StockpickerAiStockpickerAi

๐Ÿงฎ Binary Exhaustion Lab

DEDICATED ยท ALGO + LLM ยท hybrid_v1
โ“ What is the Binary Exhaustion Lab? โ€” click to expand whitepaper ยท read

Dedicated, ground-up rebuild of the candle-counter screen reader. The Binary Exhaustion engine reads consecutive runs, MA stacking, distance-from-mean, ZigZag swings, and emits a directional probability for the next 1โ€“3 candles. An LLM optionally confirms or overrides. One armed session locks one symbol โ€” switching charts mid-stream cannot contaminate another session. Every observation can auto-fire a Monte-Carlo paper wager sized by per-bucket Wilson lower-bound ร— ยผ-Kelly.

Think of this as a betting workbench for short-term candle exhaustion. You point it at one symbol/timeframe, and on every "observation" it runs a deterministic candle-counting algorithm (plus an optional LLM second opinion) to predict where the next N bars are headed. Those predictions get auto-resolved against real price action so the lab learns its own calibrated win rate, not a textbook one.

  1. ARM SESSION โ€” locks the lab to one symbol (e.g. EUR/CHF, BTCUSD, NVDA). A unique DB index makes it physically impossible to have two active sessions on the same symbol, so your stats can never get contaminated by another chart.
  2. ANALYZE NOW โ€” pulls fresh OHLC bars and runs the Binary Exhaustion engine (consecutive candles, MA stacking, distance-from-mean, ZigZag swings). Output: direction + confidence % + reasoning. Optionally paste/capture a chart screenshot and the LLM gets to confirm or override.
  3. Auto-resolve โ€” N bars later (your "Horizon"), a background worker compares entry vs. close and stamps the prediction win / loss / flat.
  4. Calibration table โ€” outcomes get bucketed by the confidence you predicted (50โ€“60%, 60โ€“70%, โ€ฆ). The Wilson lower bound column tells you the conservative true win rate per band. Until any band hits โ‰ฅ50 samples, the table shows you it's not yet trustworthy.
  5. TAKE THE TRADE โ€” sizes a paper wager at ยผ-Kelly ร— Wilson p_lower, capped at 2% of bankroll. No edge โ†’ tiny 0.1% floor wager. Real edge โ†’ it ramps up automatically. Tick the "Auto-charge MC" box and every observation fires a paper trade on its own.

Field cheat-sheet:  Symbol = the only ticker this session will track.  Asset = routing hint (forex/crypto/stock/futures) so we hit the right OHLC provider.  TF = candle timeframe (1m/5m/15m/1h/4h/1d).  Horizon (bars) = how many bars ahead the prediction is graded on.  Bankroll $ = the simulated account size that the wager sizer scales from.

NO ACTIVE SESSION lab: โ€ฆ

๐Ÿ“ธ SCREEN READ

๐Ÿ“ธ Click to upload or paste (Ctrl+V).
or use Capture Screen to share a live chart tab.
$0.0000 ยท 0 tok ๐Ÿ“ฆ Binary samples: 0 / 50 ยท pฬ‚ โ€” ยท drift โ€”
No verdict yet. Arm a session, then capture a chart and Analyze. This panel will fill with the engine's direction call, confidence %, consecutive-candle count, MA stacking, ZigZag swing context, and (if enabled) the LLM pattern read โ€” exactly like the original Candlevision verdict card, just sturdier.

๐Ÿ’ฐ NEXT WAGER (MONTE CARLO)

Suggested size
$0.00
โ€” ยท โ€” of bankroll
basis: no signal yet
Waiting for a verdict. Until any confidence band reaches the calibration floor, wagers stay at 0.1% of bankroll.
Optional. With Auto-charge MC on (default), wagers fire automatically on every observation.

๐Ÿ“Š LIVE CALIBRATION (THIS SESSION)

Confidence bandnwinslosses pฬ‚pฬ‚ low (Wilson)armed?
no data yet

๐Ÿ“œ OBSERVATION FEED (THIS SESSION)

no observations yet

๐Ÿงพ PAPER WAGER LOG

whendirconfsizebasisstatus
no wagers yet