diagnostics

.idea/copilotDiffState.xml

@@ -0,0 +1,34 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="CopilotDiffPersistence">
+    <option name="pendingDiffs">
+      <map>
+        <entry key="$PROJECT_DIR$/README.md">
+          <value>
+            <PendingDiffInfo>
+              <option name="filePath" value="$PROJECT_DIR$/README.md" />
+              <option name="originalContent" value="---&#10;title: Hyperliquid Adaptive Bot&#10;emoji: &quot;&quot;&#10;colorFrom: blue&#10;colorTo: purple&#10;sdk: docker&#10;pinned: false&#10;---&#10;&#10;# Hyperliquid Adaptive Bot&#10;&#10;This project now includes an adaptive ML bot that uses your existing Hyperliquid framework.&#10;&#10;## What it does&#10;&#10;- Trades **only** `BTC`, `ETH`, and `SOL`&#10;- Runs in **paper mode by default**&#10;- Learns from closed paper trades with an online ML model (`SGDClassifier`)&#10;- Persists everything learned to disk (`learning_state.joblib`) so restart resumes learning&#10;- Persists paper runtime state (equity, open positions, entry features) so restart does not break learning continuity&#10;- Uses a profitability gate before any live execution is allowed&#10;- Sends Telegram alerts on trade open/close and a midnight UTC summary&#10;- Sends a readiness alert when paper performance is strong enough for live trading&#10;- Optional startup live transaction test (tiny BUY then SELL) to verify live order execution path&#10;- Can mirror to live orders only when:&#10;  - `--enable-live` is set&#10;  - paper performance is consistently profitable&#10;  - manual approval is granted (`ALLOW_LIVE_TRADING=1` or approval flag file exists)&#10;  - `HL_PRIVATE_KEY` is configured&#10;&#10;## Files&#10;&#10;- `hbot.py`: existing Hyperliquid API framework and order placement helpers&#10;- `adaptive_bot.py`: new adaptive ML paper/live-gated trading bot&#10;- `smoke_test.py`: offline persistence and learning smoke test&#10;- `paper_report.json`: runtime status and metrics output (created at runtime)&#10;&#10;## Health endpoints (for HF + UptimeRobot)&#10;&#10;- `GET /` -&gt; lightweight status response (`200`) for uptime checks&#10;- `GET /healthz` -&gt; same as root (`200`)&#10;- `GET /status` -&gt; concise overview (equity, open position count, performance, live gate)&#10;- `GET /status/full` -&gt; full runtime JSON (equity, open positions, metrics, thresholds)&#10;- `GET /trades` -&gt; open positions list and count&#10;&#10;If your Space is `https://chigginrice-hyperliquid-bot.hf.space/`, your existing root monitor will work as-is.&#10;&#10;## Quick start&#10;&#10;1. Install dependencies:&#10;&#10;```bash&#10;python3 -m venv .venv&#10;source .venv/bin/activate&#10;pip install -r requirements.txt&#10;```&#10;&#10;2. Run the offline smoke test:&#10;&#10;```bash&#10;python smoke_test.py&#10;```&#10;&#10;3. Start the adaptive paper bot:&#10;&#10;```bash&#10;python adaptive_bot.py --interval 15m --poll-seconds 45&#10;```&#10;&#10;Optional knobs for early growth from scratch:&#10;&#10;```bash&#10;python adaptive_bot.py --interval 15m --exploration-rate 0.08&#10;```&#10;&#10;## Optional: allow live mirror only after paper consistency gate&#10;&#10;```bash&#10;export HL_PRIVATE_KEY=&quot;your_private_key&quot;&#10;# optional if using delegated account&#10;export HL_ACCOUNT_ADDRESS=&quot;0x...&quot;&#10;python adaptive_bot.py --interval 15m --enable-live&#10;```&#10;&#10;By default, live arm still requires manual approval even after the paper gate is passed:&#10;&#10;```bash&#10;export ALLOW_LIVE_TRADING=1&#10;# or: touch allow_live_trading.flag&#10;python adaptive_bot.py --interval 15m --enable-live&#10;```&#10;&#10;Strict two-step approval mode (recommended for safety):&#10;&#10;```bash&#10;export ALLOW_LIVE_TRADING=1&#10;touch allow_live_trading.flag&#10;python adaptive_bot.py --interval 15m --enable-live --live-approval-mode both&#10;```&#10;&#10;Optional startup live execution test (real tiny transaction):&#10;&#10;```bash&#10;python adaptive_bot.py --interval 15m --enable-live --live-startup-test --live-startup-test-coin BTC --live-startup-test-notional 12&#10;```&#10;&#10;## Docker (Hugging Face)&#10;&#10;`Dockerfile` is included for Docker-based hosting.&#10;&#10;Build locally:&#10;&#10;```bash&#10;docker build -t hyperliquid-adaptive-bot .&#10;```&#10;&#10;Run locally in paper mode:&#10;&#10;```bash&#10;docker run --rm \&#10;  -e BOT_INTERVAL=15m \&#10;  -e BOT_POLL_SECONDS=45 \&#10;  -v &quot;$(pwd)/data:/data&quot; \&#10;  hyperliquid-adaptive-bot&#10;```&#10;&#10;Run with live mode allowed after paper gate passes:&#10;&#10;```bash&#10;docker run --rm \&#10;  -e BOT_ENABLE_LIVE=1 \&#10;  -e HL_PRIVATE_KEY=&quot;your_private_key&quot; \&#10;  -e HL_ACCOUNT_ADDRESS=&quot;0x...&quot; \&#10;  -v &quot;$(pwd)/data:/data&quot; \&#10;  hyperliquid-adaptive-bot&#10;```&#10;&#10;Environment variables used by the container:&#10;&#10;- `BOT_INTERVAL` (default `15m`)&#10;- `BOT_POLL_SECONDS` (default `45`)&#10;- `BOT_STATE_FILE` (default `/data/learning_state.joblib`)&#10;- `BOT_REPORT_FILE` (default `/data/paper_report.json`)&#10;- `BOT_ENABLE_LIVE` (set any non-empty value to append `--enable-live`)&#10;- `BOT_HEALTH_HOST` (default `0.0.0.0`)&#10;- `BOT_HEALTH_PORT` (default `7860`)&#10;- `PORT` (default `7860`, used by Hugging Face Spaces)&#10;- `ALLOW_LIVE_TRADING` (set to `1/true/yes` to manually approve live trading)&#10;- `BOT_LIVE_STARTUP_TEST` (set non-empty to add `--live-startup-test`)&#10;- `BOT_LIVE_STARTUP_TEST_COIN` (default `BTC`)&#10;- `BOT_LIVE_STARTUP_TEST_NOTIONAL` (default `12`)&#10;- `TELEGRAM_BOT_TOKEN` (required for Telegram notifications)&#10;- `TELEGRAM_CHAT_ID` (required for Telegram notifications)&#10;- `PROXY_URL` (required by `telegram_notifier.py` to forward messages)&#10;- `HL_PRIVATE_KEY` (required for live mirror)&#10;- `HL_ACCOUNT_ADDRESS` (optional)&#10;&#10;## Telegram notifications&#10;&#10;- Entry alert: sent when a paper position opens&#10;- Exit alert: sent when TP/SL closes a position&#10;- Midnight summary: sent once per day at `00:00 UTC`&#10;- Readiness alert: sent when profitability gate is reached and bot is ready to request live approval&#10;&#10;Summary includes last-24h trades, win rate, PnL, and overall cumulative stats.&#10;&#10;## How the ML learns&#10;&#10;At a high level, this bot uses online learning so it improves continuously while paper trading:&#10;&#10;1. For each coin (`BTC`, `ETH`, `SOL`), it builds a feature vector from latest candles:&#10;   - short-horizon returns (`ret_1`, `ret_3`, `ret_8`)&#10;   - trend/momentum (`macd`, scaled `rsi`)&#10;   - volatility and activity (`atr14`, `vol_z`)&#10;2. `SGDClassifier(loss=&quot;log_loss&quot;)` predicts probability of a profitable long setup.&#10;3. Entry logic maps probability to actions:&#10;   - `p_long &gt;= threshold` -&gt; long&#10;   - `p_long &lt;= 1 - threshold` -&gt; short&#10;   - otherwise hold&#10;   - when model is cold, bootstrap mode uses weak momentum direction to generate first labels&#10;   - optional exploration (`--exploration-rate`) continues collecting data even during neutral model states&#10;4. When a trade closes (TP/SL), outcome is the supervision label:&#10;   - profitable -&gt; label `1`&#10;   - losing -&gt; label `0`&#10;5. The model is updated immediately using `partial_fit(...)