USD/JPY TRIX research data, January-February 2024

Data: published Forexite daily MetaStock quote OHLC.
Included unchanged minute rows preserve source CET and converted UTC.
Original ZIPs are not bundled; source URLs and SHA256 are in sources.json.
Original archive hashes and CRC were checked before extraction.
No prices, Ask, spread, intraminute path or order fills are inferred.

Offline replay (Python 3 standard library):
  python3 reproduce.py
  python3 verify_calculation.py

The first command recreates bars.json and usdjpy-h4-trix.csv byte-for-byte.
The second independently checks OHLC grouping and compares TRIX with a
Decimal precision-50 direct exponential-weight calculation, tolerance 2e-10.
N = 9, 15, 30; EMA signal length = 9.
Every EMA starts at the mean of its first N available inputs.
First EMA3 at zero-based index 3*(N-1), first TRIX one group later.
Signal is seeded with the first nine valid TRIX readings.
TRIX = 100*(EMA3/current_previous_EMA3-1); histogram = TRIX-signal (pp).

Grouping: record timestamps in (end-4h,end], labelled by end in UTC.
CET minus one hour in this winter interval. No filling; short opening
and closing groups retained, last cut group omitted. Broker candles may differ.
The visible chart range does not reset any EMA. This is not a trading backtest.

Optional charts: python3 make_charts.py (PNG export requires Inkscape).
