GBP/USD H4 KAMA(10,2,30) pullback/recovery strategy — fixed retrospective 2024 replay

Python 3 standard library; system Europe/Berlin timezone database required.
  python3 reproduce.py
  python3 verify_replay.py

The first command reconstructs H4 and every reference-price trade from the
included unchanged GBPUSD minute records. The second uses a separate minute-
stream state machine with Decimal cost arithmetic and independently checks
aggregation, KAMA, CET/CEST conversion, prefix invariance and exit boundaries.
The downloadable CSV outputs were tested byte-for-byte after offline replay.

Read protocol.json for the complete frozen specification. Its hash was created
before calculating trade results. This is fixed retrospective research; the
chronological halves are descriptive, not genuinely unseen validation.
All settings remain unchanged after seeing the negative result.

397 original daily Forexite MetaStock archives, Dec 2023-Dec 2024. Original
ZIP files are not bundled. Their public URLs and original SHA-256 hashes are
in research/sources.json. Hashes and CRC were checked before extraction.
The included GBPUSD rows preserve original four-decimal OHLC and source times.
The original extraction includes warm-up and source boundary observations;
the H4 builder omits cut boundary groups and stops at 2024-12-31 20:00 UTC.
No missing prices or weekend bars are filled. H4 record timestamps use
(end-4h,end], not a claim to reproduce any broker's native H4 candles.

These are reference-price minute OHLC simulations, not historical broker fills.
No verified Bid/Ask or tick path is reconstructed. Stop-first handles both-level
minute ambiguity; a stop gap uses open and a target gap receives target only.
Costs are illustrative total round-trip plus calendar UTC-boundary allowances:
zero 0+0; base 1.5+0.2; stress 3.0+0.5 pips. Not measured spread or broker swaps.
R is initial gross stop distance; drawdown is closed-trade net R only.

research/v208-live-audit.json records that the preceding site upload was verified
before this strategy was prepared; replay.py requires its passing status.
The data and code contain no credentials, private account or trade information.