` on that single labeled example.&#10;6. The confidence threshold adapts from recent win rate (last 30 outcomes), becoming stricter or looser.&#10;7. Learned state is persisted to `learning_state.joblib` and reloaded on restart.&#10;8. Open paper positions and captured entry features are also persisted, so trades crossing restarts still train the model on close.&#10;&#10;So each closed paper trade directly trains the next decision, and training survives restarts.&#10;&#10;## Notes&#10;&#10;- The consistency gate currently requires:&#10;  - at least 30 paper trades&#10;  - win rate &gt;= 58%&#10;  - profit factor &gt;= 1.25&#10;- State persistence file is configurable with `--state-file`&#10;- Runtime report file is configurable with `--report-file`&#10;- Live approval can be controlled with:&#10;  - `--require-live-approval / --no-require-live-approval`&#10;  - `--live-approval-mode either|both` (default `either`)&#10;  - `--live-approval-file` (default `allow_live_trading.flag`)&#10;- Startup live order test can be controlled with:&#10;  - `--live-startup-test / --no-live-startup-test`&#10;  - `--live-startup-test-coin` (default `BTC`)&#10;  - `--live-startup-test-notional` (default `12`)&#10;&#10;&#10;" />
+              <option name="updatedContent" value="---&#10;title: Hyperliquid Adaptive Bot&#10;emoji: &quot;&quot;&#10;colorFrom: blue&#10;colorTo: purple&#10;sdk: docker&#10;pinned: false&#10;---&#10;&#10;# Hyperliquid Adaptive Bot&#10;&#10;This project now includes an adaptive ML bot that uses your existing Hyperliquid framework.&#10;&#10;## What it does&#10;&#10;- Trades **only** `BTC`, `ETH`, and `SOL`&#10;- Runs in **paper mode by default**&#10;- Learns from closed paper trades with an online ML model (`SGDClassifier`)&#10;- Persists everything learned to disk (`learning_state.joblib`) so restart resumes learning&#10;- Persists paper runtime state (equity, open positions, entry features) so restart does not break learning continuity&#10;- Uses a profitability gate before any live execution is allowed&#10;- Sends Telegram alerts on trade open/close and a midnight UTC summary&#10;- Sends a readiness alert when paper performance is strong enough for live trading&#10;- Optional startup live transaction test (tiny BUY then SELL) to verify live order execution path&#10;- Can mirror to live orders only when:&#10;  - `--enable-live` is set&#10;  - paper performance is consistently profitable&#10;  - manual approval is granted (`ALLOW_LIVE_TRADING=1` or approval flag file exists)&#10;  - `HL_PRIVATE_KEY` is configured&#10;&#10;## Files&#10;&#10;- `hbot.py`: existing Hyperliquid API framework and order placement helpers&#10;- `adaptive_bot.py`: new adaptive ML paper/live-gated trading bot&#10;- `smoke_test.py`: offline persistence and learning smoke test&#10;- `ml_diagnostics.py`: offline + live learning diagnostics report&#10;- `diagnostics_smoke_test.py`: quick smoke test for diagnostics logic&#10;- `paper_report.json`: runtime status and metrics output (created at runtime)&#10;&#10;## Health endpoints (for HF + UptimeRobot)&#10;&#10;- `GET /` -&gt; lightweight status response (`200`) for uptime checks&#10;- `GET /healthz` -&gt; same as root (`200`)&#10;- `GET /status` -&gt; concise overview (equity, open position count, performance, live gate)&#10;- `GET /status/full` -&gt; full runtime JSON (equity, open positions, metrics, thresholds)&#10;- `GET /trades` -&gt; open positions list and count&#10;&#10;If your Space is `https://chigginrice-hyperliquid-bot.hf.space/`, your existing root monitor will work as-is.&#10;&#10;## Quick start&#10;&#10;1. Install dependencies:&#10;&#10;```bash&#10;python3 -m venv .venv&#10;source .venv/bin/activate&#10;pip install -r requirements.txt&#10;```&#10;&#10;2. Run the offline smoke test:&#10;&#10;```bash&#10;python smoke_test.py&#10;```&#10;&#10;Optional: run diagnostics smoke test:&#10;&#10;```bash&#10;python diagnostics_smoke_test.py&#10;```&#10;&#10;3. Start the adaptive paper bot:&#10;&#10;```bash&#10;python adaptive_bot.py --interval 15m --poll-seconds 45&#10;```&#10;&#10;4. Check ML learning quality from saved state:&#10;&#10;```bash&#10;python ml_diagnostics.py --state-file learning_state.joblib&#10;```&#10;&#10;If the bot is running locally, include live status snapshot:&#10;&#10;```bash&#10;python ml_diagnostics.py --state-file learning_state.joblib --status-url http://127.0.0.1:7860/status/full&#10;```&#10;&#10;JSON output (useful for storing artifacts in CI/HF workflows):&#10;&#10;```bash&#10;python ml_diagnostics.py --state-file learning_state.joblib --json&#10;```&#10;&#10;Optional knobs for early growth from scratch:&#10;&#10;```bash&#10;python adaptive_bot.py --interval 15m --exploration-rate 0.08&#10;```&#10;&#10;## Optional: allow live mirror only after paper consistency gate&#10;&#10;```bash&#10;export HL_PRIVATE_KEY=&quot;your_private_key&quot;&#10;# optional if using delegated account&#10;export HL_ACCOUNT_ADDRESS=&quot;0x...&quot;&#10;python adaptive_bot.py --interval 15m --enable-live&#10;```&#10;&#10;By default, live arm still requires manual approval even after the paper gate is passed:&#10;&#10;```bash&#10;export ALLOW_LIVE_TRADING=1&#10;# or: touch allow_live_trading.flag&#10;python adaptive_bot.py --interval 15m --enable-live&#10;```&#10;&#10;Strict two-step approval mode (recommended for safety):&#10;&#10;```bash&#10;export ALLOW_LIVE_TRADING=1&#10;touch allow_live_trading.flag&#10;python adaptive_bot.py --interval 15m --enable-live --live-approval-mode both&#10;```&#10;&#10;Optional startup live execution test (real tiny transaction):&#10;&#10;```bash&#10;python adaptive_bot.py --interval 15m --enable-live --live-startup-test --live-startup-test-coin BTC --live-startup-test-notional 12&#10;```&#10;&#10;## Docker (Hugging Face)&#10;&#10;`Dockerfile` is included for Docker-based hosting.&#10;&#10;Build locally:&#10;&#10;```bash&#10;docker build -t hyperliquid-adaptive-bot .&#10;```&#10;&#10;Run locally in paper mode:&#10;&#10;```bash&#10;docker run --rm \&#10;  -e BOT_INTERVAL=15m \&#10;  -e BOT_POLL_SECONDS=45 \&#10;  -v &quot;$(pwd)/data:/data&quot; \&#10;  hyperliquid-adaptive-bot&#10;```&#10;&#10;Run with live mode allowed after paper gate passes:&#10;&#10;```bash&#10;docker run --rm \&#10;  -e BOT_ENABLE_LIVE=1 \&#10;  -e HL_PRIVATE_KEY=&quot;your_private_key&quot; \&#10;  -e HL_ACCOUNT_ADDRESS=&quot;0x...