We test whether crypto is predictable over minutes — and publish what fails.
Most trading content shows you the wins. This shows everything: the hypotheses we registered before we had the data, every prediction the engine made, every outcome, and the studies that killed our own best ideas.
The engine is right 68.3% of the time. The market was right 69.7%.
M6E reads order flow and momentum and, five minutes into every 15-minute Bitcoin cycle, commits to a direction. Over 597 forward predictions it called direction correctly 68.3% of the time.
That number sounds good. It is not an edge — and why is the most useful thing here. At the moment the engine committed, the prediction market's own price already implied 69.7%. We were not beating the market. We were tracking it, slightly behind, and paying fees for the privilege.
We write down what would prove us wrong — before we have the data.
Every engine is registered here first with a numeric gate it must clear and the sample size required. The register is timestamped and append-only. Once a gate is published in advance, a failed result cannot be quietly reinterpreted as a success.
| Hypothesis | Registered | Gate to promote | n | Result | Status |
|---|---|---|---|---|---|
| M6E-COMMITTED-v1all committed tiers · minute 5 | 2026-07-03 | accuracy ≥ 0.7178 at n ≥ 100 | 597 | 0.683 | FAILED |
| M6E-E3-v1highest-conviction tier only | 2026-07-06 | edge lower bound > 0 | 152 | −0.036 | RUNNING |
The first hypothesis failed its own gate. We publish that, because the alternative — moving the goalposts after seeing the data — is how most trading claims on the internet are manufactured.
No. Accuracy minus price minus fees was negative in every tier. The market had already priced in what the model knew.
No — worse in every configuration tested. On a high-probability binary the losing side drifts then gaps, so a stop triggers on noise or arrives far too late.
No — the dip is a loss detector, not a discount. Dips ≥4c appear in 51–55% of winners but 78–95% of losers, three to five times deeper.
Not this one. It converges at the same speed as the order book, not ahead of it. Tested commitment at minutes 1–5; no horizon cleared the price with confidence.
No. Look-ahead bias — it priced a late, matured signal at an early, cheap price. We found it ourselves. It is the reason this site exists.
Strikingly often. Across every probability band, win rate matched price. A contract at 70c wins about 70% of the time — the bar everything else must clear.
Order-flow tooling is normally the expensive tier — footprint charts, liquidation maps and real volume delta sit behind €40–200/month subscriptions, or at minimum a signup form. We already run this infrastructure for our own research, so we are opening it up.
Cumulative volume delta
Real taker buy/sell imbalance from a millisecond trade tape — measured, not estimated from candles.
Live trade tape
Every print as it lands, filterable by size, so large orders are visible in real time.
Liquidation feed & map
Forced liquidations as they fire, and where leverage is stacked.
Footprint & volume profile
Volume at price inside each candle.
Funding · OI · basis
What perpetual traders are paying to hold their positions.
Cross-venue divergence
Where venues disagree on price, in real time.
Pre-registration
Hypotheses and their pass/fail gates are published before results exist, so they cannot be adjusted afterwards.
Every call, including losses
The full ledger — wins, losses, abstentions — downloadable as CSV. Not a highlight reel.
Always against the market
Every accuracy figure is shown beside what the market's price implied at the same instant.
Published method
The feature definitions behind the engine are being open-sourced, so the work can be reproduced and criticised.
Nothing for sale
No signals, no subscriptions, no private group, no affiliate links. There is no account to create.
Stated sample sizes
Every number carries its n and date range. A result without a sample size is not a result.