ワンクリックで
phira-baseline-parity
Checklist for proving baseline behavior is unchanged (or changes are intentional) via controlled A/B runs.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Checklist for proving baseline behavior is unchanged (or changes are intentional) via controlled A/B runs.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | phira-baseline-parity |
| description | Checklist for proving baseline behavior is unchanged (or changes are intentional) via controlled A/B runs. |
Use this skill when implementing a feature that should be opt-in, when defaults must remain identical, or when reviewers ask "did you change baseline behavior?".
This skill is intended to be lazy-loaded on demand: only run parity A/B commands if the researcher approved running them.
Goal: demonstrate parity between baseline and modified code under the same configuration.
If the acceptance criteria does not explicitly change defaults, your implementation should preserve baseline behavior by default.
Baseline parity
- Default behavior preserved: <yes/no/unknown>
- Evidence:
- A/B commands: <commands>
- Compared: <what you compared>
- Outcome: <match/tolerance/diff>
- If not preserved: <why + where documented + migration notes>
Deterministic Graphviz/DOT derivation + validation for archive records.
Canonical per-record archive format (YAML front matter + Markdown body).
Pointer semantics for "since last record" and detected-change bookkeeping.
Fast, compute-aware verification checklist for implementation changes (imports, configs, tiny runs).
Workflow for conference/venue paper search and short reading lists.
Procedural checklist to infer and state the repo's effective runtime environment and dependency policy.