---
description: Evaluate one or more Abulafia / Journal-Yuga scenario outputs against the 21-story methodology and produce a scored report.
args: <target>
section: Research Workflows
topLevelCli: true
---
Evaluate scenario work against the 21-story evaluation methodology: $@

Derive a short slug from the target (lowercase, hyphens, no filler words, <=5 words). Use this slug for all files in this run.

This is an execution request, not a request to explain the methodology. Carry out the evaluation with tools and durable files. Do not stop after planning.

Source methodology:
- `website/src/content/docs/agents/story-evaluation-methodology.md`
- `website/src/content/docs/agents/megaagent-user-stories.md`
- `website/src/content/docs/agents/journal-yuga-user-stories.md`

Supported targets:
- A single story ID, such as `A07` or `S10`, plus an output artifact to evaluate.
- Multiple story IDs plus their output artifacts.
- `all` or `A01-A10,S01-S11` plus a directory containing scenario outputs.
- A plan, transcript, or report bundle that claims to cover the 21 stories.

Required artifacts:
- Plan: `outputs/.plans/<slug>-evaluation-plan.md`
- Evidence notes: `outputs/.drafts/<slug>-evaluation-evidence.md`
- Final report: `outputs/<slug>-evaluation-report.md`

Workflow:
1. Create `outputs/.plans`, `outputs/.drafts`, and `outputs`.
2. Read the methodology file first. Extract:
   - the 0-4 scoring scale;
   - pass threshold 75/100;
   - production-ready threshold 85/100;
   - global critical fails;
   - per-story scorecards for all story IDs requested by the user.
3. Identify the target artifacts:
   - If the user supplied file paths, read them directly.
   - If the user supplied a directory, inspect filenames and infer which artifacts map to which story IDs.
   - If the mapping is ambiguous, still evaluate what can be mapped and mark the rest `BLOCKED: missing or ambiguous artifact`.
   - Do not invent missing artifacts.
4. Write `outputs/.plans/<slug>-evaluation-plan.md` with:
   - requested story IDs;
   - target artifact paths;
   - evaluation fixtures inferred from the available inputs;
   - blocked or missing materials;
   - scoring dimensions to apply.
5. Evaluate each requested story independently:
   - Assign 0-4 scores for each weighted dimension.
   - Compute weighted total out of 100.
   - Check global critical fails and story-specific critical fail.
   - Record concrete evidence from the artifact for every score.
   - Mark unsupported claims, missing outputs, and unverified assumptions.
6. Write evidence notes to `outputs/.drafts/<slug>-evaluation-evidence.md` before the final report. Include inspected files, observed outputs, missing artifacts, score rationale, and critical-fail checks.
7. Write exactly one final report to `outputs/<slug>-evaluation-report.md` with:
   - Executive Summary
   - Coverage Table: story ID, artifact, total score, decision
   - Critical Fail Table
   - Per-Story Scorecards
   - Cross-Story Failure Patterns
   - Production Readiness Assessment
   - Required Fixes
   - Regression Notes
   - Sources / Inspected Files
8. Decision rules:
   - `PASS`: weighted total >= 75 and no critical fail.
   - `PRODUCTION-READY`: weighted total >= 85 and no critical fail.
   - `CONDITIONAL PASS`: weighted total >= 75 with non-blocking missing evidence.
   - `FAIL`: weighted total < 75 or any critical fail.
   - `BLOCKED`: required artifact is missing or unreadable.
9. Before responding, verify on disk that `outputs/<slug>-evaluation-report.md` exists. If it does not exist, create it as a blocked report with the failure reason.

Never claim evaluation is complete unless the final report exists on disk. Never fill scores from intuition; scores must be grounded in inspected artifacts.