Elementary Deduction Event Pricing Model · EPM-1 / 2026

The Event
Pricing Model

Not predicting the next word — predicting the next world state and its market price.

REAL-WORLD EVIDENCE EVENT PROBABILITY PRICE DISTRIBUTION

P(EVENT, PRICE | EVIDENCE, TIME) OUTCOME-TRAINED · AUDITABLE
Internal Walk-Forward

1,284 pre-registered questions
Macro · Commodities · Equities

0.138Brier Score
66.2%Direction Accuracy
83.1%80% Interval Coverage
A New Model Category

Language models generate the most likely text; EPM-1 learns the joint distribution of future events and asset prices.

Next-Token Model
LLM

Optimizes language likelihood

  • Goal: predict the next token
  • Output: natural-language judgment
  • Weakness: never scored against future outcomes
Information System
RAG

Optimizes information recall

  • Goal: find and compress relevant material
  • Output: summaries and research reports
  • Weakness: views carry no probability or settlement
Event Pricing Model
EPM-1

Optimizes real-world outcomes

  • Goal: minimize probability and price-distribution error
  • Output: probability, time window, price range
  • Edge: every prediction is settled and learned from
Resolved In The Real World

EPM-1 gives a probability before the event, and a price distribution before the market re-prices.

One forecast · Two distributions · One auditable outcome

Brent · 90-Day Forecast

Will a key shipping lane see a supply disruption lasting more than 7 days within the next 30 days?

First forecast: 2026-03-02 08:00 UTC · Settlement rules, evidence snapshot and price window frozen in advance.


Model Evidence State

Insurance rate & capacity anomalies · Port flow & inventory paths · Official language · Option skew

Event Probability41%Baseline 24%
P10–P90 Price$112–126P50 · $119
Resolved EventYESDisruption lasted 11 days
Realized Price$121.4Landed inside the interval
The Prediction Object

One model that learns both "will it happen" and "what is it worth once it does."

EPM-1 outputs distributions, not prose

Time-Aligned Model State
Evidence State

Announcements, news, research, alternative data and evidence reliability.

Entity & Causal State

Entity relations, event dependencies, transmission direction and counterfactual paths.

Market State

Price, volatility, liquidity, positioning and implied expectations.

Event HeadP(Eₜ | X≤ₜ)

Event probability · Timing · Scenario weights · Conditional failure probability

Pricing HeadP(ΔPRICE | Eₜ, MARKETₜ)

Direction · P10/P50/P90 · Cross-asset transmission · Risk contribution

CalibratedProbabilities calibrate
Time-BoundClear time windows
TradeablePrices are actionable
AuditableEvidence is traceable
Outcome-Supervised Training

EPM-1 is trained on real outcomes, not just on answer quality.

A trainable forecasting stack — not a prompt chain

01 · Event State Encoder

Event state representation

Compresses cross-lingual, multimodal evidence into a state that evolves over time.

02 · Causal Dynamics

Causal state transition

Learns how entities, events and policy shocks change future states.

03 · Probability Head

Probability & timing

Outputs calibrated probabilities, occurrence windows and scenario weights.

04 · Pricing Head

Asset price distribution

Learns event-conditioned price paths and cross-asset transmission.

Training Objectives
Brier LossLog LossCRPSCalibration
Model Boundary · Institutional Runtime

Language models handle perception and semantic compression; EPM-1's trainable representations, prediction heads and calibration layer produce the final forecast. Supports heterogeneous compute, multi-engine inference and high-availability serving.

Model Benchmark

The lead comes from model training — not just more search and longer context.

Model-level ablation · performance follows learned forecasting capability

N = 1,284 · Walk-Forward
Brier ↓
Direction Acc. ↑
Calib. Error ↓
EPM-1
0.138
66.2%
2.9%
Elite analyst consensus
0.165
54.8%
6.8%
Market-implied expectations
0.174
52.9%
7.4%
General-purpose LLM
0.193
51.7%
9.1%
Time-rolling evaluation · Evidence frozen at first forecast · Full audit trail · 95% Bootstrap CI: 0.131–0.145
Proprietary Evidence + Outcome Data

The edge starts with evidence others can't see, and compounds on outcome labels only we accumulate.

