Retain rc5 host reproduction and accepted R2a replay evidence - #760

Scope

Retain two separate native research results: rc5 SPY host reproduction with the full qualification failure, and accepted offline R2a classification totals/provenance folded from #789. All original result bytes remain unchanged; no new model, simulation or classification run is introduced.

The unchanged rc5 SPY one_zero case reproduces economic parity on NativeStack2604. This run is host reproduction, not qualifying (5f, 2026-10-06T05:36Z): replay exit0, full comparator exit1,135PASS/1FAIL. Both attempts passed106/106execution checks, with byte-identical native stdout and end cash90734.08USD equal to the retained reference. Both AAPL cases remain blocked by absent retained input; five other SPY cases have no mapping.

5f declined waiver of#35. The existing rc5 qualification is bwrap-cc4503f-qualifying/receipt-v2.json; deviation-acceptance-v2.json admits only the first replay from cc4503f, and replay-history-v2.json explicitly prevents a retry from qualifying. The full FAIL remains unchanged. The gate-owner's disposition is committed BYTE-IDENTICALLY, SHA256a900056304f0ae536457eae05a98a771f3860d6352d2aec76626fcb755b5df08, with provenance as an owner decision, not a user signature.

Byte identity is bounded: all four normalized economic records and account CSVs match; the two native stdout captures also match. Raw fills/positions reports, report JSON and engine logs differ. The owner record remains verbatim; curated reporting does not extend its phrase to every raw output.

• Base commit: ecfa112 (trading: NativeStack2604 research runtime, EdgarTools 5.60.0 (runtime and catalog), hash-pinned build backend #703).
• Lane: lane:trading.
• Owned evidence path: blueprints/us-equities/engine-nautilus/evidence/j2-sim-rc5-20261005/. This follow-up adds the exact owner record and labels the curated launcher receipt as directed; original native outputs remain byte-identical.
• Prior native24/24 runtime acceptance remains retained. No harness, launcher, mapping, history, tolerance, preregistration or catalog change.
• Trading owner5f supplies the cross-family read and owns landing. This stays draft.

Retained R2a receipt folded from #789

This draft also retains the accepted offline R2a totals/provenance receipt, byte-identically from #789 at 4650b6c. Both unchanged classify/verify arms exited 0; their three classification output hashes match, with 715 package events and N2=594. The verify hashes differ only by HEAD provenance. The receipt preserves the first outer-tool exit 1/native exit 0 discrepancy and D5.1 acquisition limits. This adds no holdout, predictive, strategy, broker or paper evidence, and reruns neither research arm nor simulation.

• Receipt: blueprints/us-equities/mover-coverage-asof/evidence/native-e2-replay-2604-20261006.json, SHA256 de1d09f2b8d9479d9ace878b169d8688758bb81f110e208846822bb8bbdc2b03.
• Shared file: the native register_file API adds this one receipt entry to manifests/evidence.json; all retained J2 records and existing registry entries stay unchanged.
• Folded original: evidence(trading): retain accepted E2 replay totals and provenance #789. The co-op verifies content and closes its original; its branch stays.

Landing path

Trading owner 5f supplies the cross-family read; the CC ACK names the exact checked head; 5f lands this retained draft at the CC-approved rank 31. Rebase only on 5f's relayed landing cue. No new PR or additional acceptance campaign.

SOTA sources

• Frozen R2a Wave1 plan and D2/D3 deviations.

• Unchanged Arm A refetch.py and Arm B refetch.py, with acquisition/interpretation limits retained in the receipt.

• nautechsystems/nautilus_trader v2.0.0rc5, resolved1b0a49d2792a9432a3aca3fcb617ce7a630d905e; unchanged quickstart.

• Harness atecfa11276: spy-parity/run.py, compare.py:814–817 and README.md:82–102.

• Runtime acceptance isolation, including read-only managed.python beside.venv.

• Sealed SPY preregistration v2, mapping-manifest-v2.json, replay-history-v2.json and acceptance-plan.md:64–74 at the same pin.

• Owner-staged retained LEAN985ef30ad3ac774218c5ac516b4cb0aa2655730f inputs; all5frozen SHA256 values verified before execution. No acquisition.

Evidence-class table

Local commands run

Fold validation at 3f7a8cd:

rtk proxy nice -n 19 ionice -c3 python3 scripts/validate.py
# 0; 10,275 hashed files; integrity/scope only.
rtk git diff --check
# 0.

The R2a artifact matches its source SHA256 byte-for-byte. Its registry entry was produced with the repository's native register_file API in the last commit; 10,274 prior entries and all 13 J2 artifacts are unchanged. These checks rerun no harness, simulation, model or upstream test suite. Hosted CI at this head is pending.

Retained original execution records:

rtk proxy sha256sum -c MANIFEST.sha256
# 0; all five frozen input files OK.

The exact replay/comparison vectors with named private-root substitutions and original argv hashes are in the launcher-equivalent receipt. Both use nice19, bwrap --unshare-all/--clearenv, read-only.venv/.python/repository/data and one owned writable output. The final environment uses the five documented harness values; no host home, credentials, broker socket or provider config is mounted.

Replay04:41:27–04:41:30Z:0. Comparison04:41:47Z:1, complete=true, no skipped checks. Native end cash90734.08USD includes428.64USD distributions. The comparator's sole failing key is precondition_review.no_prior_replay_ran_the_reviewed_harness#35. Two recorded prior replays share the reviewed hashes; only one qualified. Existing deviation acceptance does not waive this condition.

Initial replay04:40:32–04:40:35Z:0; comparison04:40:52–04:40:53Z:1. Its five additional benign runtime-acceptance environment names failed the exact environment check. A fresh output corrects that route mismatch; no predicate/history/harness edit masks either failure.

rtk proxy nice -n 19 python3 scripts/validate.py
rtk git diff --check

Both0. New public native receipt/verdict/stdout files match the returned bytes; full private outputs and both original argv vectors are retained in the owned ignored runtime directory. No new custom test runner or mapping was introduced.

Decision record

No new architecture selection or qualification. The trading owner's approved launcher-equivalent route, exact declined-waiver record and full native FAIL are retained. This is host reproduction, not qualifying (5f, 2026-10-06T05:36Z). Committed history remains an unchanged source snapshot and excludes these two new attempts, separately retained under the receipts-only scope; it is not an exhaustive updated history.

Host evidence

No evidence/hosts/ file or platform/gate status is changed. Native historical replay, pinned local harness checks, prior synthetic quickstart and prior runtime acceptance remain distinct. AAPL source absence and launcher limitations stay recorded; no fresh provider fetch, substitution, actual order or broker connection.

Checklist

• Exact gate-owner record added; curated launcher receipt explicitly labelled; harness/mapping/predicates and original native outputs unchanged.
• No secrets or personal paths in committed evidence.
• No hosting, billing, merge, runtime reinstall or configuration change.
• J4's rc6 runtime/worktree and peer-owned files preserved.
• New owner record and labelled curated receipt registered in the LAST commit under the hot-file protocol.
• Trading owner's cross-family read before ready and owner-controlled landing.