Anyone can post picks. A signal worth the name carries a fire timestamp, a model probability, the market price at that exact moment, the side — and later, a graded outcome nobody edited. Here is the full standard, and our record measured against it.
Verify the record →The fire timestamp. When the alert went out — provably before the outcome was known.
The model probability. What our win-probability model read from the live game state at that second.
The Kalshi price at fire. What the market was charging for the same side at the same moment. Without this, no signal can ever be evaluated.
The side. The specific team or contract the model pointed at. Not “watch this game” — a falsifiable call.
The graded outcome. Filled in at settlement — win or loss — and never touched again.
An 89% model read lost — that is what 89% means. The log even holds a 100% model read that lost: May 30, DET@CWS, model at certainty on Detroit with Kalshi at 79¢, and the White Sox walked it off 4–3. A record with no entries like these is not a record.
Every fired signal enters the public dataset before the outcome is known. The row exists while the game is still live.
The outcome field is filled at settlement. Win or loss, from the settled Kalshi market — not from our opinion of what “should” have counted.
Nothing is edited and nothing is removed. No re-grades, no “that one didn't count,” no quietly deleted losers.
One dataset, three views. The same file renders /results/, the CSV download, and the homepage tiles. The numbers cannot disagree with each other, because there is only one source.
Under those rules, the MLB dataset currently holds 630 logged signals across 429 games. Counting the first signal per game: 350 wins, 79 losses — 81.6%, which means about 1 in 5.4 games is a loss — and losses sometimes arrive in clusters. The BTC 15-minute dataset runs under the same rules: 92.8% across 8,233 resolved windows under the published filter — roughly 1 in 14 loses.
Grading also surfaces things a highlight reel would hide. At fire, the median gap between model probability and the Kalshi price is 12.5 points, and the median price of the picked side is 80¢. Break the record down by gap size and you get this:
| Gap at fire | Signals | Win rate | Loss rate |
|---|---|---|---|
| 10–15 points | 320 | 83.8% | 16.2% |
| 15–20 points | 82 | 73.2% | 26.8% |
| 20+ points | 27 | 81.5% | 18.5% |
First signal per game, all V2.2 MLB signals. Note what this does not show: a wider gap is not a safer signal. When the market disagrees with the model by more, part of that disagreement is information the model doesn't see — injuries, bullpen state, weather. Why the gaps exist at all is the subject of /kalshi-lag/; the only edge we claim is speed.
About 10 minutes after each signal fires, a locked post goes up in our public Telegram channel — game identified, side hidden. At settlement, the post is revealed: the side, the price at fire, and the graded outcome, losses included. Telegram stamps the post time; we can't backdate it, and we can't delete a reveal without the gap being obvious. The channel has run this way since June 13, 2026.
This means you can watch the record build in real time without paying anything. If the reveals ever stopped matching the log, that would be visible to everyone. That is the point.
Locked ~10 minutes after fire, revealed at settlement, every loss shown. No payment, no email, no signup.
Watch @DegenHedgeProof →Proof the call existed before the result did — locked posts, an append-only log, any third-party stamp. Unverifiable records tend to be flattering ones.
Every service wins in its own screenshots. Ask where the losses live. If you can't click through to a complete, sortable list of them, walk away.
A CSV you can open, sort, and recount yourself beats any claim on a landing page. Recounting is the entire point.
The same call at 60¢ and at 90¢ are completely different events. A pick without the market price at the moment it fired cannot be evaluated at all.
“81.6% over 429 games, about 1 in 5.4 losing” and “90% over ten picks” are not the same claim. If the sample size is missing, assume it's small.
Run this checklist on us first: /results/, results.csv, and @DegenHedgeProof exist so you can. How the model itself works — inputs, filters, and known failure modes — is documented at /method/. Then run the checklist on anyone else selling Kalshi signals.
Every signal can lose. About 1 in 5.4 MLB first-signals has lost, and losses arrive in clusters — a run of five losses is possible at this loss rate. If a losing streak like that at your chosen stake would hurt, the stake is too big.
Hit rate is not profit: your entry price, sizing, and discipline decide your actual outcome, and the price you get is usually not the price at fire. The model measures one specific phenomenon — a short lag between live game state and the Kalshi price. It is not an independent predictor, and it is not advice to bet. Past performance never guarantees future results. This is entertainment for adults 18+ who already trade on Kalshi.
Read the full model documentation →