Live Dashboard
Recorded by the always-on worker. Updates automatically every few seconds.
Current 15-minute interval
Prediction
Experimental model: random-walk baseline from recent volatility plus a small momentum/news tilt. Not a guarantee.
BTC-USD
- Interval open
- —
- Open source
- —
- Price source
- —
- 1m volatility
- —
BTC price · 1-minute candles Interval open
Market prices · Polymarket
| Outcome | Market | Buy price | Sell price | Bot |
|---|---|---|---|---|
| ▲ UP | — | — | — | — |
| ▼ DOWN | — | — | — | — |
Market % is Polymarket's displayed price, not a fill price. Buy/sell prices are the best ask/bid in ¢ per $1 share; larger orders walk deeper into the book. Bot % is the model's own estimate. Polymarket resolves on Chainlink BTC/USD.
Order book depth
Model signal SIMULATED
Edge = bot probability − (buy price + slippage buffer + estimated fee): UP — · DOWN —
Simulated trading Paper trading
| Direction | Entry price | Current sell price | Net P/L % | Stop-loss price | Take-profit price | Status | |
|---|---|---|---|---|---|---|---|
| No simulated positions open. | |||||||
Fills are simulated against the live order book; no real orders or funds. Stop-loss and take-profit are % changes from the actual entry price, checked against the executable sell price. Net P/L includes fees.
Open a simulated position PAPER
Probability this interval EXPERIMENTAL
Recent intervals All history →
| Interval | Final call | P(UP) | Outcome | |
|---|---|---|---|---|
| No completed intervals yet | ||||
Outcome source: PM = official Polymarket resolution, Ref = Coinbase open/close (pending official result).
Prediction History
One prediction per 15-minute round: the direction frozen at minute 5, scored against Polymarket's official result. Later updates never replace it.
Entry price is Polymarket's buy price (best ask) for the predicted side, saved at minute 5. It's a market quote, not an actual trade fill, and doesn't include fees or slippage. Final payout is $1 per share if that side won the official result, $0 if it lost.
Follow every minute-5 pick Hypothetical
Hourly P&L Hypothetical
| Day / hour Los Angeles | Total predictions made | Wins | Losses | Net P&L ($) | Units |
|---|---|---|---|---|---|
| Loading… | |||||
| Round | Minute-5 prediction | Saved confidence | Entry price quote | Final payout | Official outcome | Result |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Performance
Computed only from recorded predictions and confirmed outcomes. Nothing is estimated or back-filled.
Early Prediction Accuracy EXPERIMENTAL
One frozen prediction per interval at the start and at minutes 1–5, each predicting the 15-minute close. Scored only against Polymarket's official result. Minute 5 is the main score, chosen in advance.
Accuracy by early checkpoint shaded = 95% confidence interval
Calibration by confidence
Early checkpoint results
| Checkpoint | Recorded | Missed | Coverage | Resolved | Correct | Incorrect | Pending | Accuracy | 95% CI | Market fav. | Brier | Sample |
|---|
Coverage = checkpoints frozen on time ÷ checkpoints due. A checkpoint counts only if recorded within 45 s of its minute mark; otherwise it is marked missed and never filled in. "No official result" intervals (Polymarket never settled within 3 h) are excluded from accuracy and listed separately.
Confidence buckets
| Stated confidence | Predictions | Won | Lost | Avg stated | Actual win rate | 95% CI | Verdict |
|---|
First five minutes: P(UP) path vs final outcome
Recent intervals, minute by minute
| Interval | 0m | 1m | 2m | 3m | 4m | 5m | Official |
|---|
Cells show the frozen call and P(UP). ✕ = checkpoint missed.
Rolling early accuracy (last 50 resolved intervals)
Chronological testing
Learning from results
Every incorrect minute-5 call gets a short review built only from what was recorded. Recorded facts are kept apart from possible causes. The model only changes when a refit beats it on later rounds it never saw, and you adopt it.
Incorrect minute-5 calls, reviewed
Patterns across wins and losses known at minute 5
| Group | When | Scored | Correct | Accuracy | 95% CI | Hypothetical P&L | Sample |
|---|
These are what the review lessons quote. Groups with fewer than 30 rounds are not evidence of anything; the 95% interval shows how much the true rate could differ.
Model versions
| Version | Status | Coefficients | Live scored rounds | Live accuracy | 95% CI | Hypothetical P&L |
|---|
Price-aware strategy vs following every pick SIMULATED · POLYMARKET US
A separate, pre-registered test. At minute 5 it bets UP, DOWN or nothing, using a calibrated probability and the price you could actually pay on Polymarket US (book depth + 1¢ slippage + taker fee). Only rounds that start after registration count. It never places orders; live auto-trading is unchanged.
Entry timing: minutes 1–5 STUDY · HYPOTHETICAL
Compares buying at minute 1, 2, 3, 4 or 5, and a rule fixed in advance that picks the entry minute (Adaptive A1). Each uses only the prediction and quote recorded at that minute. Auto-trading stays on minute 5 unless the pre-registered test passes on new rounds.
Later checkpoints (minutes 6–15)
Kept separate from early accuracy. Late calls are easy because little time is left, so they say little about entry quality.
| Checkpoint | n | Accuracy | 95% CI | Brier | Market acc. |
|---|
Scored only against the official Polymarket result. Rounds without one stay pending and are never graded on the Coinbase price.
Paper trading SIMULATED
Fills are simulated against the live order book: real average entry price, fees, partial fills. Positions exit on stop-loss / take-profit / manual close, or settle on the official result.
Paper-trading equity SIMULATED
Recent simulated positions SIMULATED · NO REAL ORDERS
| Opened | Origin | Side | Avg entry | Shares | Total cost | SL / TP | Exit | Avg exit | Net P&L | Status |
|---|
News
Headlines from public RSS feeds, scored with a transparent keyword lexicon.
Features used by the latest prediction
News enters the model only as a small drift tilt: weight × score, in units of remaining-time volatility. Set the weight to 0 in Settings to disable.
Feeds
Headlines
Auto-trading
One bet per 15-minute round on your chosen exchange (Polymarket US or Kalshi), following the frozen minute-5 prediction. Runs on the always-on backend. Starts in simulation; real money only after you turn it on.
Controls
Real money · Polymarket US
Real money ACTUAL FILLS
Simulation SIMULATED
Bets by round newest first
| Round | Mode | Pick | Status | Contracts | Avg price | Fee | Cost | Result | P&L |
|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||
P&L Calendar
Real-money results on Polymarket US: actual fills and confirmed settlements only, after fees, by settlement day in Los Angeles time. Simulated bets are not included.
Settings
Changes are saved to the database; the worker applies them on its next loop.