&quot; \&#10;  -v &quot;$(pwd)/data:/data&quot; \&#10;  hyperliquid-adaptive-bot&#10;```&#10;&#10;Environment variables used by the container:&#10;&#10;- `BOT_INTERVAL` (default `15m`)&#10;- `BOT_POLL_SECONDS` (default `45`)&#10;- `BOT_STATE_FILE` (default `/data/learning_state.joblib`)&#10;- `BOT_REPORT_FILE` (default `/data/paper_report.json`)&#10;- `BOT_ENABLE_LIVE` (set any non-empty value to append `--enable-live`)&#10;- `BOT_HEALTH_HOST` (default `0.0.0.0`)&#10;- `BOT_HEALTH_PORT` (default `7860`)&#10;- `PORT` (default `7860`, used by Hugging Face Spaces)&#10;- `ALLOW_LIVE_TRADING` (set to `1/true/yes` to manually approve live trading)&#10;- `BOT_LIVE_STARTUP_TEST` (set non-empty to add `--live-startup-test`)&#10;- `BOT_LIVE_STARTUP_TEST_COIN` (default `BTC`)&#10;- `BOT_LIVE_STARTUP_TEST_NOTIONAL` (default `12`)&#10;- `TELEGRAM_BOT_TOKEN` (required for Telegram notifications)&#10;- `TELEGRAM_CHAT_ID` (required for Telegram notifications)&#10;- `PROXY_URL` (required by `telegram_notifier.py` to forward messages)&#10;- `HL_PRIVATE_KEY` (required for live mirror)&#10;- `HL_ACCOUNT_ADDRESS` (optional)&#10;&#10;## Telegram notifications&#10;&#10;- Entry alert: sent when a paper position opens&#10;- Exit alert: sent when TP/SL closes a position&#10;- Midnight summary: sent once per day at `00:00 UTC`&#10;- Readiness alert: sent when profitability gate is reached and bot is ready to request live approval&#10;&#10;Summary includes last-24h trades, win rate, PnL, and overall cumulative stats.&#10;&#10;## How the ML learns&#10;&#10;At a high level, this bot uses online learning so it improves continuously while paper trading:&#10;&#10;1. For each coin (`BTC`, `ETH`, `SOL`), it builds a feature vector from latest candles:&#10;   - short-horizon returns (`ret_1`, `ret_3`, `ret_8`)&#10;   - trend/momentum (`macd`, scaled `rsi`)&#10;   - volatility and activity (`atr14`, `vol_z`)&#10;2. `SGDClassifier(loss=&quot;log_loss&quot;)` predicts probability of a profitable long setup.&#10;3. Entry logic maps probability to actions:&#10;   - `p_long &gt;= threshold` -&gt; long&#10;   - `p_long &lt;= 1 - threshold` -&gt; short&#10;   - otherwise hold&#10;   - when model is cold, bootstrap mode uses weak momentum direction to generate first labels&#10;   - optional exploration (`--exploration-rate`) continues collecting data even during neutral model states&#10;4. When a trade closes (TP/SL), outcome is the supervision label:&#10;   - profitable -&gt; label `1`&#10;   - losing -&gt; label `0`&#10;5. The model is updated immediately using `partial_fit(...)` on that single labeled example.&#10;6. The confidence threshold adapts from recent win rate (last 30 outcomes), becoming stricter or looser.&#10;7. Learned state is persisted to `learning_state.joblib` and reloaded on restart.&#10;8. Open paper positions and captured entry features are also persisted, so trades crossing restarts still train the model on close.&#10;&#10;So each closed paper trade directly trains the next decision, and training survives restarts.&#10;&#10;## ML diagnostics&#10;&#10;`ml_diagnostics.py` helps answer whether the model signal is actually improving.&#10;&#10;It reports:&#10;&#10;- overall/per-coin/per-side win rate and expectancy&#10;- directional sanity checks:&#10;  - long winners should have higher `p_long` than long losers&#10;  - short winners should have lower `p_long` than short losers&#10;- direction-aware Brier score (lower is better)&#10;- optional runtime snapshot from `/status/full` including rolling Brier/expectancy/drawdown&#10;&#10;This is useful to catch label/specification issues where the model keeps repeating failing trade patterns.&#10;&#10;## Notes&#10;&#10;- The consistency gate currently requires:&#10;  - at least 30 paper trades&#10;  - win rate &gt;= 58%&#10;  - profit factor &gt;= 1.25&#10;- State persistence file is configurable with `--state-file`&#10;- Runtime report file is configurable with `--report-file`&#10;- Live approval can be controlled with:&#10;  - `--require-live-approval / --no-require-live-approval`&#10;  - `--live-approval-mode either|both` (default `either`)&#10;  - `--live-approval-file` (default `allow_live_trading.flag`)&#10;- Startup live order test can be controlled with:&#10;  - `--live-startup-test / --no-live-startup-test`&#10;  - `--live-startup-test-coin` (default `BTC`)&#10;  - `--live-startup-test-notional` (default `12`)&#10;&#10;&#10;" />
+            </PendingDiffInfo>
+          </value>
+        </entry>
+        <entry key="$PROJECT_DIR$/diagnostics_smoke_test.py">
+          <value>
+            <PendingDiffInfo>
+              <option name="filePath" value="$PROJECT_DIR$/diagnostics_smoke_test.py" />
+              <option name="updatedContent" value="#!/usr/bin/env python3&#10;&quot;&quot;&quot;Smoke test for ml_diagnostics report builder.&quot;&quot;&quot;&#10;&#10;from __future__ import annotations&#10;&#10;import tempfile&#10;from pathlib import Path&#10;&#10;import joblib&#10;&#10;from ml_diagnostics import build_report&#10;&#10;&#10;def main() -&gt; None:&#10;    payload = {&#10;        &quot;closed_trades&quot;: [&#10;            {&#10;                &quot;coin&quot;: &quot;BTC&quot;,&#10;                &quot;side&quot;: &quot;long&quot;,&#10;                &quot;entry_ts&quot;: 1000,&#10;                &quot;pnl_usd&quot;: 5.0,&#10;                &quot;entry_p_long&quot;: 0.72,&#10;            },&#10;            {&#10;                &quot;coin&quot;: &quot;BTC&quot;,&#10;                &quot;side&quot;: &quot;short&quot;,&#10;                &quot;entry_ts&quot;: 2000,&#10;                &quot;pnl_usd&quot;: -3.0,&#10;                &quot;entry_p_long&quot;: 0.81,&#10;            },&#10;            {&#10;                &quot;coin&quot;: &quot;ETH&quot;,&#10;                &quot;side&quot;: &quot;short&quot;,&#10;                &quot;entry_ts&quot;: 3000,&#10;                &quot;pnl_usd&quot;: 4.0,&#10;                &quot;entry_p_long&quot;: 0.22,&#10;            },&#10;        ],&#10;        &quot;trade_events&quot;: [&#10;            {&quot;event&quot;: &quot;open&quot;, &quot;coin&quot;: &quot;BTC&quot;, &quot;entry_ts&quot;: 1000, &quot;p_long&quot;: 0.72},&#10;            {&quot;event&quot;: &quot;open&quot;, &quot;coin&quot;: &quot;BTC&quot;, &quot;entry_ts&quot;: 2000, &quot;p_long&quot;: 0.81},&#10;            {&quot;event&quot;: &quot;open&quot;, &quot;coin&quot;: &quot;ETH&quot;, &quot;entry_ts&quot;: 3000, &quot;p_long&quot;: 0.22},&#10;        ],&#10;    }&#10;&#10;    with tempfile.TemporaryDirectory() as td:&#10;        state_path = Path(td) / &quot;learning_state.joblib&quot;&#10;        joblib.dump(payload, state_path)&#10;        loaded = joblib.load(state_path)&#10;&#10;    report = build_report(loaded)&#10;&#10;    assert report[&quot;state_samples&quot;][&quot;closed_trades&quot;] == 3&#10;    assert &quot;overall&quot; in report&#10;    assert &quot;directional_sanity&quot; in report&#10;    assert report[&quot;brier_direction_aware&quot;] is not None&#10;&#10;    print(&quot;diagnostics_smoke_test: OK&quot;)&#10;&#10;&#10;if __name__ == &quot;__main__&quot;:&#10;    main()&#10;" />
+            </PendingDiffInfo>
+          </value>
+        </entry>
+        <entry key="$PROJECT_DIR$/ml_diagnostics.py">
+          <value>
+            <PendingDiffInfo>
+              <option name="filePath" value="$PROJECT_DIR$/ml_diagnostics.py" />
