# Agent-Friendly CLI Experiment Dataset

## Core docs

- `docs/plan.md` — original experiment plan
- `docs/agent-friendly-cli.md` — principle report
- `docs/agent-friendly-cli-paper.md` — research-paper draft
- `questions/questions.yaml` — question bank

## Experiments

- `experiments/00-baseline-agenteval/` — baseline measurement, retry, calibration, Phase 0 probe
- `experiments/01-categories-truncation-warning/` — truncation warning paired experiment
- `experiments/02-count-wrong-column-hint/` — wrong-column hint paired experiment
- `experiments/03-wrong-type-hint-arm-d/` — Arm D wrong-type paired experiment and plan

## Arms

- `arms/arm-a-loop-as-skill/` — full-bank candidate screening by loop-as-a-skill
- `arms/arm-b-prompt-gepa/` — prompt-only GEPA run and candidate tree
- `arms/arm-c-full-repo-gepa/` — full-repo code GEPA run plus invalidated early runs
- `arms/arm-e-failure-routed-search/` — failure-class routing, `categories --search`, semantic vocabulary runs

## Synthesis and provenance

- `synthesis/` — derivation/blog principle notes
- `orchestrator/` — handovers and run-control notes
- `previous-publication/` — previous publication bundle snapshot
- `archive/older-optimizer-runs/` — older optimizer/noise runs retained for completeness

## Privacy note

This dataset intentionally excludes local process files and external trace-link manifests.