The two parts of the system

There are two separate trees, and it is worth being clear about which is which.

How the repository is located

On startup, candidate folders are tested in order and the first one that looks like the repo wins:

1. the parent of the package folder      # <repo>/FxMath_LR_Studio -> <repo>
2. the package folder itself             # exe dropped into the repo root
3. two levels up
4. the current working directory
5. the parent of the working directory

A folder counts as the repository only if both of these files exist:

scripts/feature_search.py
scripts/gen_ea.py

Application layout

FxMath_LR_Studio/
├── run.py                  # double-clickable launcher / frozen entry point
├── __main__.py             # python -m FxMath_LR_Studio
├── __init__.py             # exports APP_NAME, APP_VERSION, Settings
├── app.py                  # the window, navigation, output routing
├── config.py               # settings, discovery, path conventions
├── runner.py               # Task, JobManager, Chain, Presets
├── state.py                # read artefacts off disk, judge freshness
├── mt5.py                  # compile, parse the log, deploy
├── mt5env.py               # find terminals, pair install ↔ data dir
├── pyrunner.py             # the frozen interpreter stand-in
├── theme.py                # palette and fonts
├── widgets.py              # Card, KPI, LogConsole, charts
├── pages/                  # one module per screen
├── FxMath_LR_Studio.spec   # the PyInstaller build recipe
├── build_exe.bat           # build the standalone exes
├── run_studio.bat          # run from source
├── requirements.txt
├── settings.json           # persisted settings
└── README.md

The pages

The sidebar order is the pipeline order, which is deliberate — the navigation doubles as a workflow diagram.

Repository layout

<repo>/
├── scripts/                # the pipeline
│   ├── download_ohlc_all.py
│   ├── check_ohlc_csv.py
│   ├── feature_search.py   # hard marker for repo detection
│   ├── build_allstar.py
│   ├── best_hours.py
│   ├── holdout_tail.py
│   ├── gen_ea.py           # hard marker for repo detection
│   ├── report_m5.py
│   ├── report_tail.py
│   ├── broker_run_report.py
│   ├── backtest_lr.py      # the engine
│   ├── feature_sets.py     # the 11 sets + primitives
│   ├── lr_hours_engine.py  # exact hour-subset replay
│   ├── ExportOHLC_MQL5.mq5 # broker-side exporter
│   └── …                   # study scripts
├── data/                   # OHLC CSVs
│   └── XAUUSD_M5.csv
├── logs/                   # caches and run logs
├── strategies_lr_m5/       # XAUUSD M5 output root
│   ├── <set>/              # one folder per feature set
│   ├── all9/               # the portfolio
│   ├── best_hours.json
│   └── report*.html
└── README.md

Output roots and naming

Magic numbers

Every slot has a unique magic number, built from the set's base plus a timeframe offset plus the slot index:

magic = (LR_MAGIC_BASE[set]) + TF_MAGIC_OFFSET[tf] + slot_index

where LR_MAGIC_BASE[set] = FEATURE_SETS[set]["magic_start"] + 100

So base9 on M5 occupies 401–405 and the same set on M15 occupies 1401–1405. Several timeframes can therefore run on one account without their positions being confused.

Artefact schemas

data/<SYMBOL>_<TF>.csv

time,open,high,low,close,tick_volume,spread,real_volume

<root>/<set>/summary.json

A per-set result. The header keys, shared with portfolio summaries:

Each entry in slots[]:

<root>/<set>/configs.json

A list of parameter dicts — one per slot, including its magic. This is the input the EA generator consumes, and it is the file to inspect when you want to know what a run actually selected.

<root>/<set>/slots.txt

The human-readable equivalent of configs.json: a metrics line per slot, then every parameter on its own line. Useful for diffing two runs.

<root>/<set>/equity_<magic>.csv

One column, equity. Used by the correlation filter and the charts.

<root>/<folder>/equity_combined.csv

The combined portfolio curve, after warmup alignment and summation.

<root>/best_hours.json

{
  "engine": "lr",
  "symbol": "XAUUSD",
  "tf": "M5",
  "data": "XAUUSD_M5.csv",
  "<set>": {
    "best_windows": [
      { "hours": [...], "len": 5, "pnl": 812, "spec": "08-12", "n": 210,
        "in_pnl": 671, "in_n": 158, "oos_pnl": 141, "oos_n": 52 }
    ],
    "hour_pnl": { "0": ..., "1": ..., …, "23": ... },
    "hour_n":   { "0": ..., …, "23": ... },
    "slot_magic": 401
  }
}

best_windows[0].spec is what the EA generator bakes into InpSlot<N>Hours.

<root>/<set>/trades.csv

The trade statement for a set's best slot. Columns: side, entry, exit, pnl, reason, prob, bars_held, bar_entry, bar_exit.

reason is "SL", "TP" or "EOData" — the last meaning the position was still open at the end of the data and was closed at the final price.

<root>/<folder>/tail<N>_<mode>/tail_report.json

{
  "folder": "all9",
  "mode": "warm",
  "tail_bars": 2500,
  "n_bars": 10000,
  "tail_start": 7500,
  "combined": { "np": 57, "max_dd_pct": -0.41, "sharpe": 1.23,
                "buys": 400, "sells": 414, "trades": 814 },
  "slots": [
    { "magic": 401, "set": "base9", "np": 9, "dd": -0.12, "sharpe": …,
      "trades": 142, "wins": 76, "buys": 70, "sells": 72,
      "pnl_per_trade": 0.06,
      "sl": 2.0, "rr": 1.5, "la": 3, "tb": 1000, "ri": 500,
      "blr": 0.01, "th": 0.55 }
  ]
}

The generated EA

Reports

Settings

settings.json lives next to the app, or next to the frozen executable, so a rebuild does not lose it. It is a flat JSON object, sorted by key.

Repairs applied on load

Stale values are corrected every time settings load, and the corrections explain several surprising behaviours:

Which files are stale when

The app derives stage status from artefact modification times. This table is the complete rule set.

Collision rules

What protects runs from overwriting each other:

Which files are safe to delete