Edge Hint Extractor is the first stage of the split edge research workflow: observe, abstract, design, pipeline. It reads market_summary.json, anomalies.json and optional news reaction files from a detector run, and a Python script generates deterministic rule-based hints.
You can add ideas written by a language model, either from a pre-written YAML file or from an external command. The result is hints.yaml with metadata and rule and LLM hint counts. It structures research input and gives no investment advice.
What You Get
- A hints.yaml file with a list of structured hints
- Generation metadata and counts of rule-based and LLM-generated hints
- Constrained ideas, because model hints are tied to the day’s anomalies and news context
- A clean input file for the concept synthesizer or the candidate detector
Common Use Cases
- Turning a day’s anomalies and news reactions into reusable hint objects
- Letting the model propose ideas that stay close to the observed data
- Preparing the input for concept synthesis in a daily research routine
What Makes It Unique
It supports two ways to add model ideas, an external command and a hints file, so it works with a separate LLM tool and with Claude writing hints inside the session. Rule-based hints are always generated first, which keeps the base output repeatable.
Under The Hood
The agent runs a local Python script that reads JSON and CSV observation files and writes hints.yaml. With the command option the script starts the command you supply as a subprocess and reads its output, so that command runs on your machine with whatever access it has. The file option reads a local YAML file. It needs Python and PyYAML, and a hints schema reference loads when needed.