[jadenfix]

What changed

• replace branchy long/short bookkeeping with signed position deltas
• preserve average cost on partial closes, realize P&L for partial/full closes, and price only a reversal residual at the new fill
• reject invalid orders/fills before they can mutate cash, commission, fills, or positions
• add direct transition regressions for long adds, partial/full closes, short-to-long reversal, and rejection immutability
• publish a source-pinned GIF/MP4 generated from the actual native accounting replay, with SHA-256 receipts and a full media verifier

Verification

• python3 scripts/accounting-e2e.py — strict -Werror and UBSan executables pass
• python3 scripts/backtester-e2e.py — exactly 68 strategy/dataset configurations complete on 11 generated 64-row OHLCV files
• python3 scripts/bridge-e2e.py — trained BayesianRidge round trips through the C++/Python bridge; concurrent calls differ and missing models reject
• python3 scripts/verify-accounting-demo.py — pinned source blobs, replay receipt, artifact hashes, H.264 MP4, and all 72 GIF frames verify/decode

Recorded source: d4b0acc6cde5475f532c76a7f310e9f2a357d7cb (tree f73563a87d30a16632156985ad8730073632c59f).

Scope

This is deterministic simulated accounting over fixture fills. It does not place live orders and makes no profitability, execution-quality, or latency claim.

[jadenfix]

Independent completion receipt — final exact head 880bbba9b3c6b3d6253851331855fec61dc1a14b (tree 68cbfc25f796ad890979740cadd688d424f52de1; base 8a2331f03203b01c615d8e13952cd44cfb28ae96).

Verified locally and again from a clean exact-head checkout:

• strict -Wall -Wextra -Wpedantic -Werror transition executable: pass
• separately compiled UBSan transition executable: pass
• direct regressions: add/increase, partial close, full close, short-to-long reversal, and rejected order/fill immutability
• trained BayesianRidge C++/Python bridge, concurrent prediction isolation, and missing-model rejection: pass
• real executable smoke: exactly 68 strategy/dataset configurations on 11 generated 64-row OHLCV fixtures
• source-pinned replay receipt and full media verifier: pass
• clean clone had no LFS checkout warning or status drift; docs/demo/accounting-receipt.csv resolves to ordinary text and matches blob 7dc7650c05c0c8158937826837a0c3e039153232, while historical dataset CSVs remain LFS-managed
• all 13 tracked __pycache__/*.pyc artifacts were removed; narrow __pycache__/ and *.py[cod] ignores were added
• the full hosted test sequence now ends with an explicit clean-worktree assertion

Hosted run 38366346396 passed both Ubuntu and macOS lanes, including bridge, strict/UBSan accounting, exact 68-configuration smoke, provenance/full media decode, and the new post-test clean-checkout gate.

Captured runtime source remains commit d4b0acc6cde5475f532c76a7f310e9f2a357d7cb (tree f73563a87d30a16632156985ad8730073632c59f); the packaging/hygiene-only follow-ups did not change capture inputs or media:

• H.264 MP4: 1280×720, 12.0 s, SHA-256 cda6a3044aec91f8c636fab57c579262b1e54a9a278d9dcf46b57fc68ed0a7f4, Git blob 6810714066596ba1383f41df3282d1909fdb8c45
• GIF: 960×540, 72 fully decoded frames, SHA-256 06d606fa11fb74064dc40cf9d4bbe106fb0eda3053b85a80f018eae4a90f3995, Git blob 8f22b1d72c3fbd6fdd39d3f9b5f1fb85542a1d64

Scope remains deterministic simulated accounting over fixture fills. No live orders, profitability, execution-quality, latency, or production-readiness claim. Ready for the designated portfolio merger; I did not merge.

[jadenfix]

Independent final-head verification (session3eef root): fetched exact head 880bbba9b3c6b3d6253851331855fec61dc1a14b, tree 68cbfc25f796ad890979740cadd688d424f52de1, into a fresh detached worktree. Confirmed no tracked __pycache__/.pyc remains; the small demo receipt is explicitly outside Git LFS while dataset CSV rules remain intact. Reran make accounting-test PYTHON=python (strict -Werror and separate UBSan transition executables), python scripts/bridge-e2e.py (persisted BayesianRidge round trips, concurrent isolation, missing-model rejection), and python scripts/verify-accounting-demo.py (recorded source/blob provenance, replay receipt, artifact hashes, full GIF/MP4 decode). The checkout remained clean after tests. I also independently ran the exact 68 strategy/dataset bounded backtest on the preceding content-identical runtime head; final hosted run 38366346396 reran that gate plus all other checks on Ubuntu and macOS, including a clean-tree assertion, and passed. Media remains deterministic simulated fixture accounting only—no live brokerage/exchange, profitability, latency, research-performance, or production-readiness claim. No merge performed; session100e79 remains sole merger.