USD/JPY Wilder DMI/ADX research data — January–February 2024

Original unchanged Forexite MetaStock quotes, two-decimal USDJPY OHLC.
60 original daily archives were checked by SHA-256 and ZIP CRC. Source
filenames, URLs and hashes are in research/sources.json. Used minute records
preserve their original CET timestamps and converted UTC timestamps.
No Ask, historical spread, fill or trading return is inferred.

Offline replay, Python 3 standard library only:
  python3 reproduce.py
  python3 verify_calculation.py
The first recreates bars.json, examples.json and the H4 CSV byte-for-byte.
The second uses a separate Decimal70 implementation of smoothed AVERAGES,
not the producer's smoothed sums, plus independent minute grouping.

Declared calculation:
  Up = High[t]-High[t-1]; Down = Low[t-1]-Low[t]
  +DM = Up if Up>0 and Up>Down, otherwise 0
  -DM = Down if Down>0 and Down>Up, otherwise 0
  Equal extensions produce both DM=0.
  TR = max(High-Low, abs(High-prevClose), abs(Low-prevClose))
Price group 0 has no preceding transition. Seed each sum S at index N
with transitions 1..N. Then S[t] = S[t-1] - S[t-1]/N + raw[t].
  +DI = 100*S_+DM/S_TR; -DI = 100*S_-DM/S_TR
  DX = 100*abs(+DI-(-DI))/(+DI+(-DI))
First DI/DX is at index N. First ADX is the mean of DX[N..2N-1],
at index 2N-1. Then ADX[t]=(prevADX*(N-1)+DX[t])/N.
If S_TR=0, both DI are 0; if their sum=0, DX=0 by explicit convention.
N is 7, 14 or 28. Recursion retains history from the first included group;
changing the visible chart edge must not reset the seeds.

Source timestamps are grouped in (end-4h,end], labelled by the UTC end.
Winter CET minus one hour. No missing quotes filled. Short opening/closing
groups remain; the source-boundary-cut last group is omitted. Broker H4
boundaries and indicator initialization can differ from this protocol.

Optional original re-extraction: put the 60 named original ZIPs in raw/, then
run python3 prepare_data.py. FBR_FOREXITE_ARCHIVES may name another directory.
Optional SVG generation uses make_charts.py; PNG export needs Inkscape.
No network or external packages are required for the offline numerical replay.
