Research
The Ledger
The specifications in our research record, each with its verdict and the gate that decided it. It is the record behind every number on this site, published so you can check it for yourself.
- 105,204
- specifications screened
- 562
- cleared the full gate stack
- 1 in 187
- the pass rate, measured not estimated
- 101,322
- listed below, one row each
Measured 28 October 2026. The figure only ever grows. A screen ordered through this site is counted in it, yours included, and never listed: your specification stays yours.
Search the ledger ↓What one row is
A row is one complete specification — a market, a signal, an entry clock, a stop, a target and a flatten time — run end to end against tick data and scored. Not an idea, a specification: the same idea with a 0.8 ATR stop and with a 1.2 ATR stop is two rows, because it is two tests, and both of them count.
Nearly every row carries one of three verdicts. Passed cleared every gate. Failed was stopped by one, and the deciding gate says which. Not advanced was measured but not taken on to the full stack.
Why publish it, and what is not listed
That is the whole point of publishing this. If you test enough specifications, some will look excellent by chance. Reading a winner beside the number of tests it came from is what tells you it is real, and that number is right here: 105,204 tests, 562 passes, 1 in 187.
101,322 of the 105,204 are listed below, one row per recorded test. The other 3,882 are counted and not listed: 102 are screens run on this site, which count in every total while each specification belongs to whoever ran it; the remaining 3,780 are the difference made by early research, entered after the fact as summary rows, some standing for a whole batch of tests. Pre-registrations and holdout reads, 23,930 of them, are in neither figure: writing a hypothesis down is not a test.
Roughly 68,807 of these are first-pass cuts: the cheap first look of a search, where a specification whose signal was too weak to matter (its t-statistic under a floor) stopped before any expensive test ran. They are real tests and they stay in the count, but they swamp the table, so there is a button to hide them. The most common deciding gate across everything is first pass: below the t >= 2 floor, at 32,509 rows.
No price, bar, tick or per-trade record appears here or leaves our research machine. What you are reading is our own parameter choices and the summary statistics each run produced.
That is the whole point of publishing this. If you test enough specifications, some will look excellent by chance. Reading a winner beside the number of tests it came from is what tells you it is real, and that number is right here: 105,204 tests, 562 passes, 1 in 187.
101,322 of the 105,204 are listed below, one row per recorded test. The other 3,882 are counted and not listed: 102 are screens run on this site, which count in every total while each specification belongs to whoever ran it; the remaining 3,780 are the difference made by early research, entered after the fact as summary rows, some standing for a whole batch of tests. Pre-registrations and holdout reads, 23,930 of them, are in neither figure: writing a hypothesis down is not a test.
Roughly 68,807 of these are first-pass cuts: the cheap first look of a search, where a specification whose signal was too weak to matter (its t-statistic under a floor) stopped before any expensive test ran. They are real tests and they stay in the count, but they swamp the table, so there is a button to hide them. The most common deciding gate across everything is first pass: below the t >= 2 floor, at 32,509 rows.
No price, bar, tick or per-trade record appears here or leaves our research machine. What you are reading is our own parameter choices and the summary statistics each run produced.
Screen your own idea through the same gates
The rows above came out of the research engine you can point at your own idea. You describe it in plain English and the desk turns it into an exact specification; it runs through the same gates, and it is counted in this ledger’s totals, so the count behind your report is always the real one. It is never listed here.