| name | lofn-evaluation |
| description | Evaluate, score, rank, and select Lofn outputs using panels, facets, eligibility scoring, and platform fit. Use after 24 outputs exist or when choosing finalists. Do NOT use to generate songs/images or perform QA formatting cleanup. |
SKILL: Lofn Evaluator — Panel of Experts Selection & Ranking
PREREQUISITES:
0. Load resources/panel-of-experts.md to understand the panel of experts prompting you will use.
- Load
skills/lofn-core/SKILL.md for personality and Panel system.
- Load
skills/lofn-core/PIPELINE.md for the MANDATORY execution pipeline.
- Load
skills/evaluation/TASK_TEMPLATE.md for exact evaluation requirements.
⚠️ This skill handles the critical panel selection and ranking phases of the pipeline. The evaluator makes or breaks the creative direction.
🎯 PURPOSE
The evaluator has two core functions:
- Panel Generation — Select and configure the 3 panels (concept, medium, context) with appropriate experts and transformations
- Selection & Ranking — Score and rank the 24 outputs from any modality, selecting the best N for delivery
📊 SELECTION & RANKING
Use Select_Best_Pairs.md for ranking outputs.
Requirements
After the pipeline generates 24 outputs:
- Define scoring facets (from Step 06)
- Score each output against the facets (1-10 scale)
- Score each output on the 7 Eligibility Properties (see QA SKILL.md §0A) — classify as ACCESSIBLE or AMBITIOUS
- Apply weights based on platform/goal
- Rank all 24 by weighted score
- Select top N based on request (default: best 4-6)
- Ensure pattern alignment — selections must align with the orchestrator's chosen pattern (BARBELL, ALL-ACCESSIBLE, ALL-AMBITIOUS, GRADIENT, CONTRAST-PAIRS). When BARBELL is selected, at least 40% of selections should lean toward the run's intended barbell classification.
Scoring Dimensions
| Dimension | Weight (Competition) | Weight (Social) |
|---|
| Originality | 25% | 15% |
| Technical Execution | 20% | 15% |
| Emotional Impact | 20% | 25% |
| Bold Choice Quality | 15% | 10% |
| Platform Fit | 10% | 25% |
| Viral Potential | 10% | 10% |
Eligibility Scoring (EMBEDDED 2026-05-02 — 14-panel meta-analysis)
In addition to creative scoring, every output is scored on 7 eligibility properties (full rubric: QA SKILL.md §0A):
| Property | Weight |
|---|
| Body in the song | 15% |
| Adoptable hook | 20% |
| Vast emotional TAM | 15% |
| Specificity paradox | 15% |
| Cognitive ease | 10% |
| Vocal co-discovery | 15% |
| Sonic threshold | 10% |
Accessible runs: eligibility average ≥3.5 is a HARD GATE. Below threshold → flagged for revision.
Ambitious runs: eligibility scoring is informational only (artistic identity, not mass reach).
Output Format
## Ranking Results
### Top Selections
| Rank | Title | Pair | Variation | Score | Key Strengths |
|------|-------|------|-----------|-------|---------------|
| 1 | ... | A | 3 | 8.7 | ... |
| 2 | ... | C | 1 | 8.4 | ... |
| 3 | ... | B | 4 | 8.2 | ... |
| 4 | ... | D | 2 | 8.0 | ... |
### Scoring Breakdown (Top 4)
#### Rank 1: [Title]
- Originality: 9/10 — [rationale]
- Technical: 8/10 — [rationale]
- Emotional: 9/10 — [rationale]
- Bold Choice: 8/10 — [rationale]
- Platform Fit: 8/10 — [rationale]
- Viral Potential: 9/10 — [rationale]
- **Weighted Total: 8.7**
[repeat for top 4]
### Panel Notes on Selection
[Key insights from the panel debate about why these won]
Release / Falsification Logging (ADDED 2026-05-02)
Before any candidate is published or treated as a formula test:
- Read
references/release_falsification_logging.md.
- Copy
assets/release_record.template.json into the run/release directory as release_record.json.
- Fill route, archetype, seed source, AI vocabulary level, pre-release eligibility scores, and hypothesis BEFORE metrics arrive.
- Validate with
scripts/validate_release_record.py.
- Update day 7 / 14 / 30 metrics later without rewriting the pre-registered hypothesis.
⚡ ACTIVATION
For Panel Generation:
- Research context — Current trends, cultural moment, platform requirements
- Select baseline experts — 6 per panel following composition rules
- Run panel debate — Full dissent, backtracking, synthesis
- Apply transformations — Group + Skeptic choices
- Output panel specification
For Selection & Ranking:
- Receive 24 outputs from pipeline
- Define/confirm scoring facets
- Score all 24 against facets
- Calculate weighted totals
- Select top N with rationale
The evaluator's judgment shapes the final output. Choose wisely. Score rigorously.
