Inside EdgeLab's Evaluation Bench
EdgeLab evaluates agents and strategies with datasets, task functions, custom evaluators, hard gates, and dogfood scoreboards before research can mutate portfolio state.
EdgeLab Engineering
How a trading research system is built, evaluated, and operated. Start with a bar, a research idea, an eval, or a portfolio mutation instead of a pile of disconnected reference pages.
Trace a live bar from provider websocket through storage, aggregation, indicators, and the execution queue.
1 entry Follow a research idea Research LoopMove from scanner and replay evidence into verdicts, research-cell state, and paper realization.
1 entry Follow an eval Evaluation BenchTrace datasets, cases, task functions, agents, evaluators, reports, gates, and keep/kill decisions.
2 entries Follow a portfolio mutation Portfolio HealingFollow observation, diagnosis, research-cell admission, paper realization, and promote/reject decisions.
1 entry Latest
Follow a Portfolio Mutation
Portfolio healing starts with observation and diagnosis, then moves through research-cell admission, paper realization, shadow behavior, and promotion or rejection.
Follow an Eval
An EdgeLab eval starts as a dataset case, runs through a task function and agent/runtime, then becomes evaluator output that can gate research behavior.
Follow a Bar
A live bar moves from Alpaca into one-minute storage, RabbitMQ, ValKey aggregation state, indicator tables, and the execution queue before it can influence a strategy.
Follow a Research Idea
A research idea moves from scanner evidence to replay, verdicts, admission state, and paper realization before it can become portfolio behavior.
EdgeLab Runtime Map
A map of the API, engine, queues, stores, broker paths, research loop, and public docs runtime as one connected system.
Inside EdgeLab's Evaluation Bench
EdgeLab evaluates agents and strategies with datasets, task functions, custom evaluators, hard gates, and dogfood scoreboards before research can mutate portfolio state.
Series
Series
AutoResearch
Deep dives and run notes about EdgeLab's strategy research loop, from baseline diagnosis to keep or discard decisions.
6 entries
Engine Foundations
Notes and deep dives around the execution model, job processing, and service boundaries that keep the platform moving.
4 entries
Knowledge Bank
Deep dives on ingestion, retrieval, and the data paths behind EdgeLab's book and research corpus.
1 entry
Reference shelves
Still useful, not the front door
These older shelves remain available for continuity while the public site moves toward learning paths and evidence-backed deep dives.