Signal Accuracy — How We Measure
We track every recommendation. We publish misses. Here's the methodology.
Last updated: —. Updated monthly.
Full breakdown by signal type →What we track
Every time CardMind generates a recommendation for a card, we record the card price at that moment and the recommendation type. Seven days and thirty days later, an automated process checks whether the price moved in the direction the signal indicated. That outcome — correct or incorrect — is recorded against the original recommendation.
What HIGH / MEDIUM / LOW confidence means
| Confidence | What it means | Historical accuracy |
|---|---|---|
| HIGH | 3+ signals aligned, strong pattern match | ~74%* |
| MEDIUM | 1–2 signals, mixed pattern | ~58%* |
| LOW | Insufficient signals to form a view | Not tracked separately |
* Placeholder — real data at Month 3.
Accuracy by signal type
| Signal type | Accuracy* | Notes |
|---|---|---|
| Buylist drop (CK) | 81% | Most reliable single signal |
| Volume anomaly | 67% | Stronger when combined with buylist |
| Ban window + meta | 71% | Weakest alone, strongest combined |
* Placeholder.
What we don't claim
CardMind does not predict bans. Wizards of the Coast announces bans without warning. No tool outside WotC has advance knowledge. CardMind detects the market behavior patterns that historically precede bans — buylist drops, volume spikes, tournament dominance — and surfaces that risk clearly. You decide what to do with it.
Why we publish this
Most tools don't publish accuracy data because it's uncomfortable to show misses. We think the opposite is true — showing misses is what makes the hits credible. If HIGH confidence signals are 74% accurate, that's meaningfully actionable. If we hid the misses, you'd have no reason to trust the 74%.