Almost nobody in finance does this: we wrote down a testable prediction and a strict pass/fail rule in advance, set a public deadline, and locked our daily scores in Bitcoin so we can't quietly change them. A bet we could genuinely lose — watched in the open.
The claim — pre-registered 2026-07-06
Framler's top-scored stocks out-return its bottom-scored stocks.
Rank all 1,000+ stocks by our score, then compare the realised return of the top 20% vs the bottom 20%. If the score is real, the top beats the bottom — repeatedly, beyond chance.
PRE- REGISTEREDON RECORD2026-07-06
Does the top beat the bottom?
Daily top-minus-bottom return · 7-day horizon · 123 published days
NOT YET PROVEN
Average gap
+0.35%
Days top won
70/123
Independent windows
25
Each bar is one published day: above the line our top-scored stocks beat the bottom-scored, below they lost. It's mixed — 70 up, 53 down — tilted slightly positive (dashed line = +0.35% average). A real edge would sit clearly above zero; this doesn't yet.
Is that tilt real, or luck? · 95% confidence interval25 indep. windows
0%
The interval still crosses zero (lower bound -0.25%). Until it clears zero, the tilt could be luck, not skill — so the verdict stays NOT YET PROVEN. We're not claiming an edge; we're testing one in public, and we'll say plainly if it fails. Recomputed from live data every 30 min; significance uses non-overlapping windows only. The independent windows are counted back from the newest matured day, so shifting that end date by a day re-draws the whole set. Of the 7 alignments of this same history, 1 clear zero. The pre-registered rule fixes one of them and we do not get to pick — but it is the honest weight to put on any reading before 2027-01-01.
30-day horizon · the definitive test
PROVEN — SO FAR
+2.36%
Clears the bar on 5 independent windows. One independent window is one month here, so only ~13 exist by 2027-07-01 — the deadline can expose a large gap, not certify a small one.
First verdict
2027-01-01 · in ~87 days
~37 independent windows expected by then
The rules can't move — that's what makes it a bet
Fixed 2026-07-06 · published · Bitcoin-anchored
Fixed in advance
The claim, the 95% pass mark and the deadline were set on 2026-07-06. No changing the rule after seeing the result.
Scientific bar
A pass needs the gap to be 95% statistically certain — not "looks positive". No cherry-picked windows.
Recomputable
The headline gap is the plain average over every published day. Recompute it yourself, get our exact number. Significance is tested only on non-overlapping windows, because overlap fakes certainty.
₿ anchored
The daily scores feeding the test are hashed into Bitcoin in our signal ledger — we can't rewrite a past score to help the bet.
Public either way
At each deadline we stamp the verdict — PROVEN or NOT PROVEN — here and in the ledger. A failure gets said out loud, not buried.
What a pass proves — and what it doesn't. It proves our score ranks stocks better than chance. Not profit, not investment advice, not beating the market after costs. Proof of honesty and ranking skill — nothing more. And what a failure won't prove. Each deadline arrives with a fixed budget of independent windows — one per horizon length, so roughly 37 at 7 days and 13 at 30. That is enough to expose a large gap and not enough to certify a small one, so a NOT PROVEN verdict will mean the gap was never demonstrated at this sample size — never that we measured the edge to be zero. We report the bar we actually cleared, in both directions. Follow the maturing record on track record.
If you came here to poke holes
The strongest objections we know — answered straight
“Your own track record page says the calls are roughly 50/50 so far. Why does this bet matter?”
Because we claim honest measurement, not a proven edge. The 50/50 figure is our own published number, graded by a fixed rule. This page is the pre-registered instrument that settles whether the ranking skill is real: a fixed bar, a public deadline, and a verdict either way.
“A +0.35% average gap is tiny — after trading costs it's nothing.”
Correct — and no trading claim is being made. A pass proves the score ranks stocks better than chance (top-vs-bottom quintile, the standard academic yardstick), not that a strategy survives costs. Honesty and ranking skill; nothing more.
“25 independent windows is far too few for a 95% verdict.”
Agreed — that is exactly why the verdict above reads NOT YET PROVEN instead of something friendlier. Roughly 37 independent windows will exist by 2027-01-01; until the interval clears zero, "not proven" stays on the page.
“If the bet fails, you’ll quietly delete this page.”
The daily scores feeding it are hashed into Bitcoin via OpenTimestamps, and anyone can download the day-files and proofs from the ledger. Copies exist that we cannot recall — deleting the page wouldn’t unpublish the record, it would only prove the failure louder. The deadline verdict gets stamped here and in the ledger either way.
“Why should anyone trust your arithmetic?”
Don’t trust it — recompute it. The headline gap is a plain average of the daily top-minus-bottom return over every published day; significance is tested only on non-overlapping windows, because overlapping ones manufacture certainty. Same public data, same fixed rule, same number to the digit.