stress

README

Stress test harness

Non-benchmark stress testing of recoil against real repositories.

Layout

• manifest.json — repo list with pinned SHAs, applied config, ground-truth answer paths.
• queries/universal.json — query archetypes applied to every repo.
• queries/<repo-key>.json — repo-specific queries layered on top.
• configs/ — per-Phase-3 config variants (classify.override, rank, include/exclude).
• synth/ — transcript synthesizer for Phase 2b.
• run.sh — end-to-end driver per repo.
• aggregate.py — produces the matrix view from per-repo reports.
• reports/ — per-repo and aggregated JSON outputs.

Clones

Clones live OUTSIDE this repo at ~/recoil-stress-clones/<repo-key>/<sha>/. Downloaded via GitHub tarball API (faster than git clone for our purposes, no .git overhead).

Documentation

Overview

stress is a non-benchmark stress-test harness for recoil. It exercises the CLI binary against real cloned repositories (path-pinned in manifest.json) across a query battery, multiple config variants, lifecycle scenarios, adversarial cases, and latency sweeps.

Subcommands:

go run ./stress run-repo <key> [--config <name>] [--label <name>] [--mine-limit N] [--include-hidden]
go run ./stress aggregate [--label <name>] [--floor 0.99]
go run ./stress phase3
go run ./stress phase4
go run ./stress phase5
go run ./stress phase6
go run ./stress phase7
go run ./stress synth <repo-key> [--n 6] [--seed 42]
go run ./stress report