Rule lab
Research tool · not investment advice. Full disclaimer →
Write a rule; get the honest answer.
Descriptive record only — past data and published method.
Compose
every choice is in the URLSource: closed vocabulary.
Submitting only changes the URL — the whole rule is the query string, so the result is shareable and reproducible.
Verdict
an example answersampleSource: demo · synthetic.
rule SELECT liquid500 WHERE not_extended RANK BY mom12 TAKE 25 HOLD quarterly
| net return/vol | 0.61 |
|---|---|
| gross return/vol | 1.02 |
| flat-cost return/vol | 0.84 |
| half 1 (2012-18) | 0.55 |
| half 2 (2019-26) | 0.66 |
| placebo p95The 95th percentile of the null distribution: the same book rebuilt N times with RANDOM picks (same universe, same size, same costs). A rule must beat this, not merely beat zero — luck has a high bar. source: the placebo machinery. | 0.40 |
| observed | 0.61 |
| empirical p | 0.02 |
| benchmark (Nifty-500 net) | 0.89 |
| max drawdown | -41.0% |
| annual cost % | 6.00 |
| turnover/rebalance | 0.38 |
| capacity breakpointThe AUM (₹ cr) at which participation-cost impact kills the edge — the largest AUM whose net return/vol still beats the benchmark. A rule with no stated capacity is not a result. | unstated |
prereg eb64d10999f899b4… · demo — synthetic numbers, not a real run
Synthetic example — these are not the numbers of a real run. Compose a rule above to check yours against the ledger.
Check your rule before a backtest
the same gauntlet we run on ourselvesSource: research ledger.
Pick a universe, a ranking signal, how many names and how often you rebalance. Before anything runs, we check your rule against the ledger of approaches already recorded dead. This form checks the specification and existing evidence; it does not run a new backtest. An untested rule remains untested.
The vocabulary is deliberately closed. You cannot express a rule we cannot honestly evaluate, which is why there is no free-text box here.
How the gauntlet judges
what a verdict has to surviveSource: published boundary.
- Both halves. A rule must beat the Nifty 500 net benchmark in 2012-18 and again in 2019-26; one good half is noise.
- The placebo. It must beat the 95th percentile of randomly selected books drawn from the same tables, not merely beat zero.
- Capacity. The breakpoint is the portfolio size at which trading impact erases the result; with no stated capacity there is no result.
- Registration first. The rule text is frozen and hashed before anything runs, and the first registration stands.
- Known-dead shapes. A rule matching a shape already recorded as dead is cited from the ledger before any run: a dead idea can be re-tested, never re-walked silently.