Negative results · forward-tested

The graveyard. Where trading ideas go to be proven or die.

Finance publishes winners and buries the rest. We publish the burials: every hypothesis this desk runs is settled on the sealed forward book, and a cohort whose 95% interval sits below zero at n≥30 gets a headstone — with the data that killed it.

Derived, not written. Verdicts are computed from the same settled trades in the ledger and the hash chain on /verify. Educational research, not advice.

No burials yet

Nothing falsified yet — the book is young.

A verdict either way needs n≥30 settled trades per cohort. Cohorts below the bar are listed as sampling — no claims, no burials, until the data arrives.

ENTRYNO EDGE SHOWN

Gate-passing momentum names beat a passive hold over two weeks, net of costs.

n=67 · hit 63% (CI 51–73%)
mean +0.65%/trade · 95% CI −0.3 to +1.6%

Indistinguishable from zero at 95% — the honest reading is "no demonstrated edge yet".

EARLY COILNO EDGE SHOWN

Coiled bases flagged before breakout resolve upward within the hold window.

n=31 · hit 68% (CI 50–81%)
mean +0.51%/trade · 95% CI −0.6 to +1.7%

Indistinguishable from zero at 95% — the honest reading is "no demonstrated edge yet".

WINNERSAMPLING

Names clearing every gate plus the premium filter beat the base ENTRY cohort over a two-week hold.

n=22 · hit 50% (CI 31–69%)
mean −1.39%/trade · 95% CI −5.0 to +2.2%

Still sampling — n=22, no verdict until n=30.

Method: cohorts are the engine's own verdict bands, settled at the recorded exits, equal notional. Mean-return intervals are 95% t-intervals on raw settled trades; hit rates carry Wilson intervals. A falsified cohort is never deleted or re-run under a friendlier spec — see the post-mortem → for the narrative and the chain → for the receipts.