How it works
Everything on this site is automated, impersonal, and reproducible from public data. This page is the complete specification — no black box.
What we publish
Once a day at 00:20 UTC, the model posts one call per tracked coin: higher or lower over the day, with a confidence between 50% and 65%. The Kraken price at post time is frozen into the row. That's the whole product — no price targets, no trade sizing, no advice.
The model (version mom-1.0)
A deliberately small momentum blend, computed from the last 120 daily Kraken candles:
· Momentum over 3, 10, and 30 days, each measured in units of the coin's own daily
volatility so coins are comparable.
· Blended 50/30/20 — short momentum dominates at a one-day horizon.
· Dampened when recent volatility spikes above trend: momentum is less trustworthy in
chaos, so confidence shrinks toward a coin flip.
· Confidence is capped at 65%. A one-day crypto call never deserves more, and we'd
rather under-claim than over-claim.
There is no machine learning and no curve fitting. The constants are published in the model version, and any retune bumps the version — you can see exactly which calls came from which model.
Grading
Every call is graded automatically at 00:10 UTC the next day against the Kraken daily candle close (the candle that closes at 00:00 UTC):
· Win — the close moved at least 0.1% in the called direction from the posted price.
· Loss — it moved at least 0.1% the other way.
· Push — the move was smaller than 0.1% either way; noise, not a call.
· Void — no close could be observed by 06:00 UTC (exchange outage). Rare, and it
counts as neither win nor loss.
The measurement window runs from the 00:20 UTC post to the next 00:00 UTC close — about 23 hours 40 minutes. Both prices are stored in the row, so every grade can be re-checked against public data.
Calibration — the honesty mechanism
A model that says "60%" should be right about 60% of the time. The record page buckets every graded call by claimed confidence and shows the actual hit rate next to it, plus a single "pulse" number: actual minus claimed. Around zero is healthy. If it drifts persistently negative, the model is overconfident and gets retuned — publicly, with a version bump noted here.
What this is not
VerdictCrypto is a publication, not an adviser. The calls are automated, identical for every reader, and never tailored to anyone's circumstances. We don't manage money, we don't know your situation, and nothing here is a recommendation to buy or sell anything. In liquid markets a coin flip wins about half the time — the record page exists so you can see precisely how much (or little) the model adds, before you care about it.