Why KELLY can be stronger here
KELLY is designed to win by being scoreable: every forecast should connect to source evidence, causal assumptions, falsifiers, settlement, calibration, replay, and memory lineage.
why it mattersA forecast platform is only as good as its ability to be wrong in public, adjust from reality, and improve the next forecast without hiding the chain.
sector edgeShows environmental shocks, lag hypotheses, forecast calibration, and observed outcomes.
best claim boundaryKELLY should be called better only where its scorecards, calibration, replay receipts, and source coverage outperform alternatives. This page is built to make that measurable.
Where the proof lives
proofSource plan: the approved feed rows on this page
proofForecast method: the doctrine timeline on this page
proofAuthority evidence: Authority Index and replay receipts
proofScore evidence: settlement and calibration records
proofSystem evidence: Brain Fleet, generality evidence, and advanced proof labs
Professional layout pattern
The lab layout follows successful product patterns: fast first read, watchable objects, citations beside claims, source freshness, score history, and direct proof links.
patternWeather map pattern: layers for rain, wind, temperature, pressure, air quality, and alerts.
patternLocation pattern: local status first, regional context second, historical closure third.
patternCitation-forward answer pattern: put sources and proof next to claims.
patternForecasting track-record pattern: keep probability, resolution rule, score, and history together.
patternWatchlist pattern: let users inspect a saved set without writing ground truth.
patternAlert/status pattern: show what changed, what is stale, and what KELLY is watching.
Research references
TradingViewCharts, alerts, screeners, replay, multi-chart layouts, and fast command/search patterns make market work feel immediate. MetaculusForecast questions need clear resolution rules, track records, scoring, calibration, and aggregation history. Windy / Windy.appWeather users need maps, layers, local points, alerts, history, model comparison, and fast location context. PerplexityAnswer users expect citations beside claims, follow-up context, and checkable source trails.
KELLY's specialized forecast method
Location, public weather observations, forecast-vs-actual, pressure changes, anomaly history, closure and memory eligibility.
1Approved source intake only
2Tag source, freshness, domain, and authority
3The Book of Why: ask what caused the change, what would have happened otherwise, and what evidence can falsify the causal story
4Thinking in Systems: map feedback loops, delays, stocks, flows, constraints, and downstream effects
5Thinking, Fast and Slow: separate fast pattern recognition from slower deliberate verification
6Superforecasting: start with base rates, update incrementally, keep probabilities explicit, and score the forecast later
7Thinking in Bets: treat every forecast as a wager against reality with a clear settlement rule
8Build causal pressure nodes and forecast cones
9Separate KELLY forecast from user forecast
10Track falsifiers, uncertainty, and competing explanations
11Close against reality
12Score calibration and error direction
13Apply continuous micro-adjustments after settlement to improve future accuracy
14Create replayable memory evidence when eligible
What this lab is for
Demonstrating causal hypotheses without financial or betting advice baggage.
safe claimShows environmental shocks, lag hypotheses, forecast calibration, and observed outcomes.
boundaryNo emergency instruction replacement and no certainty from sparse sensors.
Live proof status
proof score43 / early
visible cells0
readiness score0%
sources wired5/7
public writesblocked
Proof links
Every location view should show source time, observed variables, forecast target, closure rule, and replay receipt.
Approved source plan
| source | status | cost | role |
|---|
No simulated source
synthetic-none |
wired |
none |
The engine can run manual ticks without any external API key. |
Open-Meteo
open-meteo |
wired |
free_no_key |
Current public weather observations and forecasts. Best default proof demo. |
Open-Meteo Air Quality
open-meteo-air-quality |
wired |
free_no_key |
No-key air-quality observations for PM2.5, PM10, ozone, nitrogen dioxide, carbon monoxide, and UV where coverage exists. |
National Weather Service
nws |
available |
free_no_key |
US forecast grid and alerts. Requires a responsible User-Agent string, not an API token. Outside U.S. coverage returns no coverage. |
OpenAQ
openaq |
available |
free_key_limited |
Optional air quality observations. Coverage varies by location. |
NOAA CO-OPS Water Levels
noaa-water |
wired |
free_no_key |
Water-level/tide observations when a station is configured. |
OpenAI
openai |
wired |
paid_optional |
Only for narrative phrasing. The proof engine does not require an LLM. |
User questions this lab can inspect
| question lane | plain question | node |
|---|
| Rain |
Is it going to rain? |
WEATHER:RAIN_NOW |
| Snow |
Is it going to snow? |
WEATHER:SNOW_NOW |
| Hail |
Is hail a risk? |
WEATHER:HAIL_RISK |
| Wind |
Will wind be a problem? |
WEATHER:WIND_SPEED |
| Air quality |
Is air quality getting worse? |
AQ:PM25 |
| Heat |
Is heat a risk? |
WEATHER:TEMP_2M |
| Overall risk |
Is weather risk rising? |
WEATHER:RISK |
Default focuses
| focus | meaning | context |
|---|
| New York |
rain and city weather risk |
40.7128, -74.006 |
| Miami |
rain and storm risk |
25.7617, -80.1918 |
| Dallas |
storm and heat risk |
32.7767, -96.797 |
| Austin |
storm and heat risk |
30.2672, -97.7431 |
| Atlanta |
storm and humidity risk |
33.749, -84.388 |
| Boston |
coastal weather risk |
42.3601, -71.0589 |
| Denver |
hail and storm risk |
39.7392, -104.9903 |
| Seattle |
rain and wind risk |
47.6062, -122.3321 |
Causal pressure map
WEATHER:TEMP_2MWEATHER:HUMIDITYWEATHER:PRESSUREWEATHER:WIND_SPEEDWEATHER:WIND_GUSTWEATHER:PRECIP_NOWWEATHER:RAIN_NOWWEATHER:SNOW_NOWWEATHER:HAIL_RISKAQ:PM25AQ:PM10AQ:OBSERVATION_DENSITY
causal linkWEATHER:PRESSURE -> WEATHER:RISK
causal linkWEATHER:WIND_SPEED -> WEATHER:RISK
causal linkWEATHER:WIND_GUST -> WEATHER:RISK
causal linkWEATHER:PRECIP_NOW -> WEATHER:RISK
causal linkWEATHER:RAIN_NOW -> WEATHER:RISK
causal linkWEATHER:SNOW_NOW -> WEATHER:RISK
causal linkWEATHER:HAIL_RISK -> WEATHER:RISK
causal linkWEATHER:HUMIDITY -> WEATHER:PRECIP_NOW
causal linkAQ:PM25 -> WEATHER:RISK
causal linkAQ:PM10 -> WEATHER:RISK
causal linkAQ:OBSERVATION_DENSITY -> AQ:PM25
Boundary receipt
{
"readOnly": true,
"publicUserWriteAllowed": false,
"noAdviceSubstitute": "No emergency instruction replacement and no certainty from sparse sensors.",
"noClaimOfPerfectPrediction": true,
"proofRequiredForEveryForecast": true
}