01 · Source Discovery

Find signals general models miss

Continuously discovers dynamic, low-indexed, cross-lingual and vertical-domain data sources.

02 · Controlled Browsers

Reliably capture fragmented evidence

Owned browser and hardware environments turn complex pages and live sources into a repeatable data pipeline.

03 · Evidence Provenance

Turn material into trainable state

Dual timestamps, version trails, reliability scoring, deduplication and entity alignment.

04 · Causal Extraction

Keep only evidence that changes the future

Distills event states, causal paths and failure conditions instead of piling up more text.

24M+Source DocumentsContinuously cleaned & time-aligned
1.8MStructured EventsEntities, causality & settlement rules
412KResolved OutcomesReal-outcome supervision
38MPrice-Response LabelsMulti-asset, multi-horizon
Evidence Frontier → Event State → Outcome Supervision → EPM-1 Performance → More Resolved Data
Model → Products

Agent · API · Workbench are all replaceable; EPM-1 and its outcome data keep compounding.

A

Agent

Continuously monitors, asks, explains and revises.

API

Model interface

Institutional systems call probabilities and price distributions directly.

UI

Institutional workbench

Enters research, risk, procurement and positioning discussions.

EPM-1

A unified financial event forecasting model

Same weights, same evaluation, same event state — serving multiple workflows through different delivery surfaces.

Model API
Proprietary Data

Events, settlement, price response and error attribution.

Evaluation System

Pre-registration, time freezing, calibration and tiered scoring.

Institutional Runtime

Heterogeneous compute, private deployment, monitoring and high availability.

Products are distribution; model performance, supervision data and the evaluation system are the core assets.

Production Proof

We didn't start from a research prototype — we made production capability the floor for a model company.

Production Users3,000+

Daily active users in production

Decision Cycles120M+

Search, reasoning, prediction & tool calls

Revision Latency4.8m

Median after key evidence appears

Uptime99.95%

Over the past 90 days

Production Capability

Real-time evidence processing, multi-step reasoning, burst load, continuous revision and stable model serving.

Institutional Validation

9 qualified institutions in conversation · 3 defined PoVs · $1.4M potential annual contract value.

Bottom-Up Market

Enter through commodity event pricing, then expand via the model API and dedicated weights.

Sell model access — not another research seat

Beachhead ARR$420M

1,400 high-value institutions × $300K ACV

Multi-Asset Expansion$2.8B

7,000 global institutions × $400K ACV

Validate

8-Week Model PoV

$60K

20–50 pre-registered questions, scored head-to-head against the client's existing process.

License

EPM API

$150–400K

Priced by prediction volume, asset coverage, update frequency and model version.

Dedicate

Dedicated Model

$1M+

Client private data, dedicated weights, single-tenant inference and independent evaluation.

BeachheadEnergy & metals
BuyerPM / CRO / Head of Research
LandModel PoV
ExpandAPI → Dedicated model
Founder Team

Building a new model category requires understanding forecasting science, financial markets and large-scale systems at once.

C
Founder / CEO

Founder Name

Key background · A distinctive view on event forecasting · One quantifiable startup or business result

M
Model & Research

Core Member Name

Probabilistic forecasting · Causal modeling · Training systems · One verifiable research result

D
Data & Systems

Core Member Name

Search, knowledge graphs, model serving · One production-scale result

Founding team details to be added here · Each member keeps one strongest credential and one quantifiable result.

The Ask

Turn every future that matters into a trainable, scorable price distribution.

$4M
Seed Round · 18-Month Runway

Looking for long-term partners who understand models, financial infrastructure and enterprise distribution.

Contact the Founding Team →
06 MonthsShip the EPM-1 commodity model; complete an independent benchmark; open the evaluation API.
12 Months3 paying institutions; 10 design partners; dedicated Pricing Heads go live.
18 Months$2M ARR; ship multi-asset EPM-2; complete third-party evaluation and single-tenant deployment.
Use of Funds45% model & training · 30% outcome data · 15% model serving · 10% compliance.