+              <option name="updatedContent" value="#!/usr/bin/env python3&#10;&quot;&quot;&quot;Offline and live diagnostics for adaptive bot learning quality.&#10;&#10;This script helps validate whether the model is learning directionally useful signals.&#10;&quot;&quot;&quot;&#10;&#10;from __future__ import annotations&#10;&#10;import argparse&#10;import json&#10;import math&#10;import os&#10;from collections import defaultdict&#10;from typing import Any, Dict, List, Optional, Tuple&#10;from urllib.error import URLError&#10;from urllib.request import urlopen&#10;&#10;import joblib&#10;&#10;&#10;def _safe_float(v: Any, default: float = 0.0) -&gt; float:&#10;    try:&#10;        return float(v)&#10;    except Exception:&#10;        return default&#10;&#10;&#10;def _mean(values: List[float]) -&gt; Optional[float]:&#10;    return (sum(values) / len(values)) if values else None&#10;&#10;&#10;def _trade_p_long(trade: Dict[str, Any], open_probs: Dict[Tuple[str, int], float]) -&gt; float:&#10;    if &quot;entry_p_long&quot; in trade and trade[&quot;entry_p_long&quot;] is not None:&#10;        return float(max(0.0, min(1.0, _safe_float(trade[&quot;entry_p_long&quot;], 0.5))))&#10;    key = (str(trade.get(&quot;coin&quot;, &quot;&quot;)), int(_safe_float(trade.get(&quot;entry_ts&quot;, 0), 0)))&#10;    return float(max(0.0, min(1.0, open_probs.get(key, 0.5))))&#10;&#10;&#10;def _summarize_basic(trades: List[Dict[str, Any]]) -&gt; Dict[str, float]:&#10;    n = len(trades)&#10;    wins = sum(1 for t in trades if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &gt; 0)&#10;    pnl = sum(_safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) for t in trades)&#10;    return {&#10;        &quot;trades&quot;: float(n),&#10;        &quot;wins&quot;: float(wins),&#10;        &quot;win_rate&quot;: (wins / n) if n else 0.0,&#10;        &quot;total_pnl_usd&quot;: pnl,&#10;        &quot;expectancy_usd&quot;: (pnl / n) if n else 0.0,&#10;    }&#10;&#10;&#10;def _directional_sanity(trades: List[Dict[str, Any]], open_probs: Dict[Tuple[str, int], float]) -&gt; Dict[str, Optional[float]]:&#10;    longs = [t for t in trades if str(t.get(&quot;side&quot;, &quot;&quot;)).lower() == &quot;long&quot;]&#10;    shorts = [t for t in trades if str(t.get(&quot;side&quot;, &quot;&quot;)).lower() == &quot;short&quot;]&#10;&#10;    long_win = [_trade_p_long(t, open_probs) for t in longs if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &gt; 0]&#10;    long_lose = [_trade_p_long(t, open_probs) for t in longs if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &lt;= 0]&#10;    short_win = [_trade_p_long(t, open_probs) for t in shorts if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &gt; 0]&#10;    short_lose = [_trade_p_long(t, open_probs) for t in shorts if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &lt;= 0]&#10;&#10;    # For a healthy long-probability model:&#10;    # - long winners should have higher p_long than long losers&#10;    # - short winners should have lower p_long than short losers&#10;    return {&#10;        &quot;long_winners_avg_p_long&quot;: _mean(long_win),&#10;        &quot;long_losers_avg_p_long&quot;: _mean(long_lose),&#10;        &quot;short_winners_avg_p_long&quot;: _mean(short_win),&#10;        &quot;short_losers_avg_p_long&quot;: _mean(short_lose),&#10;    }&#10;&#10;&#10;def _direction_aware_brier(trades: List[Dict[str, Any]], open_probs: Dict[Tuple[str, int], float]) -&gt; Optional[float]:&#10;    scores: List[float] = []&#10;    for t in trades:&#10;        side = str(t.get(&quot;side&quot;, &quot;&quot;)).lower()&#10;        if side not in {&quot;long&quot;, &quot;short&quot;}:&#10;            continue&#10;        p_long = _trade_p_long(t, open_probs)&#10;        won = 1.0 if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &gt; 0 else 0.0&#10;        # Target for long probability:&#10;        # - long trade: y=1 if win&#10;        # - short trade: y=1 if short loses (long would've won)&#10;        y_long = won if side == &quot;long&quot; else (1.0 - won)&#10;        scores.append((p_long - y_long) ** 2)&#10;    return _mean(scores)&#10;&#10;&#10;def _calibration_table(&#10;    trades: List[Dict[str, Any]], open_probs: Dict[Tuple[str, int], float], side: str&#10;) -&gt; List[Dict[str, Any]]:&#10;    bins: Dict[int, List[Dict[str, Any]]] = defaultdict(list)&#10;    for t in trades:&#10;        t_side = str(t.get(&quot;side&quot;, &quot;&quot;)).lower()&#10;        if t_side != side:&#10;            continue&#10;        p = _trade_p_long(t, open_probs)&#10;        decile = min(9, max(0, int(math.floor(p * 10.0))))&#10;        bins[decile].append(t)&#10;&#10;    rows: List[Dict[str, Any]] = []&#10;    for decile in range(10):&#10;        arr = bins.get(decile, [])&#10;        if not arr:&#10;            continue&#10;        n = len(arr)&#10;        wins = sum(1 for t in arr if _safe_float(t.get(&quot;pnl_usd&quot;, 0.0), 0.0) &gt; 0)&#10;        avg_p = _mean([_trade_p_long(t, open_probs) for t in arr])&#10;        rows.append(&#10;            {&#10;                &quot;decile&quot;: decile,&#10;                &quot;range&quot;: f&quot;{decile / 10:.1f}-{(decile + 1) / 10:.1f}&quot;,&#10;                &quot;trades&quot;: n,&#10;                &quot;avg_p_long&quot;: avg_p,&#10;                &quot;win_rate&quot;: (wins / n) if n else None,&#10;            }&#10;        )&#10;    return rows&#10;&#10;&#10;def _open_prob_lookup(events: List[Dict[str, Any]]) -&gt; Dict[Tuple[str, int], float]:&#10;    out: Dict[Tuple[str, int], float] = {}&#10;    for e in events:&#10;        if e.get(&quot;event&quot;) != &quot;open&quot;:&#10;            continue&#10;        coin = str(e.get(&quot;coin&quot;, &quot;&quot;))&#10;        entry_ts = int(_safe_float(e.get(&quot;entry_ts&quot;, 0), 0))&#10;        p_long = float(max(0.0, min(1.0, _safe_float(e.get(&quot;p_long&quot;, 0.5), 0.5))))&#10;        out[(coin, entry_ts)] = p_long&#10;    return out&#10;&#10;&#10;def _fetch_status(url: str) -&gt; Optional[Dict[str, Any]]:&#10;    try:&#10;        with urlopen(url, timeout=3) as response:&#10;            if response.status != 200:&#10;                return None&#10;            body = response.read().decode(&quot;utf-8&quot;)&#10;            return json.loads(body)&#10;    except (URLError, TimeoutError, json.JSONDecodeError, ValueError):&#10;        return None&#10;&#10;&#10;def build_report(payload: Dict[str, Any], status_payload: Optional[Dict[str, Any]] = None) -&gt; Dict[str, Any]:&#10;    closed = payload.get(&quot;closed_trades&quot;, []) or []&#10;    events = payload.get(&quot;trade_events&quot;, []) or []&#10;    open_probs = _open_prob_lookup(events)&#10;&#10;    by_coin: Dict[str, List[Dict[str, Any]]] = defaultdict(list)&#10;    by_side: Dict[str, List[Dict[str, Any]]] = defaultdict(list)&#10;    for t in closed:&#10;        by_coin[str(t.get(&quot;coin&quot;, &quot;unknown&quot;))].append(t)&#10;        by_side[str(t.get(&quot;side&quot;, &quot;unknown&quot;)).lower()].append(t)&#10;&#10;    per_coin = {coin: _summarize_basic(trades) for coin, trades in sorted(by_coin.items())}&#10;    per_side = {side: _summarize_basic(trades) for side, trades in sorted(by_side.items())}&#10;&#10;    report: Dict[str, Any] = {&#10;        &quot;state_samples&quot;: {&#10;            &quot;closed_trades&quot;: len(closed),&#10;            &quot;trade_events&quot;: len(events),&#10;        },&#10;        &quot;overall&quot;: _summarize_basic(closed),&#10;        &quot;per_coin&quot;: per_coin,&#10;        &quot;per_side&quot;: per_side,&#10;        &quot;directional_sanity&quot;: _directional_sanity(closed, open_probs),&#10;        &quot;brier_direction_aware&quot;: _direction_aware_brier(closed, open_probs),&#10;        &quot;calibration&quot;: {&#10;            &quot;long&quot;: _calibration_table(closed, open_probs, &quot;long&quot;),&#10;            &quot;short&quot;: _calibration_table(closed, open_probs, &quot;short&quot;),&#10;        },&#10;        &quot;runtime_status&quot;: status_payload,&#10;    }&#10;&#10;    # High-level warnings to speed up triage.