fast-analysis

Analyze game export files to assess gameplay and model decisions without raw logs. This workflow generates structured summaries covering llm_events and llm_reasoning, enabling fast triage and quick identification of bugs across exported games.

Batch analyze exported games using a parallelized, script-driven workflow to process multiple files simultaneously. This generates per-game json outputs containing game_overview and llm_events, enabling quick identification of issues across multiple game exports.

Analyzing exported game files produces structured outputs: game_overview, game_narrative, llm_events, and llm_reasoning. These compact summaries surface decision quality, errors, and tool usage, exported to per-game json files for quick triage.

Evaluate LLM reasoning by running script-driven analysis on game exports to extract llm_reasoning and llm_events. This workflow surfaces decision quality, errors, and tool usage patterns without requiring access to the original raw logs.

No, raw server logs are not needed to assess gameplay and identify likely bugs. The fast-analysis workflow operates directly on exported game files to evaluate model decisions and surface errors using reusable analysis scripts.