I.60% avg fill
Gap Fill
What fraction of overnight gaps of size X get filled intraday? Bucketed by gap size and direction; filterable by day of week and expiry.
Buckets · 6Lookback · 180d
→
Asset
NIFTY
Every setup is a distribution. The question is which one, and how often it has paid.
Edgeful · India pre-computes the historical probabilities of recurring intraday setups — gap fills, opening-range breakouts, initial-balance breaks, prior-day pivots — across NIFTY, BANKNIFTY, and the ten most liquid NIFTY constituents. Stack them. Filter them. Trade only where the data agrees.
NIFTY · Gap Fill · last 180 trading days · by gap-size bucket
← higher fill · lower fill →
Probability Workbench
Pick an instrument. Pick a setup. Read the table.
Instrument
Report
Gap Fill · NIFTY
Gap Fill probabilities · — → —
≤ 40%
40–80%
≥ 80%
| Bucket | Direction | N | Fill rate | CI · 95% | Avg min | Med min | Recent 30d |
|---|
Live · 09:42 IST
Setup firing now
Setup
0.1–0.25% gap up
Hist. fill
76%
Recent decay
50% ⚠
The Differentiator
Probability stacking, not strategy backtesting.
Single setups have edges. Stacked setups have asymmetries. Combine two or three reports across the same lookback window and read the joint probability that all signals fired in the same direction — and what happened next.
Setup A
Gap Fill
76%
UP · 0.1–0.25% · n=25
×
Setup B
ORB Continuation
52%
UPSIDE breakout · n=88
=
Joint
Both aligned
81%
n=18 · CI 60–93% · last 30d 100%
The market doesn't reward conviction. It rewards conditional conviction — the willingness to wait for two or three uncorrelated edges to align in the same direction.
— House note · April 2026The Library
Six reports, each a self-contained edge.
Every report ships with instance count, raw probability, 95% Wilson confidence interval, and a recency check (last 30 days vs. full window) that flags decayed edges before you take them.
II.53% continuation
Opening Range Breakout
Define the opening range as the first 15 minutes. What fraction of days break that range, and of those, how many continue in the breakout direction by close?
OR window · 15mLookback · 180d
→
III.48% continuation
Initial Balance
The first hour's high and low. Same logic as ORB but on the wider window prop traders care about — the IB rejects more strongly, but breaks more decisively.
IB window · 60mLookback · 180d
→
IV.false-break heavy
Prior-day High / Low
When today breaks yesterday's high or low, does it tend to continue (breakout) or revert (false break)? Filterable by gap context.
Two pivotsLookback · 180d
→
V.open-relative
Session Bias
Given today's open relative to yesterday's close, what is the probability the session closes green vs. red? A simple, durable directional read.
4 quadrantsLookback · 180d
→
VI.timeframe-aware
Engulfing Reversals
On a chosen timeframe — 5m, 15m, or 1h — how often does a bullish or bearish engulfing candle actually mark the local reversal?
3 timeframesLookback · 180d
→
Methodology
Clean statistics. Transparent windows. No machine learning.
The whole point of this project is explainable edges. Every probability is reproducible from a clean clone of the repository — no black boxes, no hidden parameters.
Rolling windows. The default lookback is 180 trading days. Long histories overfit; the rolling window keeps the regime current.
Wilson confidence. Every probability ships with a 95% Wilson interval — narrow bands reflect both high frequency and high stability.
Recency check. Every row reports the last-30-day fill rate alongside the full window. Divergence is flagged with a ⚠.
Indian calendar. NSE holidays via pandas-market-calendars. No assumptions about Mon–Fri.
One source of data. Upstox Analytics token. 1-minute bars. No synthetic fallback when the API is down — the report fails loudly instead.
IST throughout. Every timestamp is in Asia/Kolkata internally; conversion happens only at the API boundary.