Probability intelligence for prediction markets

Prediction markets give you a price.
We grade it.

Bucephus combines prediction-market prices with news, public data and crowd signals to produce calibrated probabilities across 37 categories, and grades every forecast against the outcome.

Scroll
Reads signals from
01  The problem

A price is not a probability
you can trust.

Polymarket and Kalshi show you where money sits. They don't tell you how often prices like that turn out right, and nobody publishes a public scorecard. Analysts, media teams and trading desks end up building their own models.

I

Noisy prices

Thin markets swing on a single order. The number moves, the truth doesn't.

II

No track record

Nothing tells you whether a 70% price has historically meant 70%.

III

Scattered data

Over 100,000 live markets across dozens of topics, and no single place to compare them.

02  How it works

One probability,
built from many signals.

A market price is one input. Bucephus also reads news, public datasets and the crowd, weighs each by how reliable it has proven, calibrates the result against past outcomes, and grades itself when the market resolves.

Why 68%Illustrative
Market price
62%
News
64%
Model
57%
Public data
71%
Crowd
65%
Bucephus
68%
+6 pts vs marketConfidence 0.81 · 5 sources · low dispersion

Format illustration, not a live forecast.

  1. 01

    Ingest

    Market prices, news, public datasets and crowd signals, from 21 sources.

  2. 02

    Weigh

    Each signal counts for more or less by how reliable it has proven. It is not a plain average.

  3. 03

    Calibrate

    The pooled number is adjusted against how past forecasts actually turned out.

  4. 04

    Grade

    When the market resolves, the forecast is scored and added to a public record.

03  What it looks like

The product, as it runs today.

The product's homepage carousel, rebuilt live: a match with its model, then spotlight markets with their history. It rotates by itself and pauses when you hover. Click and hover everything.

Real values from the running product. Market history is AI-reconstructed from coverage. The live site is being relaunched.

04  The public record

Scored against reality, not opinion.

Brier score is the standard measure of forecast accuracy, and lower is better. Across 1,788 resolved markets, Bucephus scored 0.076, against 0.136 for simply guessing the base rate.

Straight talk. On this measure Bucephus is level with the market price, not ahead of it. Today the value is calibrated probabilities, broad coverage and a public graded history in one place, and we say so openly.

Brier score · 1,788 resolved markets

Bucephus0.076
Guessing the base rate0.136
00.050.100.15

Lower is better. Bars share one scale from 0 to 0.15.

As of 2026-09-20. Resolved binary prediction markets (16.2% resolved YES). The benchmark is guessing the base rate. The football World Cup run is a separate test.

05  The product

One layer above the exchanges.

i.

Calibrated probabilities

Model-adjusted odds next to the market price, so you can see exactly where they disagree.

ii.

A graded history

Every forecast is stored and scored when its market resolves. The record is built in public.

iii.

Feeds, API and embeds

Built for teams who want probabilities inside their own product. Built, not yet sold. Not an exchange, no real-money wagering.

bucephus / forecastILLUSTRATION
Example market
Will the outcome resolve YES?
Market price62%
Bucephus68%+6 pts vs market
0%Where the two disagree100%
Format illustration, not a live forecast
How a forecast is graded
From forecast to public record
1

Stored

The forecast is saved before the outcome is known.

2

Resolved

The market settles on the real outcome.

3

Scored

The Brier score joins the public history.

Curve illustrates calibration: forecast vs outcome
For builders
Probabilities inside your product
REST APIEmbedsData feeds
GETEmbedFeed
1 2 3 4 5 6GET /forecast/{market} { "price": 0.62, "model": 0.68, "graded": true }
Built, not yet sold. Shape illustration.
06  Who it's for

Built for people who need probabilities, not just prices.

Researchers and traders

Compare a market's price with a calibrated probability and its track record.

Media and analysts

Put a probability, and why it moved, next to the story.

Developers

Feeds, an API and embeds, so probabilities live inside your own product.

Decision teams

One consistent probability layer across dozens of topics.

How it can earn: API access, embeds and data feeds. Built, not yet sold.

07  Traction

Proof from a single World Cup run.

Built by one founder directing AI agents. Live 13 June to 21 July 2026, with no ad spend.

3,432unique visitors on the peak day, 9 July
~314Kpage views while live
187registered users
$0spent on ads. All organic.

Every figure comes from our own analytics.

08  Let's talk

See the numbers behind the record.

The investor deck covers traction, the graded record and the plan in full.

AK

Amit Kumar

Solo founder, full-time on Bucephus, based in Bangalore. Built the product, the forecasting pipeline and the public record, directing AI agents. amithrathod@protonmail.com · @BucephusHQ