ForexBestRobots.com — quote-weighted Money Flow Index, real EUR/USD

Derived historical Bid inputs; no executed transaction sizes or trade results.
Source: HistData Generic ASCII EURUSD tick quotes, January 2024.
Original clock: fixed EST, UTC-05:00. CSV clock: UTC (add five hours).
Study: 2024-01-08 00:00 <= time < 17:00 UTC.
Viewer: 07:00 <= candle start < 17:00 UTC; prior history is retained.

N counts all records. Q counts Bid or Ask changes versus the previous archive
record, continuing across candle boundaries. These are reconstructed event
weights, not traded lots or a guaranteed match for native MT4 tick volume.
Integer prices are millionths; 1 EUR/USD pip equals 100 integer units.
Spreads are Ask minus Bid in those integer units; smed is the sample median.

TP = (High + Low + Close)/3. The sign compares consecutive TP values.
For the last L candles including the selected one, sum TP*W into F+ when
TP rises, F- when TP falls, neither when equal. W is N or Q.
Negative flow is stored as a positive magnitude. MFI=100*F+/(F++F-).
One-sided flow gives 100 or 0. Both sums zero gives unavailable, not 50.
The integer sum H+L+C is compared before rounding; scaling cancels in MFI.
There is no Wilder smoothing. A zero weight contributes zero.

L flows require L+1 chronological complete candles. Two adjacent available
MFI readings for a crossing require L+2. Missing required intervals or an
M5 group with fewer than five observed minutes makes that reading unavailable.
The missing minutes are 04:33, 05:29, 06:19 UTC. Missing candles are not filled.
The rolling calculation recovers after all required history passes a gap.
Minute coverage does not certify that all market updates were captured.

Bar labels are starts; intervals exclude the end. M1 labelled 07:40 is known
after 07:41; M5 labelled 07:30 after 07:35. Bid Close is not a guaranteed fill.
Zone: below lower / above upper / otherwise middle, including exact bounds.
ReturnLow: previous<=lower and current>lower.
ReturnHigh: previous>=upper and current<upper.
EnterLow: previous>=lower and current<lower.
EnterHigh: previous<=upper and current>upper.
All applicable events are retained; both readings must be available.
Levels 20/80 or 10/90 change classifications, not MFI values.

The first returns in the fixed viewer window are 07:40 (over 20) and 09:53
(under 80), with M1(14), N. They were not selected on trade outcomes.
The seven-candle above-80 run at starts 12:48-12:54 (last end 12:55) was
selected retrospectively as the longest run, to illustrate persistence.

Extract this kit, then run with Python 3 and its standard library:
  python3 replay_mfi.py
  python3 replay_mfi.py --start 2024-01-08T09:53:00Z
  python3 replay_mfi.py --timeframe 5 --start 2024-01-08T07:30:00Z
  python3 replay_mfi.py --period 28 --weight q --lower 10 --all

The raw monthly archive is identified by download page and SHA-256 in
source-and-method.json, not redistributed. No orders, fills, win rate,
backtest performance or profitability estimates are included.
