Zeno Check
Check any forecast in the browser or over the API.
- —Free to try
- —Receipt with every answer
- —Weather and crypto have track records
Forecast reliability · early access
Discover when the models you trust are missing something.
Zeno checks whether a forecast deserves trust. Spotting forecasts that fail together is being tested in public, before outcomes: see the Failure Watch.
- 23.1%
- less error than each forecaster's own track record on ForecastBench, first look at unseen questions
- 62%
- of Zeno's weather warnings were real misses, on unseen data
- 2,840,152
- forecast panels with known outcomes across 21 fields, time-audited for the next model
- 10,335
- forecasters on the SurpriseBench record across 24 fields — economists, health teams, AI labs, markets
Live check · served model
Should you trust this forecast?
Pick an example or paste your own. Every answer says whether Zeno corrected it or kept the reference.
Ensemble weatherReplay · outcome already known
Sydney temperature · 24 hours ahead
Zeno moved the panel estimate toward the outcome and flagged two narrow forecasts. UKMO missed; ECMWF's range held, so that one was a false alarm.
Served answer
Zeno correction
Advanced · edit the raw request
Edit the forecasts below (each needs a 90% range or a probability) and run. Untested cases return your forecast unchanged.
panel
26.5°CZeno
25.1°Cactual 22.2°C
19.8°C30.4°C
panel
26.5°C
Zeno
25.1°C
actual
22.2°C
UKMO66% · correct warning
ECMWF88% · false alarm
% = chance that forecaster misses its own 90% range · not a chance of “surprise” or profit
Past replay · already resolved
Six systems agreed. All six were wrong.
Replay of a past Santiago temperature forecast (research data, not the served model). Switch views to see what Zeno sees.
Real case · Santiago · 18 Oct 2026 · Temperature at 2 m
How warm did it get in Santiago that afternoon?
Every model predicts the future. Here, the future judges every model. · past case, frozen data, not a live warning
1 Predictions2 Consensus3 Dependencies4 Uncertainty5 Reality
Predictions
6 forecasts
Six ensemble systems, each with its own 90% range, for one hour in Santiago.
Replay · Santiago · 2 m temperature · 2026-10-18 · tap a curve
When many forecasts agree, Zeno checks whether that agreement is earned or fragile.
How it works
Three steps. One clear answer.
- 01
Send the forecasts
Point values, ranges or probabilities from your models, vendors or agents.
- 02
Zeno reads the panel
Track records, shared families and agreement that looks too easy.
- 03
Get a decision
Trust, widen the range, or send it for review, with the evidence attached.
Ways to work with Zeno
Start free. Prove it on your data.
Forecast audit
We measure your forecasts against strong baselines on your own history.
- —One panel, one target
- —Rules locked before results
- —Private report
Failure warnings
Test whether Zeno catches your costly misses at a fixed review budget.
- —Flag the riskiest 10%
- —Scored on misses caught
- —Stop/go decision at the end
Pilot pricing is set per scope. Ask us.
Who it's for
See it in your field.
- Weatherlive check
Do our 90% ranges really hold 90%?
see it on real data → - Tradingresearch
Which confident signals deserve a second look?
see it on real data → - Satellitesexperimental
When do space-weather forecasts fail together?
see it on real data → - Energyexploratory
When should a demand forecast be escalated?
see it on real data → - AI agentsresearch
Are my agents independent, or one view repeated?
see it on real data → - Investorsearly-stage
What is built, what is proven, what is next?
see it on real data →
Is this production software?+
No. Zeno is an early-stage model. The live check runs today; everything else is labelled research or pilot, with its limits stated.
Does it make trades or forecasts for me?+
No. It tells you when a forecast you already have deserves trust, a wider range, or a second look.
What do I need to send?+
Past forecasts with timestamps and the outcomes that followed. More history means a sharper answer.
Is my data shared?+
Pilot data stays private unless you choose public receipts.