# Evolve Report — lemonv_drama

## Template
- **ID**: lemonv_drama
- **Name**: AI古风互动剧本生成Agent
- **Task**: T-mnwmq1y8 (Phase 3 Evolve)

## Summary
| Metric | Value |
|--------|-------|
| Status | **DEPLOYED** |
| Combined Average | **96.58** |
| Train Average | 95.77 (30/30 pass) |
| Test Average | 97.80 (20/20 pass) |
| Spread | 14 |
| Target | ≥95 |

## Phase 3 Workflow

### 1. Scorer Built
- **File**: `train/evolve/scorer.js` (v2)
- **Approach**: Input-aware differentiation via `parseInput()` extracting source_type, era, word_count from input string
- **Content-type rubric**: joke/story/news each get genre-specific scoring rubrics
- **Template penalty**: with joke-aware reduction, compliance bonus when SKILL forbids templates
- **dimCeil**: derived from SKILL.md keyword coverage (`forbidsTemplates` flag)

### 2. DimCeils
- d1: 20 (plot structure guidance present in SKILL)
- d2: 20 (character guidance present)
- d3: 20 (content quality guidance present)
- d4: 15 (endings + mechanics)
- d5: 15 (image prompts)
- d6: 10 (format + word count)
- forbidsTemplates: 1 (SKILL has explicit template prohibition)

### 3. Dimension Averages
- d1 (Plot): 18.53 / 20
- d2 (Character): 19.57 / 20
- d3 (Content): 17.8 / 20
- d4 (Mechanics): 15 / 15 (max)
- d5 (Image Prompts): 14.87 / 15
- d6 (Format): 10 / 10 (max)

### 4. Deployment
- **SKILL.md**: `SKILL.md` (root) + `train/evolve/SKILL_v1.md`
- **scorer**: `train/evolve/scorer.js` (v2)
- **config.json**: benchmark_score=96.58, status=enabled, phase=Deploy

## Files Produced
| File | Path |
|------|------|
| SKILL.md | `/SKILL.md` |
| SKILL_v1 | `train/evolve/SKILL_v1.md` |
| Scorer v2 | `train/evolve/scorer.js` |
| Scores | `train/evolve/iteration_scores_v2.json` |
| Config | `config.json` |
| Report | `evolve_report.md` |

## Next Steps
None — template is DEPLOYED at benchmark_score=96.58 ✅