USD/JPY H4: fresh DI cross and ADX14>=25 educational model
Actual unchanged two-decimal Forexite OHLC, January-February2024.
61,972 original M1 rows;61,792 replay rows through2024-02-29T20:00Z.
Winter CET minus1h to UTC. Observed H4 groups (end-4h,end], no fill.
Original daily-archive URLs and hashes retained. The60 daily archives
were not redownloaded. The13 stored source-kit hashes were verified.

Offline: Python3 replay.py then Python3 verify_replay.py (standard library).
Seven output files must reproduce byte-for-byte from the frozen protocol.
Separate Decimal70 indicator arithmetic and causal minute ledger verify
all decisions and20 trades. Optional verify_javascript.py requires Node.
make_charts.py requires Inkscape only for PNG export.

DI14: seed TR/+DM/-DM sums with transitions1..14 at index14. Then
S=S-S/14+new. DM winner must be positive and strictly exceed the other;
equal positive extensions produce both DM0. DI=100*DMsum/TRsum.
Zero TRsum gives DI0. DX=100*abs(+DI-(-DI))/(+DI+(-DI)); zero denom0.
ADX first at index27 =mean DX[14..27], then(previousADX*13+DX)/14.
ATR14 is mean TR[1..14] at14, then Wilder recurrence. No chart reseed.
The declared seed is explicit; exact parity with platform seeds is not claimed.

Shared start27. Long previous+DI<=previous-DI/current+DI>current-DI;
short reversed. Filter currentADX>=25 at the cross only, equality passes.
No ADX rising/fresh25cross requirement, no delayed queue, no EMA,
pullback or extreme-point breakout confirmation. One position per run.
Next observed minute Open strictly later within5min; reject busy signals.
Risk ceil(1.5*ATR pips), stop1R,target2R,six-observed-H4 cap.
Minute priority: adverse Open gap at Open; favorable gap at target;
due cap at Open; intraminute touches stop-first if both; final Close at
sample end if still open. Entry-minute touches count. See protocol.json.

Costs0/1.2/2.4 pips once per closed trade; UI0..5 step0.1 accounting only.
No Ask/spread/slippage/commission/financing or broker fill reconstruction.
Both runs lose at zero and positive transaction allowance. Smaller sample
loss after filtering does not prove an edge. R is not account percent.
First chart is first filtered trade; second first accepted control trade
whose matching filtered event fails strength. E/X/S/T on chart2 are CONTROL.
Examples were selected by chronology/event rules, never by later return.
