LP MODELNETWORK
Research Lab

Building an AI quant network for LP capital

AI models forecast
Markets keep the score

Learning from historical pool data
Testing every forecast on what comes next

Explore the research How AI earns trust
Follow the record

How AI earns a record

Research before capital

AI learns from history
Evidence earns trust

We are building an AI quantitative network for liquidity providers

The first study is designed to train a machine-learning model on past pool data and test its forecasts against unseen outcomes and simple rule-based baselines

Learning, before the outcomeMechanism illustration / Synthetic data

One unseen outcome tests all three forecasts

Data cutoff
LightGBM / AI modelHistorical volatilityEWMA

Planned study / One AI model + two rule baselines

ForecastEvidenceAllocation

The network we are building

Model contribution / LP demand

Independent intelligence
A shared proving ground

The long-term network is designed for independent AI models to compete through recorded forecasts, so LP managers can compare their evidence before adopting them

AI model researchersTrain and contribute models
The shared recordEvaluate the evidence
Liquidity managersChoose what earns trust
For liquidity managers

Understand the model before giving it a role in capital decisions

Compare forward records, inspect failure cases and assess the full economics of an LP position

For model researchers

Let a sustained record make the case for your model

The roadmap opens toward external model submissions and paid model services, with common baselines and forecasts recorded before outcomes

External model participation and model service payments are planned stages of the network

The standard

The prediction cannot follow the result

First, the forecast
Then, the verdict

A model records its range before the outcome
The market moves — the original forecast stays fixed

One forecast / One marketForecast recorded / Outcome hidden
Forecast recorded24 hours laterFrozen upper boundFrozen lower bound
Mechanism illustration / Synthetic pathRange width and coverage must be assessed together

The forecast is fixed while the outcome remains unknown

Recorded in advanceTimestamp, data cutoff and model version

Measured consistentlyCoverage, width and quantile error

Read the evaluation method

The first research scope

Start narrow / Build a record

One pool
A serious starting point

Initial pool focus

WETH / USDC

Forecast horizon24 hours

Machine-learning modelLightGBM

Rule-based baselinesHistorical volatility + EWMA

A forecast faces two tests

Prediction quality and position economics are assessed separately

Forecast quality

Coverage / Range width / Quantile error

Position economics

Inventory value / Fees / Gas / Slippage / Token swaps

Current phase / Methodology & data design

The Research Lab explains the mechanism using illustrative data
Live model results and capital execution are not available at this stage

Enter Research Lab

The path to capital

Responsibility follows evidence

Build the record
Then extend the responsibility

  1. Current focus

    Reproducible research

    Establish single-pool data, cost accounting and evaluation rules

  2. Next

    Forward observation

    Register forecasts and track new outcomes through a continuous shadow record

  3. Later

    Controlled capital

    Limited trials with team capital, with broader access dependent on results and operating controls

LP Model Network

Give intelligence
a record to stand on

Explore the research