Module aggregate 

Stage 5 — aggregate.

Compares exactly two conditions: collects pass_rate (from grading.json), total_tokens/duration_ms (from timing.json), and the skill-invocation determination per condition; computes mean/stddev and the a - b delta; accumulates validity warnings (mixed timing sources, sub-100% invocation rate, stray-write violations + live-source reads, plugin shadows); and writes benchmark.json.

Structs

Benchmark

The full benchmark.json.

Stats

Summary statistics for one metric.

Functions

aggregate

Compute and write benchmark.json for the iteration. Requires exactly two conditions and at least one eval-* directory.