Work with us

Bring a forecast. Test the decision.

We test whether better evidence improves your decision, and lock the rules before the answer exists.

  1. 01

    decision

  2. 02

    forecast panel

  3. 03

    reference

  4. 04

    deadline

  5. 05

    score

  6. 06

    stop / go

01 · three pilot paths

Pick what you want measured.

Each pilot fixes its scope, score and outcome source before it starts. Only the first path rests on measured capability today.

available

Forecast evaluation

Compare your forecasting panel with strong, established baselines.

scope
one panel, one target
evaluation
proper score vs equal average and a tree baseline, fixed in advance
outcomes
an independent source you both accept
timeframe
4–8 weeks of resolved outcomes
delivery
scored report + reproducible receipts
research direction

Failure intelligence

Test whether warnings catch costly failures at a fixed review budget.

scope
flag the riskiest 10% of forecasts
evaluation
share of group failures caught vs random
outcomes
independent, timestamped after each warning
timeframe
until about 100 group failures resolve
delivery
sealed warnings, scored later
research direction

Investigation optimization

Research whether extra evidence improves specific decisions.

scope
one decision with a known cost of error
evaluation
decision outcome with and without investigation
outcomes
your recorded decisions and results
timeframe
agreed per pilot
delivery
research memo, labelled as such

02 · receipt

One chain survives scrutiny.

Research, provenance, model, seal and delivery remain distinct.

  1. 01

    Offdiagonal

    records what was knowable

  2. 02

    Zeno

    learns when trust is earned

  3. 03

    Check · API · MCP

    delivers the same answer

  4. 04

    Arena

    seals, resolves and scores

  5. 05

    Zeno30

    adds separate forecasts

03 · where it already works

Your field, measured.

Real results from finished, judged runs, all 24 fields.

every field

Warnings beat every rival in 5 of 24 fields on the later check.

  1. RSVHealth+24%n 2k · 3k outcomes
  2. Philadelphia Fed (points)Economy+23.8%n 31k · 4k outcomes
  3. ForecastBenchGeneral forecasting+23.1%n 182k · 109k outcomes
  4. SeaSpace and sea+23%n 19k · 12k outcomes
  5. ECBEconomy+16.8%n 12k · 1k outcomes
  6. CryptoMarkets and crypto+13.4%n 50k · 84k outcomes
  7. Weather vs analysisAtmosphere+10.5%n 49k · 49k outcomes
  8. FluHealth+9%n 2k · 2k outcomes
  9. Weather vs airportsAtmosphere+6.3%n 22k · 22k outcomes
  10. COVID-19Health+1.1%n 8k · 6k outcomes
  11. Sea-level pressureAtmosphere0%n 227k · 1.1M outcomes
  12. WindAtmosphere0%n 227k · 1.1M outcomes
  13. Urban heatAtmosphere0%n 31k · 154k outcomes
  14. Prediction marketsMarkets and crypto0%n 13k · 66k outcomes
  15. Single-run weatherAtmosphereNot measured · too few misses to judgen 31k · 23k outcomes
  16. Air qualityAtmosphereNot measured · too few misses to judgen 132 · 644 outcomes
  17. Farming rainAtmosphereNot measured · too small to judgen not recorded · 31 outcomes
  18. Ensemble weatherAtmosphereNot measured · outcomes arrivingn not recorded · 0 outcomes
  19. Space weatherSpace and seaNot measured · too few misses to judgen 19k · 26k outcomes
  20. Market tradersMarkets and cryptoNot measured · too few misses to judgen 11k · 4k outcomes
  21. Cross-venue marketsMarkets and cryptoNot measured · too few misses to judgen 132 · 325 outcomes
  22. Philadelphia Fed (ranges)EconomyNot measured · too small to judgen not recorded · 2k outcomes
  23. Energy demandEnergyNot measured · too few misses to judgen 534 · 3k outcomes
  24. NuminousGeneral forecastingNot measured · too small to judgen not recorded · 154 outcomes

Practice test · historical: development and later confirmation slices of past data · n 937,644 confirmation rows

evidence +
Uncertainty · not measureddocs/v082/fields.json · open data ↗

04 · start

What do you forecast?

Bring the decision, available forecasts, and consequence of being confidently wrong.