signal-classify

Classifying session events into friction and success from M1/M2 logs involves applying a rule-based approach to categorize raw events into noise, friction, success, and neutral. This enables faster debugging and UX improvements without relying on LLMs.

Tagging tool-output streams for debugging without an LLM is achieved through deterministic, rule-based classification. The process applies CLI-driven thresholds like retry and long_chain to filter events, outputting structured signals for analytics.

Filtering classified session events is done using CLI-driven threshold options such as retry and long_chain. Running the classification script on an events.jsonl file generates a categorized output file tailored to your specific debugging needs.

Signal classification works directly with raw M1/M2 session logs to categorize events for UX analysis. It outputs structured signals suitable for dashboards and reports, tagging events into noise, friction, success, and neutral categories.

You should not use an LLM for session event classification when deterministic, reproducible results are required. A rule-based approach provides consistent tagging of M1/M2 session logs into noise, friction, and success without the variability of large language models.

Generating structured signals from events.jsonl requires running the classification script to process raw M1/M2 session logs. The script outputs categorized events that are directly suitable for dashboards, reports, and tooling integrations.