Zeno · forecast audits

Check the odds before you trust them.

Zeno checks your forecast against how thousands like it really ended.

One real forecast · Manifold · 7 days out
0%5%10%15%20%
Market 12.3%
Track record 7.6%
Ended: No
  1. Made
  2. Audited
  3. Saved
  4. Resolved
  5. Scored

Someone says 12%.

Track record n 4,397 · 95% 6.9%–8.5% · one case, not a score

240 real forecasts, from made to ended

0%50%100%7 DAYS OUTRESOLVED
━ 6 of 240 started ≥80% sure and landed the other wayn 240 real Manifold markets · a week before close · sample of 65,978

When many forecasts agree, Zeno checks whether that agreement is earned or fragile.

Audit yours.

You 20%

Your number stays yours.

Ledger— forecasts audited— in the last 24 hpublic records appear here when owners opt in

12% → 7.6%

Long shots run hot.

n 4,397 markets priced 10–15% · 95% range 6.9%–8.5% · Manifold · 65,978 in the curve

0%0%50%50%100%100%priced 0.00–0.02: resolved yes 0.3% (n=5389)priced 0.02–0.05: resolved yes 0.8% (n=7926)priced 0.05–0.10: resolved yes 3.1% (n=7910)priced 0.10–0.15: resolved yes 7.6% (n=4397)priced 0.15–0.25: resolved yes 15.1% (n=5618)priced 0.25–0.40: resolved yes 28.8% (n=6259)priced 0.40–0.60: resolved yes 49.2% (n=9276)priced 0.60–0.75: resolved yes 66.8% (n=5642)priced 0.75–0.85: resolved yes 81.3% (n=3967)priced 0.85–0.90: resolved yes 89.0% (n=2171)priced 0.90–0.95: resolved yes 94.7% (n=2784)priced 0.95–0.98: resolved yes 98.4% (n=2501)priced 0.98–1.00: resolved yes 99.4% (n=2138)7.6% happened
Price vs what happened · n=65,978 resolved markets · Manifold · 95% ranges

Real past case · Bitcoin · 1 Sep 2025

1 / 4

Where will Bitcoin be in an hour? Three forecasters agreed within $289.

3 FORECASTERSTHE CROWD · 90% SUREZENO · WIDEN ITactual $110,918

Widen it.

The price ran to $110,918, past everyone. Zeno's answer scored 12% better than the crowd's.

Earlier model replay, not a new assessment · n = 1 case · source: frozen replay record

Five people whose lines all trace back to one orange dot

01

Five voices. One rumour.

Agreement isn't evidence.

A polygraph needle that spikes once in orange

02

Zeno reads the needle.

Tested on unseen outcomes.

Three doors: open teal, wide ochre, barred orange

03

One next step.

Keep, check, or rethink.

Where crowds break together

Top weather models miss on the same day up to 3.4× more often than chance.

  • ecmwf ifs × ukmo global 20km3.4×
  • ecmwf ifs × icon3.0×
  • gem global × ukmo global 20km2.9×
  • ncep gefs × ukmo global 20km2.9×
  • icon × ukmo global 20km2.8×
  • gem global × icon2.6×

white line = chance (1×) · bracket = 90% interval

Ten people repeating one rumour sounds like consensus. It's one source. Five forecasters agreed. All five had read the same blog post.

Past record · 2026-09-12 to 2026-10-17 · n 43,490 outcomes per pair

evidence +
Uncertainty · 3.25–3.65 · 2.79–3.16 · 2.75–3.15 · 2.7–3.06 · 2.6–2.99 · 2.38–2.74download the record ↗
Zeno

crypto

299k

forecasts

panels
84k
outcomes
84k
state
in training

● correction tested · ● in training · · 2 held sources, outer ring

Less error than the crowd's own answer · held-out weeks · 95% range

  • +25%

    Crypto

    8–34% · n 428 days

  • +17%

    COVID-19

    9–23% · n 5,726 weeks · never trained on

  • +16%

    Energy demand

    13–18% · n 1,271 weeks

  • +15%

    Weather ensembles

    12–17% · n 6,542 days

  • +9%

    Weather history

    6–15% · n 286 weeks

Scope: worse on Sea state, Single-run weather — there Zeno hands your forecast back. Full record ↗