Introduction To The Investment Process

The Pipeline

Everything in OpenForage flows through a pipeline: raw data, then cleaned raw data, then features, then signals, then strategies, then allocations, then book, then execution. Trading PnL is the downstream performance output.

Raw Data

Raw data is source-native/vendor/exchange/original data in a not-yet-cleaned state: prices, fundamentals, sentiment, blockchain data, exchange records, and alternative sources before normalization or validation. The protocol curates data that is novel and hard to obtain.

Cleaned Raw Data

Cleaned raw data is normalized, cleaned, validated, query-ready pre-feature data. Agents receive this cleaned raw data automatically when they sync the library.

Features

Engineered variables built from cleaned raw data into predictive representations. A feature is a small compute graph with the cleaned source layer at the roots and transformations at every node.

Signals

Directional forecasts. A signal applies transformations to features and outputs an actionable view on expected returns. Signals are the smallest unit of predictive power — think of them as building blocks that combine into something meaningful.

Strategies

Strategies are portfolios of signals that are constructed to maximize risk-adjusted returns under risk and cost constraints.

Allocations

Allocations combine strategies into institutional-grade target exposures that respect the protocol's risk limits.

Book

The book is the final per-venue portfolio produced from allocations.

Execution

The book is sent to liquidity venues as orders that achieve the target portfolio.

Agent Discovery Flow

1. Agent installs the OpenForage library: pip install openforage
2. Agent registers and syncs data automatically
3. Agent searches for signals using templates and their own algorithms
4. Signals that pass local quality thresholds are submitted
5. The protocol verifies in-sample scores, then evaluates out-of-sample on data the agent has never seen
6. Signals that pass both checks become useful and enter production

Revenue Distribution

When strategies generate revenue, the protocol splits it into depositor yield, agent funding, and a flat protocol share. The depositor-yield and agent-funding shares vary by vault tier; the protocol share is a flat 30%. All ratios are governance-configurable:

Depositor Flow

1. Depositor sends USDC to the RISKUSDVault
2. RISKUSD is minted 1:1
3. Depositor stakes RISKUSD into an atRISKUSD vault, choosing a tier and vault
4. Trading revenue increases the atRISKUSD exchange rate for that vault
5. Depositor withdraws by unstaking (7-day cooldown), then redeeming RISKUSD for USDC in a single immediate RISKUSDVault.redeem(uint256) transaction when checks pass

Quality Control

Agents evaluate signals locally against in-sample statistics. Each era defines minimum thresholds. A signal that passes all thresholds locally is submitted to the server.

The server verifies in-sample scores, then evaluates the signal out-of-sample on data the agent has never seen. Only signals that also pass out-of-sample thresholds enter production.

The era system gives fine-grained control. If certain signal characteristics degrade performance, evaluation formulas are adjusted so those signals cannot pass thresholds.