A LITTLE MORE CONTEXT
Know what the number means.
likely.sh turns defined market questions into probabilities you can reference alongside your own execution methods.
A fixed question.
Every study identifies a frozen reference range, a measurement window and a deadline. “High touched” and “high only” are different questions: high only also requires the low to remain untouched.
A completed minute.
Estimates use information available at the end of each completed bar. The clock is explicit. Upcoming targets stay hidden until their reference period ends.
A defined outcome.
The 30-minute studies ask whether the previous candle’s high, low, or midpoint will be touched. Each is a separate Yes/No question; several can resolve Yes. The eight-hour and New York range studies use four exclusive outcomes: neither, high only, low only, and both.
Evidence, with limits.
Every estimate requires 100 distinct historical windows. Unsupported states remain unavailable. A high probability does not specify the path price takes, an entry, a stop or a profitable trade.
A live market, a fixed question
This MVP focuses on Bitcoin. All studies follow live BTCUSDT spot on Binance. The price streams continuously; probabilities and touch decisions update only after a complete one-minute candle. Windows open and resolve automatically.
Estimates use a frozen model trained through December 2024. Reference ranges and prior touches use recent completed candles retained by the live engine. Missing evidence pauses an estimate; it is never treated as a miss.
Display times default to your device’s local timezone. Use the Local / UTC switch above to change them; your choice is remembered. Study boundaries remain fixed, including across daylight-saving changes.
The 15-minute studies open at :00, :15, :30 and :45 UTC and ask whether the previous 15-minute high or low will be touched by the deadline.
The 30-minute studies open at :00 and :30 UTC, every day. Targets are the preceding completed candle’s high, low, and midpoint: (high + low) ÷ 2. Each estimate uses at least 100 separate historical days at the same UTC half-hour, with one matching minute selected per day.
The three eight-hour UTC studies use the previous eight-hour range. The New York study measures weekdays 09:30–16:00 local time against the same date’s midnight–09:30 range, including exchange holidays.
Daily and weekly classic pivots use the previous complete UTC day or Monday–Sunday UTC week. P, R1–R3 and S1–S3 are independent touch questions, each expiring at the current period’s end. Daily matches count distinct days; weekly matches count distinct weeks. Historical tables use 15-minute daily and 60-minute weekly elapsed-time buckets. Use the timeframe filter to choose the duration that suits your decision.
Targets use minute high–low evidence from the instrument’s configured source. A provided zero-trade candle advances time but establishes no touch. Missing or conflicting evidence cannot be silently treated as a miss.
The five short-window BTC targets and eleven supported daily/weekly pivot targets have frozen logistic-regression models. Weekly S2, R3 and S3 ML remain unavailable because training or calibration lacks enough independent outcomes. Experimental ML comparisons appear alongside historical estimates when complete, timely inputs are available. Live evaluation records predictions before their outcomes; historical improvements do not establish trading profitability.