scorecard (internal)

Turn gate results and design evidence into 1–10 marks, findings, and optionally issues.

Not a slash command

This is an internal procedure (plugin/prompts/scorecard.md), not something you invoke. /rhiza:quality reads and follows it once the gates have run and design-analysis has produced its evidence.

The scoping rule

The most important rule in the plugin's scoring: score only what the repo owns.

• In scope: src/, tests/, pyproject.toml, README.md, project docs, .rhiza/template.yml, locally-hardened config.
• Out of scope: everything the template delivers — .github/workflows/*, Makefile, .pre-commit-config.yaml, pytest.ini, ruff.toml, and the typecheck/mutation/fuzzing targets.

A gap in a Rhiza-owned file is fixed upstream, and is reported as "upstream/out-of-scope" rather than scored. Without this, every managed repo would be marked down for its own template — making the score meaningless and identical everywhere. A gate that was unavailable for the repo's profile is out-of-scope too, never a FAIL.

What it does

1. Assigns marks — always the two design subcategories, plus the gate-derived ones (lint, types, docs, test pass rate, coverage, deps/security, template fidelity) and any additional ones with real signal: test design quality, error handling and CLI UX, security posture, semver discipline, cross-platform robustness, idempotency, user-facing docs.
2. Requires evidence per mark — a score with a justification that cites nothing isn't earned; the subcategory is dropped instead.
3. Produces findings — one per subcategory below 10, each with a self-contained title, current→target score, the file(s)/lines to change, a done when… criterion, and an evidence snippet, ordered by leverage.
4. Offers to file them — via an AskUserQuestion multi-select, never free text. Nothing is created without an explicit selection. Filing goes through plugin/scripts/platform_cli.py issue-create, which maps to gh issue create or glab issue create — and matters because glab issue create has no body-file flag, so the body has to be passed inline.

Notes

• The coverage bar is the repo's own COVERAGE_FAIL_UNDER, not a number picked at scoring time.
• Coverage alone doesn't earn the test-design mark — a brittle suite can hit 100% and still pin internals.
• It re-runs nothing and gathers nothing. A missing number means an unscored subcategory, not a guess.

Reference

Source	plugin/prompts/scorecard.md
Invocation	not a slash command — reached with Read, never invoked
Read by	/rhiza:quality, design-analysis, known-issues