&#10;    warnings: List[str] = []&#10;    ds = report[&quot;directional_sanity&quot;]&#10;    lw = ds.get(&quot;long_winners_avg_p_long&quot;)&#10;    ll = ds.get(&quot;long_losers_avg_p_long&quot;)&#10;    sw = ds.get(&quot;short_winners_avg_p_long&quot;)&#10;    sl = ds.get(&quot;short_losers_avg_p_long&quot;)&#10;&#10;    if lw is not None and ll is not None and lw &lt;= ll:&#10;        warnings.append(&quot;Long directional signal is inverted or weak: long winners do not have higher p_long than long losers.&quot;)&#10;    if sw is not None and sl is not None and sw &gt;= sl:&#10;        warnings.append(&quot;Short directional signal is inverted or weak: short winners do not have lower p_long than short losers.&quot;)&#10;&#10;    overall_wr = report[&quot;overall&quot;].get(&quot;win_rate&quot;, 0.0)&#10;    overall_exp = report[&quot;overall&quot;].get(&quot;expectancy_usd&quot;, 0.0)&#10;    if len(closed) &gt;= 50 and overall_wr &lt; 0.45:&#10;        warnings.append(&quot;Win rate below 45% after 50+ trades; model may be mis-specified or overtrading.&quot;)&#10;    if len(closed) &gt;= 50 and overall_exp &lt; 0:&#10;        warnings.append(&quot;Negative expectancy after 50+ trades; consider reducing exploration and revisiting labels.&quot;)&#10;&#10;    report[&quot;warnings&quot;] = warnings&#10;    return report&#10;&#10;&#10;def print_report(report: Dict[str, Any]) -&gt; None:&#10;    print(&quot;=== Adaptive Bot ML Diagnostics ===&quot;)&#10;    samples = report[&quot;state_samples&quot;]&#10;    print(f&quot;Closed trades: {samples['closed_trades']}&quot;)&#10;    print(f&quot;Trade events: {samples['trade_events']}&quot;)&#10;&#10;    overall = report[&quot;overall&quot;]&#10;    print(&#10;        &quot;Overall: &quot;&#10;        f&quot;n={int(overall['trades'])}, &quot;&#10;        f&quot;win_rate={overall['win_rate'] * 100:.2f}%, &quot;&#10;        f&quot;total_pnl={overall['total_pnl_usd']:.2f}, &quot;&#10;        f&quot;expectancy={overall['expectancy_usd']:.4f}&quot;&#10;    )&#10;&#10;    print(&quot;\nBy side:&quot;)&#10;    for side, s in report[&quot;per_side&quot;].items():&#10;        print(&#10;            f&quot;  {side}: n={int(s['trades'])}, win_rate={s['win_rate'] * 100:.2f}%, &quot;&#10;            f&quot;total_pnl={s['total_pnl_usd']:.2f}, expectancy={s['expectancy_usd']:.4f}&quot;&#10;        )&#10;&#10;    print(&quot;\nBy coin:&quot;)&#10;    for coin, s in report[&quot;per_coin&quot;].items():&#10;        print(&#10;            f&quot;  {coin}: n={int(s['trades'])}, win_rate={s['win_rate'] * 100:.2f}%, &quot;&#10;            f&quot;total_pnl={s['total_pnl_usd']:.2f}, expectancy={s['expectancy_usd']:.4f}&quot;&#10;        )&#10;&#10;    ds = report[&quot;directional_sanity&quot;]&#10;    print(&quot;\nDirectional sanity:&quot;)&#10;    print(&#10;        &quot;  Long winners avg p_long: &quot;&#10;        f&quot;{ds['long_winners_avg_p_long'] if ds['long_winners_avg_p_long'] is not None else 'n/a'}&quot;&#10;    )&#10;    print(&#10;        &quot;  Long losers avg p_long: &quot;&#10;        f&quot;{ds['long_losers_avg_p_long'] if ds['long_losers_avg_p_long'] is not None else 'n/a'}&quot;&#10;    )&#10;    print(&#10;        &quot;  Short winners avg p_long: &quot;&#10;        f&quot;{ds['short_winners_avg_p_long'] if ds['short_winners_avg_p_long'] is not None else 'n/a'}&quot;&#10;    )&#10;    print(&#10;        &quot;  Short losers avg p_long: &quot;&#10;        f&quot;{ds['short_losers_avg_p_long'] if ds['short_losers_avg_p_long'] is not None else 'n/a'}&quot;&#10;    )&#10;&#10;    brier = report.get(&quot;brier_direction_aware&quot;)&#10;    print(f&quot;\nDirection-aware Brier (lower is better): {brier if brier is not None else 'n/a'}&quot;)&#10;&#10;    warnings = report.get(&quot;warnings&quot;, [])&#10;    if warnings:&#10;        print(&quot;\nWarnings:&quot;)&#10;        for w in warnings:&#10;            print(f&quot;  - {w}&quot;)&#10;&#10;    runtime = report.get(&quot;runtime_status&quot;)&#10;    if runtime:&#10;        print(&quot;\nLive status snapshot:&quot;)&#10;        metrics = runtime.get(&quot;metrics&quot;, {})&#10;        ml_progress = runtime.get(&quot;ml_progress&quot;, {})&#10;        print(&#10;            &quot;  runtime trades=&quot;&#10;            f&quot;{int(_safe_float(metrics.get('trades', 0), 0))}, &quot;&#10;            f&quot;win_rate={_safe_float(metrics.get('win_rate', 0.0), 0.0) * 100:.2f}%, &quot;&#10;            f&quot;pf={_safe_float(metrics.get('profit_factor', 0.0), 0.0):.2f}&quot;&#10;        )&#10;        if ml_progress:&#10;            print(&#10;                &quot;  runtime brier_50=&quot;&#10;                f&quot;{ml_progress.get('brier_rolling_50')}, &quot;&#10;                f&quot;expectancy_30={ml_progress.get('expectancy_usd_30')}, &quot;&#10;                f&quot;max_dd={ml_progress.get('max_drawdown_pct')}&quot;&#10;            )&#10;&#10;&#10;def parse_args() -&gt; argparse.Namespace:&#10;    p = argparse.ArgumentParser(description=&quot;Diagnose adaptive bot ML learning quality&quot;)&#10;    p.add_argument(&quot;--state-file&quot;, default=&quot;learning_state.joblib&quot;, help=&quot;Path to persisted bot state file&quot;)&#10;    p.add_argument(&#10;        &quot;--status-url&quot;,&#10;        default=&quot;http://127.0.0.1:7860/status/full&quot;,&#10;        help=&quot;Optional bot status endpoint for runtime snapshot&quot;,&#10;    )&#10;    p.add_argument(&quot;--skip-status&quot;, action=&quot;store_true&quot;, help=&quot;Skip HTTP status fetch&quot;)&#10;    p.add_argument(&quot;--json&quot;, action=&quot;store_true&quot;, help=&quot;Print full report as JSON&quot;)&#10;    return p.parse_args()&#10;&#10;&#10;def main() -&gt; int:&#10;    args = parse_args()&#10;&#10;    if not os.path.exists(args.state_file):&#10;        print(f&quot;State file not found: {args.state_file}&quot;)&#10;        return 1&#10;&#10;    payload = joblib.load(args.state_file)&#10;    status_payload = None if args.skip_status else _fetch_status(args.status_url)&#10;    report = build_report(payload, status_payload=status_payload)&#10;&#10;    if args.json:&#10;        print(json.dumps(report, indent=2))&#10;    else:&#10;        print_report(report)&#10;    return 0&#10;&#10;&#10;if __name__ == &quot;__main__&quot;:&#10;    raise SystemExit(main())&#10;" />
+            </PendingDiffInfo>
+          </value>
+        </entry>
+      </map>
+    </option>
+  </component>
+</project>

