finiov0.1HISTORICAL
Historical classifier.
The original classifier estimates the chance of a positive return over five trading sessions.
One symbol · one event · five sessions
View v0.1 →Historical model card ↗FINIO MODEL / FINANCIAL PROBABILITY
One forecast.
Many financial
questions.
Finio turns market data into probabilistic forecasts you can question.
Generate one forecast for an asset and ask about returns, outperformance, drawdowns, volatility, and more, all from the same probabilistic view.
FINIO MODELv0.3 / BETA
NVDAONE FORECAST, MANY QUESTIONS
What is the chance NVDA beats SPY over 20 trading sessions?
EXAMPLE FORECASTHISTORICAL ARCHIVE
55.7%
Chance NVDA beats SPY over the next 20 trading sessions.
55.7%
About this example
SAME FORECAST: v03-NVDA-2026-07-07. The archive key is not a saved live forecast ID. Each answer is a recorded model probability, not an observed outcome. No per-query training. This preview makes no production requests.
TRY FINIO
Start with an asset.
Choose a model, then generate your forecast in the live workspace.
MODEL EVOLUTION
Three generations.
One direction: measurable progress.
From a fixed event classifier to reusable statistical forecasts and a learned distribution. Each generation expands the questions you can ask.
Model evolution, not a performance chart.
These are research estimates, not buy or sell advice. v0.3’s Beta label comes from its historical release criteria; it does not establish live predictive accuracy.
Open the original v0.1 live classifier
v0.1 / HISTORICAL BASELINE
Original v0.1 live model
A fixed five-trading-session positive-return estimate. Successful requests create a public record. Pending means no outcome has been recorded.
Use a US stock symbol such as AAPL or NVDA. Five sessions means trading days, excluding market closures. Each successful request creates a public prediction record.
HOW FINIO WORKS
A forecast you can question.
An answer you can measure.
v0.2 and v0.3 reduce one reusable scenario distribution into probabilities for supported financial events.
Model the market state.
Generate paired asset and SPY scenarios from completed-session market data. The forecast stays fixed while you ask questions.
Change the question.
Explore return direction and thresholds, SPY outperformance, daily-close drawdown, volatility and supported return ranges over 1-20 trading sessions.
Get a probability.
A supported question becomes a typed event. Finio counts the modeled paths that satisfy it, with no additional forecast generation or training.
Return to the record.
Saved predictions carry timestamps and model versions. Later recorded outcomes can be evaluated separately from simulations and historical tests.
FINIO MODEL EVIDENCE
We measure our claims
across model generations.
Three model generations. A measurable progression from fixed-task classification to reusable, learned forecasts.
Archive scope across v0.1, v0.2 and v0.3. Row-backed totals cover v0.2 and v0.3 only; supplied v0.1 figures are kept separate. Records count each model, asset, origin and question once. Shared market cases across versions remain separate model records, not independent trials. Assets are counted once across versions.
v0.1 / HISTORICAL
Original fixed-task model
One question. One horizon.
23,774 held-out samplesSupplied figures. Primary Brier 0.251623; constant baseline 0.251618. Row artifacts unavailable here; excluded from archive totals.
View v0.1 history →v0.2 / EXPERIMENTAL
Reusable statistical forecast
One forecast. Many questions.
1,157 historical forecastsPrimary Brier 0.249858. Positive 5-session return; AAPL benchmark proxy for other events.
View v0.2 evidence →v0.3 / BETA
Learned probabilistic model
Shared learned asset and SPY distribution.
1,148 historical forecastsPrimary Brier 0.246779. Positive 5-session return; actual SPY benchmark for other events.
View v0.3 evidence →Row-backed evaluation period: 2026-07-02 to 2026-10-29. Scores, baselines and protocols stay per model; these snapshots are not a head-to-head ranking. No overall predictive edge is established. Prospective production totals require an audited export.
Inspect the original v0.1 model card and historical figures
These figures were supplied by Finio. The underlying v0.1 evaluation artifacts are absent from this checkout, so they could not be independently reproduced here.
MODEL SPECIFICATION / v0.1 HISTORICAL
finio-model-v0.1
Shared cross-asset classifier
- Primary horizon
- 5 trading days
- Training securities
- 295
- Training samples
- 314,234
- Held-out samples
- 23,774
- Training period
- Jan 2022 – Jun 2026
- Held-out period
- Jul 2026 – Oct 2026
KNOWN LIMITATION
Historical testing has not yet demonstrated a meaningful predictive edge.
Inspect historical evaluation METRICS + BASELINES +
| Measure | Model | Baseline |
|---|---|---|
| Accuracy | 49.84% | 49.16% momentum |
| Brier score | 0.251623 | 0.251618 constant |
299 dataset securities; 295 training securities. Lower Brier is better. The model's Brier score is slightly worse than the constant baseline. These historical results do not establish predictive edge. Figures supplied by Finio; evaluation artifacts are not hosted in this website repository.
BUILD WITH FINIO
Financial-event probabilities.
In your software.
Add versioned forecasts to an app, agent or research workflow. Generate one distribution, then ask multiple typed questions without rebuilding the forecasting stack.
01 One reusable forecast
02 Structured probability answers
03 Saved records and evidence
Public experimental endpoints. Successful requests create public records. Availability and data permissions require separate verification.
Versioned contracts and saved records
v0.3 uses /api/v3; v0.2 uses /api/v2. POST /forecasts creates a forecast, POST /queries answers a typed question, and GET /forecasts/:id or /predictions/:id retrieves saved evidence.
Original v0.1: POST https://api.tryfinio.com/v1/predict with {"symbol":"NVDA","horizon":"5d"}. Its fixed-event contract and ledger are separate.
v0.3 / JAVASCRIPT / REAL API CONTRACT
One forecast, then a question
const base = 'https://tryfinio.com/api/v3';
const post = async (route, body) => {
const response = await fetch(base + route, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify(body)
});
if (!response.ok) throw new Error('Request failed');
return response.json();
};
const forecast = await post('/forecasts', {
asset: 'NVDA'
});
const answer = await post('/queries', {
forecast_id: forecast.forecast_id,
query: {
horizon: 20,
event: { metric: 'outperformance',
operator: '>', value: 0 }
}
});
console.log(answer.probability);The example asks whether NVDA outperforms SPY after 20 trading sessions. No response value is invented. Asking another supported question reuses forecast.forecast_id.
FINIO ON IPHONE / COMPANION APP
Take your research
with you.
The Finio iPhone app brings market research, learning and simulated portfolios into a companion interface.
The web model exposes explicit forecast versions and evidence. App signal scores and AI projections are distinct outputs and do not imply that every screen uses v0.3.
