Crypto · 2025-09-01
coinbase-btc-usd-1h: value at 2025-09-01 18:00 UTC
- History behind it
- 48 past records
- Warning
- 58%
Surprise Intelligence · a model of forecast fragility
Is this agreement
earned or fragile?
When many forecasts agree, Zeno checks whether that agreement is earned or fragile. It never changes your forecast without proof, and always says why: trust, widen, adjust or look again.
Real case · Weather · 24 hours ahead
Sydney temperature, 5 ensemble models
ECMWF
21.2° · 88% risk · held
GEM
26.5° · missed
ICON
29.0° · missed
NCEP
21.6° · held
UKMO
27.2° · 66% risk · missed
ZENO
90% range
20.0°24.0°28.0°32.0°
forecasts
26.5°
Zeno
25.1°
actual
22.2°
Zeno flagged the two confident models most likely to miss. One did miss, and one held. Zeno moved toward the answer.
01 · why it matters
Confident forecasts fail without warning.
Every forecast says how sure it is; Zeno learns which forecasters fail together, then must prove that knowing this improves decisions. Here is one real case, start to finish.
01 · Forecasts arrive
26.5°
5 weather models forecast Sydney's temperature 24 hours ahead. Their middle answer is 26.5°.
02 · Zeno reads track records
2 flagged
ECMWF 88% · UKMO 66% chance of a confident miss, based on how each model has failed before.
03 · Zeno answers
25.1°
An adjusted forecast with a 90% range of 21.1–29.1°, plus the reason for each flag.
04 · What happened
22.2°
The panel was off by 4.3°; Zeno by 2.9°. Zeno's range held. Of the flagged models, UKMO missed; ECMWF held — a false alarm, shown too.
one frozen example · not an averagewarnings measured on ensemble weather only
02 · how it works
It learns when confidence is earned.
Zeno remembers how each model has failed before. It adjusts only where that memory has been proven on later weeks, and otherwise keeps your forecast.
In
Your forecasts
any models, with or without ranges
Memory
Their track record
only outcomes known before the decision
Gate
Has Zeno earned it here?
tested on later weeks it never saw
yes · adjust
Adjusted forecast, 90% range and miss risk per model.
no · keep yours
Your forecast returned unchanged, with the reason.
03 · what Zeno caught
Real warnings, real misses, false alarms too.
Sealed weeks the model never saw. Every case is shown with what really happened.
real misses per 100 alerts
62
random alerts would catch
≈15
share of all confident misses caught
33%
ensemble weather · sealed confirmation weeks10% review budgetserved model v0.57
Crypto · 2025-12-11
coinbase-btc-usd-1h: value at 2025-12-11 15:00 UTC
- History behind it
- 48 past records
- Warning
- 60%
Crypto · 2025-10-07
coinbase-btc-usd-1h: value at 2025-10-07 21:00 UTC
- History behind it
- 48 past records
- Warning
- 54%
Research replay from frozen run v044-h200-20261004T1335Z, not live Arena questions. Confirmation period, main model seed 11, continuous targets of judged sources; cases at the 10th / 50th / 90th percentile of CRPS(candidate) - CRPS(reference). Pool: 8,654 cases; the rule picked three Bitcoin hours because crypto dominates the pool. The candidate is not gate-passed and is not what the API serves.
04 · the evidence
Tested on weeks it never saw.
Each source is judged on later periods. Where Zeno lost, the source's own forecast is kept.
Crypto price ranges
Ensemble weather
Deterministic weather
COVID-19 hospitalisations
Sea conditions
Prediction markets
RSV
FluSight
sourceheld-out gain over its referenceserved
Crypto price ranges
+11.7%
Ensemble weather
+5.6%
Deterministic weather
+6.2%
COVID-19 hospitalisations
+11.8%
Sea conditions
+3.2%
Prediction markets
kept
RSV
kept
FluSight
kept
Held-out gains are source-specific. They do not transfer automatically to a new domain.
05 · use it
Try it, call it, or connect it.
The page, the API and the MCP server return the same answer from the same published model.
curl -X POST https://zenodivergent.dev/api/public/v1/check \
-H 'content-type: application/json' \
-d '{"question": { ...your forecasts... }, "mode": "proven"}'06 · the ecosystem
Evidence becomes the next test.
Offdiagonal, Zeno, Zeno30 and Arena form one auditable loop; API and MCP deliver its answer.
- 01
Offdiagonal
Records forecasts, timing, histories and outcomes.
- 02
Zeno
Learns trust, corrections and shared failure patterns.
- 03
Check · API · MCP
Returns one answer, its range and its evidence.
- 04
Arena
Seals the answer before reality and scores it afterward.
- 05
Zeno30
Adds separately made forecasts and new investigations.
- 06
Resolved evidence
Improves the next test without rewriting the past.