.idea/workspace.xml

@@ -5,153 +5,7 @@
   </component>
   <component name="ChangeListManager">
     <list default="true" id="4d8c1aed-8693-45b5-a8e3-cfb3e1553feb" name="Changes" comment="">
-      <change beforePath="$PROJECT_DIR$/.env.example" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/.gitattributes" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/.gitignore" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/Dockerfile" beforeDir="false" afterPath="$PROJECT_DIR$/Dockerfile" afterDir="false" />
-      <change beforePath="$PROJECT_DIR$/GOOGLE_APPS_SCRIPT_FULL.js" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/README.md" beforeDir="false" afterPath="$PROJECT_DIR$/README.md" afterDir="false" />
-      <change beforePath="$PROJECT_DIR$/_backtest_hype.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/_backtest_ml_auto.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/_scan_coins.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/autonomous_bot.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/backtest.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/bot_config.json" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_APT_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_APT_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_APT_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_APT_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_APT_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_APT_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ARB_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ARB_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ARB_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ARB_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ARB_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ARB_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_AVAX_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_AVAX_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_AVAX_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_AVAX_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_AVAX_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_AVAX_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_BTC_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_BTC_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_BTC_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_BTC_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_BTC_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_BTC_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_DOGE_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_DOGE_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_DOGE_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_DOGE_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_DOGE_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_DOGE_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ETH_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ETH_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ETH_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ETH_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ETH_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_ETH_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_INJ_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_INJ_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_INJ_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_INJ_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_INJ_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_INJ_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_LINK_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_LINK_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_LINK_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_LINK_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_LINK_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_LINK_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_NEAR_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_NEAR_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_NEAR_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_NEAR_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_NEAR_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_NEAR_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_OP_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_OP_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_OP_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_OP_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_OP_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_OP_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SOL_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SOL_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SOL_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SOL_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SOL_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SOL_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SUI_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SUI_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SUI_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SUI_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SUI_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_SUI_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_TIA_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_TIA_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_TIA_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_TIA_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_TIA_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_TIA_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_WIF_h12_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_WIF_h12_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_WIF_h3_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_WIF_h3_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_WIF_h6_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/ml_auto_WIF_h6_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_APT_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_APT_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_ARB_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_ARB_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_AVAX_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_AVAX_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_BTC_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_BTC_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_DOGE_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_DOGE_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_DOT_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_DOT_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_ETH_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_ETH_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_INJ_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_INJ_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_LINK_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_LINK_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_NEAR_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_NEAR_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_OP_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_OP_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_SOL_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_SOL_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_SUI_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_SUI_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_TIA_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_TIA_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_WIF_long.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/models/scalp_WIF_short.joblib" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/optimize_ml_coins.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/regime.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/.gitkeep" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/ai_entry_decisions.jsonl" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/autonomous_backtest_report.json" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/funding_arb.log" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/funding_monitor.log" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/ml_auto_backtest_seed.json" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/reports/ml_auto_memory_BTC.jsonl" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/run_all_backtests.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/setup.sh" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/strategy_bb_reversal.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/strategy_conservative.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/strategy_hype.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/strategy_ml_auto.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/strategy_scalp.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/strategy_zlema.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/tests/.gitkeep" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/tests/test_autonomous_bot.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/tests/test_hbot_copy_execution.py" beforeDir="false" />
-      <change beforePath="$PROJECT_DIR$/track_ml_learning.py" beforeDir="false" />
+      <change beforePath="$PROJECT_DIR$/.idea/copilotDiffState.xml" beforeDir="false" />
     </list>
     <option name="SHOW_DIALOG" value="false" />
     <option name="HIGHLIGHT_CONFLICTS" value="true" />

README.md

@@ -33,6 +33,8 @@ This project now includes an adaptive ML bot that uses your existing Hyperliquid
 - `hbot.py`: existing Hyperliquid API framework and order placement helpers
 - `adaptive_bot.py`: new adaptive ML paper/live-gated trading bot
 - `smoke_test.py`: offline persistence and learning smoke test
+- `ml_diagnostics.py`: offline + live learning diagnostics report
+- `diagnostics_smoke_test.py`: quick smoke test for diagnostics logic
 - `paper_report.json`: runtime status and metrics output (created at runtime)
 
 ## Health endpoints (for HF + UptimeRobot)
@@ -61,12 +63,36 @@ pip install -r requirements.txt
 python smoke_test.py
 ```
 
+Optional: run diagnostics smoke test:
+
+```bash
+python diagnostics_smoke_test.py
+```
+
 3. Start the adaptive paper bot:
 
 ```bash
 python adaptive_bot.py --interval 15m --poll-seconds 45
 ```
 
+4. Check ML learning quality from saved state:
+
+```bash
+python ml_diagnostics.py --state-file learning_state.joblib
+```
+
+If the bot is running locally, include live status snapshot:
+
+```bash
+python ml_diagnostics.py --state-file learning_state.joblib --status-url http://127.0.0.1:7860/status/full
+```
+
+JSON output (useful for storing artifacts in CI/HF workflows):
+
+```bash
+python ml_diagnostics.py --state-file learning_state.joblib --json
+```
+
 Optional knobs for early growth from scratch:
 
 ```bash
@@ -189,6 +215,21 @@ At a high level, this bot uses online learning so it improves continuously while
 
 So each closed paper trade directly trains the next decision, and training survives restarts.
 
