BTCUSDT · out-of-sample
Track record
5.5 years. 2.95 million minute bars. 36 pre-registered experiments. That is what the Confirmation Signal was built and tested on — and each of those experiments had the result that would have killed it written down before it ran, so a failure could not be quietly reinterpreted afterwards.
Everything in this panel is out-of-sample. Each signal was written into the journal the moment price arrived, with the grade it arrived under, and scored 60 minutes later — recorded before its outcome was known. Nothing here is a backtest and no figure is typed into this page; it is read from the journal at request time and recomputed nightly.
Real signals, real outcomes
Every chart below is a signal we published before price arrived and scored an hour afterwards — not a backtest. Up to three per grade, and each caption says where that one ranks inside its own grade, so you can see how cherry-picked it is. WIDE shows fewer because fewer qualified: we would rather show one than pad it. The full distribution they came from is the table underneath.
- Worst adverse
- 0.9 bp
- 0.01%
- Best in favour
- 152.0 bp
- 1.52%
- Held +60m
- 140.3 bp
- 1.40%
Top 4% of 257 TIGHT signals
- Worst adverse
- 0.1 bp
- 0.00%
- Best in favour
- 115.6 bp
- 1.16%
- Held +60m
- 23.6 bp
- 0.24%
Top 2% of 257 TIGHT signals
- Worst adverse
- 0.8 bp
- 0.01%
- Best in favour
- 75.4 bp
- 0.75%
- Held +60m
- 47.4 bp
- 0.47%
Top 4% of 257 TIGHT signals
- Worst adverse
- 0.8 bp
- 0.01%
- Best in favour
- 171.8 bp
- 1.72%
- Held +60m
- 171.8 bp
- 1.72%
Top 4% of 261 NORMAL signals
- Worst adverse
- 2.0 bp
- 0.02%
- Best in favour
- 275.3 bp
- 2.75%
- Held +60m
- 164.6 bp
- 1.65%
Top 7% of 261 NORMAL signals
- Worst adverse
- 0.0 bp
- 0.00%
- Best in favour
- 89.0 bp
- 0.89%
- Held +60m
- 50.0 bp
- 0.50%
Top 2% of 261 NORMAL signals
- Worst adverse
- 33.2 bp
- 0.33%
- Best in favour
- 207.4 bp
- 2.07%
- Held +60m
- 52.3 bp
- 0.52%
Top 59% of 310 WIDE signals
- Worst adverse
- 25.5 bp
- 0.26%
- Best in favour
- 147.7 bp
- 1.48%
- Held +60m
- 101.5 bp
- 1.01%
Top 52% of 310 WIDE signals
- Worst adverse
- 28.9 bp
- 0.29%
- Best in favour
- 158.7 bp
- 1.59%
- Held +60m
- 37.2 bp
- 0.37%
Top 55% of 310 WIDE signals
What each grade has been worth
| Grade | Median adverse move @15m | vs baseline | Held ≤20 bp | Median adverse move @60m | Touches | Window | Published? |
|---|---|---|---|---|---|---|---|
| TIGHT | 20.0 bp 0.20% | -3.3 bp | 50.0% | 34.6 bp 0.35% | 256 | 2026-07-23 → 2026-10-16 | yes |
| NORMAL | 25.6 bp 0.26% | +2.3 bp | 41.9% | 39.0 bp 0.39% | 260 | 2026-07-24 → 2026-10-16 | yes |
| WIDE | 24.9 bp 0.25% | +1.6 bp | 42.9% | 37.4 bp 0.37% | 310 | 2026-07-27 → 2026-10-15 | yes |
| ALL TOUCHES | 23.3 bp 0.23% | — | 45.5% | 35.9 bp 0.36% | 844 | 2026-07-23 → 2026-10-16 | yes |
The ALL TOUCHES row is the honest baseline — what a touch does when you know nothing about the tape. A grade is only worth the distance between its row and that one. A grade with fewer than 30 scored touches is shown here but is not quoted anywhere else on the site or in alerts.
The journal underneath
A touch cannot be scored until its full 60-minute window has elapsed, so the pending count is always non-zero on a live feed. Nothing is removed from this journal once written.
See the next signal as it fires.
The record above grows every night. The free plan puts the live grade in front of you on every touch — no card, no call.
Every minute the model publishes a bracket for the next hour. This is what happened to 20,099 of them — every forecast written in the last 14 days, scored against the price an hour later.
The band scored here is the one drawn before the move and left where it was. Nothing is refitted after the fact, and no forecast is dropped for being wrong — the only ones missing are those whose outcome bar never arrived.
What the band has been worth
The middle two tiles have to be read together. A band that always holds because it is a mile wide is worth nothing, and that shows up as a median move near zero — the width tile is what stops the coverage figure being bought with air. Window: 2026-10-04 → 2026-10-18.
Individual days
Not the best days — the day at each rank, and never all of one kind. Two weekday sessions and two weekend ones, because a weekend is a different market: thinner books, a slower pace, and a band that does not behave identically across the two. Each day is ranked against its OWN kind, so the label says how that day went rather than restating which market it fell in, and one of each pair sits below that kind's median by construction.
| Day | Session | Forecasts | Closed inside | Where it ranks |
|---|---|---|---|---|
| 2026-10-06 Tue | Weekday | 1,440 | 82.7% | better than 70% of the 10 weekdays |
| 2026-10-08 Thu | Weekday | 1,440 | 73.7% | better than 20% of the 10 weekdays |
| 2026-10-10 Sat | Weekend | 1,440 | 93.1% | better than 75% of the 4 weekend days |
| 2026-10-11 Sun | Weekend | 1,440 | 83.4% | better than 25% of the 4 weekend days |
The model behind it
The bracket is not a rule of thumb, a fixed percentage, or a multiple of last hour's range. It is the output of our own machine-learning modelling, re-run every minute on the live tape and updated as the market changes underneath it.