Learn
Trust · 4 min read
Why we'd rather show nothing
Stale, blocked, untrusted, missing. The four states that stop a pick from shipping — and why a silent fallback is the worst answer a model can give.
Most predictive systems quietly degrade. A data source goes stale, a feature drifts, an upstream API rate-limits — and the model just shrugs and returns whatever it can scrape together. The user sees a number that looks like every other number, and trusts it the same way.
Veritas does the opposite. If the readiness gate fails, the pick doesn't ship at all, and the surface that would have shown it explicitly reports why.
The four states
- Stale — the price moved more than our band since the model snapshot.
- Blocked — a required upstream source returned an error or no data.
- Untrusted — a sanity check inside the pipeline failed (e.g. a player listed as out is being projected for full minutes).
- Missing — the market isn't carried by the books we know about for this user's region.