+## ML diagnostics
+
+`ml_diagnostics.py` helps answer whether the model signal is actually improving.
+
+It reports:
+
+- overall/per-coin/per-side win rate and expectancy
+- directional sanity checks:
+  - long winners should have higher `p_long` than long losers
+  - short winners should have lower `p_long` than short losers
+- direction-aware Brier score (lower is better)
+- optional runtime snapshot from `/status/full` including rolling Brier/expectancy/drawdown
+
+This is useful to catch label/specification issues where the model keeps repeating failing trade patterns.
+
 ## Notes
 
 - The consistency gate currently requires:

diagnostics_smoke_test.py

@@ -0,0 +1,63 @@
+#!/usr/bin/env python3
+"""Smoke test for ml_diagnostics report builder."""
+
+from __future__ import annotations
+
+import tempfile
+from pathlib import Path
+
+import joblib
+
+from ml_diagnostics import build_report
+
+
+def main() -> None:
+    payload = {
+        "closed_trades": [
+            {
+                "coin": "BTC",
+                "side": "long",
+                "entry_ts": 1000,
+                "pnl_usd": 5.0,
+                "entry_p_long": 0.72,
+            },
+            {
+                "coin": "BTC",
+                "side": "short",
+                "entry_ts": 2000,
+                "pnl_usd": -3.0,
+                "entry_p_long": 0.81,
+            },
+            {
+                "coin": "ETH",
+                "side": "short",
+                "entry_ts": 3000,
+                "pnl_usd": 4.0,
+                "entry_p_long": 0.22,
+            },
+        ],
+        "trade_events": [
+            {"event": "open", "coin": "BTC", "entry_ts": 1000, "p_long": 0.72},
+            {"event": "open", "coin": "BTC", "entry_ts": 2000, "p_long": 0.81},
+            {"event": "open", "coin": "ETH", "entry_ts": 3000, "p_long": 0.22},
+        ],
+    }
+
+    with tempfile.TemporaryDirectory() as td:
+        state_path = Path(td) / "learning_state.joblib"
+        joblib.dump(payload, state_path)
+        loaded = joblib.load(state_path)
+
+    report = build_report(loaded)
+
+    assert report["state_samples"]["closed_trades"] == 3
+    assert "overall" in report
+    assert "directional_sanity" in report
+    assert report["brier_direction_aware"] is not None
+
+    print("diagnostics_smoke_test: OK")
+
+
+if __name__ == "__main__":
+    main()
+

