de9c9f9f6f0b9.
• Pinned retained E2 summary, all11retained file hashes; package ZIPa0b0aabf32175f0759e5911a7258eb829b29d5b6c0a72314e086335b53a6c19c.
• A11 interpretation follows Arefetch:1181–1188 and Brefetch:1206–1213: Git provenance lives only in verify.json, outside the three classification outputs. D2verify proves same-arm repeatability.
• D5.1 supported fetch route is refs/pull/485/head; co-op reports GitHubhead28a611b396541af3c901bc2ec5f719b508e06765. This source-acquisition note is not a claim that the lane fetched that ref.

Evidence-class table

Claim	Evidence class	Receipt / retained observation
Four native classifier/verifier exits0 and exact reproduction	native_proven	new receipt's exits and8output hashes; original private stdout/meta retained
D3all11files+ZIP match; originals and each arm's7inputs preserved	local_integration	independent pre/post hash checks, source700
Python3.12.3/NumPy2.5.3, no-network isolated execution	native_proven + source_review	bwrapunshare-all/clearenv, RO runtime/source, one arm output mount; exact sanitized argv retained
A11 separates verify HEAD provenance from revision effect	source_review	bothclean HEADs; onlyverify.classify_code_revision.head differs
D5.1 code availability/ancestry with truthful transport scope	source_review + local_integration	pre-existing localobject; preclassifyGitblobSHA/exactcleanHEAD; local99df→28a ancestorcheck0
Public-content and repository integrity	local_integration	privacy scan0, full integrity validation0, diffcheck0

Observations explicitly preserved

A-classify's older meta format recorded EXIT_CODE=0 while the outer tool returned1. Its native returned hashes, new output files and successful same-copy verify agree with native0; D2governs by that native exit. The outer1 is retained separately. Later three commands already captured NATIVE_EXIT_CODE and SANITIZER_EXIT_CODE.

The private meta writer now snapshots the complete PIPESTATUS array immediately, suppresses ERR/errexit interference during capture and records both codes before returning. Trivial dry cases produced native/sanitizer0/0 and3/0, with wrapper0/3. This private logging change is not repository code or an experiment rerun; its sources are installed Bash5.3.9 bash(1) and the native-status-before-summary pattern in adoption-bootstrap.yml:998–1005 at the base.

ArmA's object was already available in the shared clone before this lane read it. The lane did NOT fetch99dfduringR2a; the original acquisition ref is not established from retained records. Preclassify source blob SHAfc6d5415 and exactcleanworktreeHEAD were checked. Co-op's D5.1 GitHubref-to28a report is cited separately; this lane independently verified local object ancestry0 and the current physical sourceSHA. No transport ref is invented, and no retry, changed input, harness change or new fetch of ArmA occurred.

The referenced private draft grew from the earlier2,746-byte review snapshot to3,724bytes with the requested provenance/older-meta observations, accepted-review state and plain wording that passes a stricter token-level scan. All native output hashes/counts remain unchanged. The published JSON has no home/tilde paths, dates or symbol-like tokens; private per-event output, historical timestamps and input paths are not included. Receipt SHA256: de1d09f2b8d9479d9ace878b169d8688758bb81f110e208846822bb8bbdc2b03.

Local commands run

# Each exact arm worktree, Python3.12.3/NumPy2.5.3; nice19/no network:
python blueprints/us-equities/mover-coverage-asof/refetch.py classify --out-dir <fresh-copy>/e2
python blueprints/us-equities/mover-coverage-asof/refetch.py classify --out-dir <fresh-copy>/e2 --verify
# Both arms0/0;4commands total, completed before10:30Z.

rtk proxy nice -n 19 python3 scripts/validate.py --scan-file blueprints/us-equities/mover-coverage-asof/evidence/native-e2-replay-2604-20261006.json
#0.
rtk proxy nice -n 19 python3 scripts/validate.py
#0; metadata integrity/scope only, no live model/GPU execution.
rtk git diff --check
#0.

The checks above used the repaired private writer; native/sanitizer0/0 retained for both scans. No full suite, build or simulation was repeated during the paper window.

Decision record

No new architecture or strategy selection. A11 interprets verification as repeatability; D5.1 acknowledges equivalent pinned-source availability. 5f's cross-family acceptance and the co-op's R2A-ACCEPTED directive govern this receipt.

Host evidence

No evidence/hosts entry, platform-status change, paper/broker acceptance, new holdout or predictive/strategy claim. R1/R2bholds remain independent. Public reviewer evidence contains only totals, boolean checks, native exit codes and hashes/provenance.

Checklist

• New totals-only receipt follows existing E2layout.
• No raw private events, credentials, home/tilde paths, dates or symbols in JSON.
• Both older-meta and fetch-ref observations explicitly disclosed.
• Native code/input/output unchanged; no new model/provider/broker execution.
• Registry in LAST commit; no merge/landing by this lane.
• Co-op GPT first-pass read and CC ACK before5f's landing queue.

[seathatflowsinourveins]

Closed with a record by the PR triage of 2026-10-07 (the command center's ruling). Not merged; the branch codex/ns2604-r2a-receipt-20261006 stays on origin at 4650b6c.

Folded into #760 at 3f7a8cd. Of this PR's 2 changed files, 1 is byte-identical there and 1 is merged; none is lost. Registry: 1 entries added or changed by this PR are all present there, 0 of them re-registered to the merged file's hash.

Reopen trigger: none while #760 carries the content. Reopen with gh pr reopen 789.