This is the public record of a systematic crypto research engine. Every hypothesis is written down with an explicit falsification criterion before it is tested. Every test result, including the ones that kill an idea, is appended to a hash-chained ledger. The ledger is verified on every rebuild, and the full outcome distribution is on this page.
Of 991 logged experiments, the overwhelming majority did not survive. That is the expected shape of an honest search: if most ideas passed, the gate would not be a gate.
| Verdict | Experiments | Share |
|---|---|---|
| discarded — did not beat the gate | 458 | 46.2 % |
| duplicate — already measured, no new evidence | 437 | 44.1 % |
| kept — beat the incumbent's gate | 64 | 6.5 % |
| baseline — reference run, not a hypothesis | 17 | 1.7 % |
| invalid — could not be evaluated | 14 | 1.4 % |
| inconclusive — not decidable at this sample size | 1 | 0.1 % |
The experiment ledger is append-only and every row carries the hash of the row before it and a hash of its own payload. Rebuilding this page re-walks the whole chain from the first row.
Each row below is a claim that was put to the test, followed by the criterion that was declared in advance as sufficient to reject it. “Qualified” means the candidate cleared every gate (matched random-entry null, multiplicity correction, out-of-sample, walk-forward). “Rejected” means it reached a full validation and did not.
| Claim under test | Runs | Qualified | Rejected |
|---|---|---|---|
| Breakouts in contracting volatility persist longer. Expectancy does not improve over the parent. | 30 | 0 | 0 |
| Breakouts are better when not already overbought. Fewer trades with no expectancy improvement. | 26 | 2 | 2 |
| Keine Einstiege in Volumen-Ausreisser. Die Volumenspitze markiert den Beginn der Bewegung, der Filter entfernt die Gewinner. | 25 | 1 | 1 |
| Breakouts that close near the bar high continue. No improvement in expectancy or halves. | 23 | 0 | 0 |
| A shorter hold captures the fast part of the move. Expectancy falls because costs are paid more often per unit of move. | 22 | 0 | 0 |
| Keine Einstiege in Explosionskerzen. Gerade die grossen Kerzen setzen sich fort; der Filter schneidet die beste Haelfte. | 21 | 0 | 0 |
| Nur nach einem Aufwaertstag einsteigen. Der Vortag ist irrelevant und der Filter kostet nur Trades. | 20 | 0 | 0 |
| Keine Einstiege nach einem Gap nach unten. Gaps nach unten werden systematisch aufgekauft, der Filter kostet die Erholung. | 18 | 1 | 1 |
| A wider stop survives noise without changing the thesis. Expectancy falls and drawdown rises. | 17 | 3 | 4 |
| Nur bis 10 % unter dem 20-Tage-Hoch handeln. Tiefe Ruecklaeufe liefern die besseren Einstiege, der Filter kostet Rendite. | 17 | 2 | 0 |
| A new 20-day high on a liquid EUR pair continues for up to ten days. Net expectancy after 80 bps taker fees is <= 0 on the training slice. | 17 | 0 | 0 |
| Taking profit at 15% improves risk-adjusted return. Expectancy falls because winners are cut. | 12 | 0 | 0 |
| A tighter stop improves risk-adjusted return. Stopped-out trades destroy more expectancy than they save. | 11 | 0 | 2 |
| Only take breakouts above the long-term mean. The filter removes the profitable half of the sample. | 11 | 0 | 2 |
A frozen version that is only evaluated on data after its freeze date is the only evidence that cannot be overfitted. There is exactly one version in that lane right now. It has produced no trades yet, and this page says so instead of showing a flat equity line as if it were a result.
| Version | Markets | Days live | Forward events |
|---|---|---|---|
| FSV-20261103-F1_trend_continuation_40_180_30d_v1-3f978114 | 84 | 3 | 0 |
total forward versions ever frozen: 22 · closed: 21 · all-time forward events: 580 · paper fills in the simulation ledger: 168
| Night | Status | Experiments | Kept | Frozen |
|---|---|---|---|---|
| 20261107 | OK | 8 | 0 | 0 |
| 20261106 | OK | 12 | 0 | 0 |
| 20261105 | OK | 10 | 0 | 0 |
| 20261104 | OK | 68 | 4 | 0 |
| 20261103 | OK | 68 | 4 | 0 |
| 20261102 | CHAIN_BROKEN | 13 | 0 | 0 |
| 20261101 | CHAIN_BROKEN | 36 | 1 | 0 |
| 20261031 | OK | 0 | 0 | 0 |
| 20261030 | OK | 22 | 1 | 0 |
| 20261029 | OK | 50 | 7 | 0 |
| 20261028 | OK | 93 | 8 | 2 |
| 20261027 | OK | 8 | 0 | 0 |
| 20261026 | OK | 0 | 0 | 0 |
| 20261025 | OK | 25 | 1 | 0 |
A candidate is only accepted if it clears four gates at once: (1) a matched random-entry null — same market, same volatility regime, same position age — at p < 0.05 after a Holm correction for the number of variants tried; (2) a deflated Sharpe ratio at or above 0.90, which penalises the number of trials; (3) out-of-sample and walk-forward consistency; (4) a full cost model with Kraken fees, spread and modelled slippage, so an edge that exists only before costs is not an edge. One entry per Vienna calendar day, no leverage, no shorting, no pyramiding. Every one of those rules is in the ledger, and the ledger is what this page renders.