ml_diagnostics.py

@@ -0,0 +1,308 @@
+#!/usr/bin/env python3
+"""Offline and live diagnostics for adaptive bot learning quality.
+
+This script helps validate whether the model is learning directionally useful signals.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import math
+import os
+from collections import defaultdict
+from typing import Any, Dict, List, Optional, Tuple
+from urllib.error import URLError
+from urllib.request import urlopen
+
+import joblib
+
+
+def _safe_float(v: Any, default: float = 0.0) -> float:
+    try:
+        return float(v)
+    except Exception:
+        return default
+
+
+def _mean(values: List[float]) -> Optional[float]:
+    return (sum(values) / len(values)) if values else None
+
+
+def _trade_p_long(trade: Dict[str, Any], open_probs: Dict[Tuple[str, int], float]) -> float:
+    if "entry_p_long" in trade and trade["entry_p_long"] is not None:
+        return float(max(0.0, min(1.0, _safe_float(trade["entry_p_long"], 0.5))))
+    key = (str(trade.get("coin", "")), int(_safe_float(trade.get("entry_ts", 0), 0)))
+    return float(max(0.0, min(1.0, open_probs.get(key, 0.5))))
+
+
+def _summarize_basic(trades: List[Dict[str, Any]]) -> Dict[str, float]:
+    n = len(trades)
+    wins = sum(1 for t in trades if _safe_float(t.get("pnl_usd", 0.0), 0.0) > 0)
+    pnl = sum(_safe_float(t.get("pnl_usd", 0.0), 0.0) for t in trades)
+    return {
+        "trades": float(n),
+        "wins": float(wins),
+        "win_rate": (wins / n) if n else 0.0,
+        "total_pnl_usd": pnl,
+        "expectancy_usd": (pnl / n) if n else 0.0,
+    }
+
+
+def _directional_sanity(trades: List[Dict[str, Any]], open_probs: Dict[Tuple[str, int], float]) -> Dict[str, Optional[float]]:
+    longs = [t for t in trades if str(t.get("side", "")).lower() == "long"]
+    shorts = [t for t in trades if str(t.get("side", "")).lower() == "short"]
+
+    long_win = [_trade_p_long(t, open_probs) for t in longs if _safe_float(t.get("pnl_usd", 0.0), 0.0) > 0]
+    long_lose = [_trade_p_long(t, open_probs) for t in longs if _safe_float(t.get("pnl_usd", 0.0), 0.0) <= 0]
+    short_win = [_trade_p_long(t, open_probs) for t in shorts if _safe_float(t.get("pnl_usd", 0.0), 0.0) > 0]
+    short_lose = [_trade_p_long(t, open_probs) for t in shorts if _safe_float(t.get("pnl_usd", 0.0), 0.0) <= 0]
+
+    # For a healthy long-probability model:
+    # - long winners should have higher p_long than long losers
+    # - short winners should have lower p_long than short losers
+    return {
+        "long_winners_avg_p_long": _mean(long_win),
+        "long_losers_avg_p_long": _mean(long_lose),
+        "short_winners_avg_p_long": _mean(short_win),
+        "short_losers_avg_p_long": _mean(short_lose),
+    }
+
+
+def _direction_aware_brier(trades: List[Dict[str, Any]], open_probs: Dict[Tuple[str, int], float]) -> Optional[float]:
+    scores: List[float] = []
+    for t in trades:
+        side = str(t.get("side", "")).lower()
+        if side not in {"long", "short"}:
+            continue
+        p_long = _trade_p_long(t, open_probs)
+        won = 1.0 if _safe_float(t.get("pnl_usd", 0.0), 0.0) > 0 else 0.0
+        # Target for long probability:
+        # - long trade: y=1 if win
+        # - short trade: y=1 if short loses (long would've won)
+        y_long = won if side == "long" else (1.0 - won)
+        scores.append((p_long - y_long) ** 2)
+    return _mean(scores)
+
+
+def _calibration_table(
+    trades: List[Dict[str, Any]], open_probs: Dict[Tuple[str, int], float], side: str
+) -> List[Dict[str, Any]]:
+    bins: Dict[int, List[Dict[str, Any]]] = defaultdict(list)
+    for t in trades:
+        t_side = str(t.get("side", "")).lower()
+        if t_side != side:
+            continue
+        p = _trade_p_long(t, open_probs)
+        decile = min(9, max(0, int(math.floor(p * 10.0))))
+        bins[decile].append(t)
+
+    rows: List[Dict[str, Any]] = []
+    for decile in range(10):
+        arr = bins.get(decile, [])
+        if not arr:
+            continue
+        n = len(arr)
+        wins = sum(1 for t in arr if _safe_float(t.get("pnl_usd", 0.0), 0.0) > 0)
+        avg_p = _mean([_trade_p_long(t, open_probs) for t in arr])
+        rows.append(
+            {
+                "decile": decile,
+                "range": f"{decile / 10:.1f}-{(decile + 1) / 10:.1f}",
+                "trades": n,
+                "avg_p_long": avg_p,
+                "win_rate": (wins / n) if n else None,
+            }
+        )
+    return rows
+
+
+def _open_prob_lookup(events: List[Dict[str, Any]]) -> Dict[Tuple[str, int], float]:
+    out: Dict[Tuple[str, int], float] = {}
+    for e in events:
+        if e.get("event") != "open":
+            continue
+        coin = str(e.get("coin", ""))
+        entry_ts = int(_safe_float(e.get("entry_ts", 0), 0))
+        p_long = float(max(0.0, min(1.0, _safe_float(e.get("p_long", 0.5), 0.5))))
+        out[(coin, entry_ts)] = p_long
+    return out
+
+
+def _fetch_status(url: str) -> Optional[Dict[str, Any]]:
+    try:
+        with urlopen(url, timeout=3) as response:
+            if response.status != 200:
+                return None
+            body = response.read().decode("utf-8")
+            return json.loads(body)
+    except (URLError, TimeoutError, json.JSONDecodeError, ValueError):
+        return None
+
+
+def build_report(payload: Dict[str, Any], status_payload: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
+    closed = payload.get("closed_trades", []) or []
+    events = payload.get("trade_events", []) or []
+    open_probs = _open_prob_lookup(events)
+
+    by_coin: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+    by_side: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+    for t in closed:
+        by_coin[str(t.get("coin", "unknown"))].append(t)
+        by_side[str(t.get("side", "unknown")).lower()].append(t)
+
+    per_coin = {coin: _summarize_basic(trades) for coin, trades in sorted(by_coin.items())}
+    per_side = {side: _summarize_basic(trades) for side, trades in sorted(by_side.items())}
+
+    report: Dict[str, Any] = {
+        "state_samples": {
+            "closed_trades": len(closed),
+            "trade_events": len(events),
+        },
+        "overall": _summarize_basic(closed),
+        "per_coin": per_coin,
+        "per_side": per_side,
+        "directional_sanity": _directional_sanity(closed, open_probs),
+        "brier_direction_aware": _direction_aware_brier(closed, open_probs),
+        "calibration": {
+            "long": _calibration_table(closed, open_probs, "long"),
+            "short": _calibration_table(closed, open_probs, "short"),
+        },
+        "runtime_status": status_payload,
+    }
+
+    # High-level warnings to speed up triage.
+    warnings: List[str] = []
+    ds = report["directional_sanity"]
+    lw = ds.get("long_winners_avg_p_long")
+    ll = ds.get("long_losers_avg_p_long")
+    sw = ds.get("short_winners_avg_p_long")
+    sl = ds.get("short_losers_avg_p_long")
+
+    if lw is not None and ll is not None and lw <= ll:
+        warnings.append("Long directional signal is inverted or weak: long winners do not have higher p_long than long losers.")
+    if sw is not None and sl is not None and sw >= sl:
+        warnings.append("Short directional signal is inverted or weak: short winners do not have lower p_long than short losers.")
+
+    overall_wr = report["overall"].get("win_rate", 0.0)
+    overall_exp = report["overall"].get("expectancy_usd", 0.0)
+    if len(closed) >= 50 and overall_wr < 0.45:
+        warnings.append("Win rate below 45% after 50+ trades; model may be mis-specified or overtrading.")
+    if len(closed) >= 50 and overall_exp < 0:
+        warnings.append("Negative expectancy after 50+ trades; consider reducing exploration and revisiting labels.")
+
+    report["warnings"] = warnings
+    return report
+
+
+def print_report(report: Dict[str, Any]) -> None:
+    print("=== Adaptive Bot ML Diagnostics ===")
+    samples = report["state_samples"]
+    print(f"Closed trades: {samples['closed_trades']}")
+    print(f"Trade events: {samples['trade_events']}")
+
+    overall = report["overall"]
+    print(
+        "Overall: "
+        f"n={int(overall['trades'])}, "
+        f"win_rate={overall['win_rate'] * 100:.2f}%, "
+        f"total_pnl={overall['total_pnl_usd']:.2f}, "
+        f"expectancy={overall['expectancy_usd']:.4f}"
+    )
+
+    print("\nBy side:")
+    for side, s in report["per_side"].items():
+        print(
+            f"  {side}: n={int(s['trades'])}, win_rate={s['win_rate'] * 100:.2f}%, "
+            f"total_pnl={s['total_pnl_usd']:.2f}, expectancy={s['expectancy_usd']:.4f}"
+        )
+
+    print("\nBy coin:")
+    for coin, s in report["per_coin"].items():
+        print(
+            f"  {coin}: n={int(s['trades'])}, win_rate={s['win_rate'] * 100:.2f}%, "
+            f"total_pnl={s['total_pnl_usd']:.2f}, expectancy={s['expectancy_usd']:.4f}"
+        )
+
+    ds = report["directional_sanity"]
+    print("\nDirectional sanity:")
+    print(
+        "  Long winners avg p_long: "
+        f"{ds['long_winners_avg_p_long'] if ds['long_winners_avg_p_long'] is not None else 'n/a'}"
+    )
+    print(
+        "  Long losers avg p_long: "
+        f"{ds['long_losers_avg_p_long'] if ds['long_losers_avg_p_long'] is not None else 'n/a'}"
+    )
+    print(
+        "  Short winners avg p_long: "
+        f"{ds['short_winners_avg_p_long'] if ds['short_winners_avg_p_long'] is not None else 'n/a'}"
+    )
+    print(
+        "  Short losers avg p_long: "
+        f"{ds['short_losers_avg_p_long'] if ds['short_losers_avg_p_long'] is not None else 'n/a'}"
+    )
+
+    brier = report.get("brier_direction_aware")
+    print(f"\nDirection-aware Brier (lower is better): {brier if brier is not None else 'n/a'}")
+
+    warnings = report.get("warnings", [])
+    if warnings:
+        print("\nWarnings:")
+        for w in warnings:
+            print(f"  - {w}")
+
+    runtime = report.get("runtime_status")
+    if runtime:
+        print("\nLive status snapshot:")
+        metrics = runtime.get("metrics", {})
+        ml_progress = runtime.get("ml_progress", {})
+        print(
+            "  runtime trades="
+            f"{int(_safe_float(metrics.get('trades', 0), 0))}, "
+            f"win_rate={_safe_float(metrics.get('win_rate', 0.0), 0.0) * 100:.2f}%, "
+            f"pf={_safe_float(metrics.get('profit_factor', 0.0), 0.0):.2f}"
+        )
+        if ml_progress:
+            print(
+                "  runtime brier_50="
+                f"{ml_progress.get('brier_rolling_50')}, "
+                f"expectancy_30={ml_progress.get('expectancy_usd_30')}, "
+                f"max_dd={ml_progress.get('max_drawdown_pct')}"
+            )
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(description="Diagnose adaptive bot ML learning quality")
+    p.add_argument("--state-file", default="learning_state.joblib", help="Path to persisted bot state file")
+    p.add_argument(
+        "--status-url",
+        default="http://127.0.0.1:7860/status/full",
+        help="Optional bot status endpoint for runtime snapshot",
+    )
+    p.add_argument("--skip-status", action="store_true", help="Skip HTTP status fetch")
+    p.add_argument("--json", action="store_true", help="Print full report as JSON")
+    return p.parse_args()
+
+
+def main() -> int:
+    args = parse_args()
+
+    if not os.path.exists(args.state_file):
+        print(f"State file not found: {args.state_file}")
+        return 1
+
+    payload = joblib.load(args.state_file)
+    status_payload = None if args.skip_status else _fetch_status(args.status_url)
+    report = build_report(payload, status_payload=status_payload)
+
+    if args.json:
+        print(json.dumps(report, indent=2))
+    else:
+        print_report(report)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
+