Skip to content

Preregister SPY/LEAN parity mapping manifest v2 (not run; gates unchanged) - #90

Merged
seathatflowsinourveins merged 2 commits into
mainfrom
claude/spy-parity-manifest-v2-20260922
Sep 23, 2026
Merged

seathatflowsinourveins merged 2 commits into
mainfrom
claude/spy-parity-manifest-v2-20260922

Conversation

@seathatflowsinourveins

Copy link
Copy Markdown
Collaborator

Summary

  • Pre-registers SPY/LEAN parity mapping manifest v2. It is frozen before any v2 run and has not been run. Files:
    • spy-parity/mapping-manifest-v2.json, sha256 1b821d7b…
    • spy-parity/PREREGISTRATION-v2.md, sha256 4393a1b7…
    • retained probes under spy-parity/probes/v2/ (synthetic engine-semantics data, each with its observed output and hash)
  • It closes v1's two unsupported mappings with engine-native NautilusTrader 2.0.0rc5 mechanisms:
    • Dividend cash: a SimulationModule whose process() returns Money, applied natively through try_adjust_account as a reported AccountState. The amount is the frozen factor file × the engine-observed position, on LEAN's next-trading-day ex-date. It is not a balancing entry.
    • Market-on-open: rc5 rejects AtTheOpen. The fill instead comes from native OCO-linked STOP_MARKET/MARKET_IF_TOUCHED orders at the decision bar's close ± 0.0001, which trigger at the engine's own open tick. For both oracle events (entry 323.58, exit 291.69) the prediction is an exact open fill. It is not a general MOO equivalent when the open equals the prior close (stated as a limit).
  • Tolerances are unchanged (tolerances.json sha256 c8bc7231…). v1 files and the v1 verdict are unchanged.
  • Three adversarial reviews covered look-ahead, fabrication and oracle equivalence; none refuted the preregistration.
  • The spy-lean-parity and dividend-sim-module gate notes record the preregistration. Statuses are unchanged.
  • A separate session implements exactly this manifest against this commit and runs it; this session reviews that verdict before any flip.
  • Tests: test_every_referenced_probe_artifact_hashes_as_recorded is now scoped to v1's top-level files, and a new test_v2_preregistration_probe_artifacts_hash_as_recorded checks every v2 probe hash.

Evidence

  • python3 -m unittest: 2093 OK (324 skipped).
  • All validate.yml steps and trading_gates.py --check rc=0.
  • Guarded gitleaks ran before the commit: no leaks.
  • 22 new files are pinned in manifests/evidence.json.

🤖 Generated with Claude Code

Closes the two unsupported v1 mappings with engine-native NautilusTrader
2.0.0rc5 mechanisms, frozen before any v2 run: dividend cash through a
SimulationModule whose process() returns Money (frozen factors x
engine-observed position, LEAN next-trading-day ex-date), and the
market-on-open fill through native OCO-linked STOP_MARKET/MARKET_IF_TOUCHED
orders triggered at the engine's own open tick. Tolerances are unchanged.
Retained probes with observed outputs and hashes are under probes/v2
(synthetic engine-semantics data). Three adversarial reviews (look-ahead,
fabrication, oracle equivalence) did not refute it; the proxy is not a
general MOO equivalent when the open equals the prior close. Gate statuses
unchanged; the run is done by a separate session against this commit.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
…s data

The blind-checkout classification test (from main) requires every value
under a label-like key to be classified; mapping-manifest-v2.json
mappings[0].decision is a methodology statement that must survive a blind
checkout, so its sha256 joins DATA_VALUE_SHA256.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
@seathatflowsinourveins
seathatflowsinourveins merged commit 12059ba into main Sep 23, 2026
7 checks passed
@seathatflowsinourveins
seathatflowsinourveins deleted the claude/spy-parity-manifest-v2-20260922 branch September 25, 2026 18:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant