Level Tuner reads a sample of your playtest logs, finds where players get stuck or breeze through, and proposes tuning changes to your level files as a patch you review like any other change. It learns your formats from the sample: it writes a small adapter that turns your logs into one event table, shows you what it parsed, and stops instead of guessing when the sample does not parse.
Usage
Put a sample of playtest logs in playtests/ and one level file in levels/. The first run writes an adapter for your formats and prints the parsed rows for you to check.
Start Tune and pick the levels to look at.
Read the report, then apply the patch, or only the hunks you agree with.
What it produces
tuning/adapter.js: turns your log format into one event table (session, level, event, position, time, outcome), yours to keep and edit
tuning/metrics.csv: per level and segment: sessions, completions, deaths, retries, median time and quit point
tuning/report.md: a difficulty curve per level, with the spikes marked and explained
tuning/proposal.patch: suggested changes to enemy counts, timings and pickups in your level files, each hunk citing the metric behind it
Configuration
targetCurve (default gentle): gentle, steady or steep, the difficulty shape to aim for.
minSessions (default 20): levels with fewer sessions are reported but not tuned.
logs and levels (default playtests/ and levels/): where the sample logs and your level files live.
adapter (optional): an adapter saved from an earlier run; without it, a new one is written from the sample.
Notes
It does not play the game; it reads the sessions you already recorded.
The parsing and the metrics are plain code; the model works from the metrics table, not from raw log text.
Your level files change only when you apply the patch.
More sessions give steadier suggestions; a single tester's run is mostly noise.
