Methodology
The order matters: a name has to survive each layer before it reaches you.
1 · Universe
Every rule runs against a named sleeve rather than the whole market, so a result is always interpretable — you know what population it came from.
| Sleeve | Purpose |
|---|---|
| Long-List | Primary long sleeve for the mean-reversion and momentum scans. |
| future_shorts | F&O names screened for short setups. No overlap with the long sleeve. |
| indices | Index underlyings, used for the index watchlist. |
| Fundamentally-Strong | Quality reference list the long sleeve is derived from. |
2 · Fundamentals filter
Screens run against point-in-time snapshots — the figures as they were known on the date in question, not today's restated numbers. This is the difference between a backtest that means something and one that quietly reads the future.
3 · Technical signals
Rules are expressed in a small declarative language over multiple timeframes (15-minute, hourly, daily, weekly, monthly). Because rules are data rather than code, the same definition drives the live scan and the backtest — they cannot drift apart.
4 · Confirmation overlays
A technical trigger on its own is weak. Overlays — relative strength against the index, options positioning, market breadth, regime — are blended into an ensemble score. The components that contributed are shown alongside the score; when an overlay has no data it is dropped and the score reflects that, rather than a neutral value being invented to fill the hole.
5 · Proof state
Each rule carries a status, and a passing backtest is deliberately not enough to call something proven:
| Stage | Question |
|---|---|
| Backtest | Did it work on the data it was designed against? |
| Out-of-sample | Does it survive data it never saw — walk-forward, per fold? |
| Forward | Does it work on data that did not exist when the rule was written? |
Only the third is real evidence, and it takes time that cannot be shortcut.
Questions
What is point-in-time fundamental data and why does it matter?
Point-in-time data is the figures as they were known on the date being tested, rather than today's restated numbers. Using restated data lets a backtest read information that was not available at the time, which inflates results. NSE Desk screens against date-keyed snapshots to avoid this look-ahead bias.
What timeframes do the rules use?
Rules are expressed over 15-minute, hourly, daily, weekly and monthly bars. Because rules are stored as data rather than code, the same definition drives both the live scan and the backtest, so the two cannot drift apart.
What is the ensemble score?
A blend of the technical trigger with confirmation overlays such as relative strength against the index, options positioning and market regime. The contributing components are listed with the score. When an overlay has no data it is dropped rather than replaced with a neutral value.
What does 'candidate, not proven' mean?
It means the rule has not yet cleared out-of-sample and forward testing. Backtest, out-of-sample and forward are tracked separately, and only forward results on genuinely unseen data are treated as real evidence.
Which universes are scanned?
Long-List for the long mean-reversion and momentum scans, future_shorts for short setups, and indices for the index watchlist. Each rule is pinned to one sleeve so results are always interpretable.
For educational and informational purposes only — NOT investment advice. Outputs are unvalidated research candidates, not trade recommendations. Trade execution is disabled by design. See the full disclaimer.