skills.rest
gozonerd
gozonerdCOMMUNITY
Explore all skills by this author →
GitHub

Comparative Ranking Methodology

Generates prompt variants, scores outputs against quality criteria, and selects the strongest result.

Updated Apr 11, 2026
Search Tags:#quality scoring#parallel subagents#thread generation#prompt variants#convergence audit#comparative ranking#output selection
One-click install
npx skills add https://github.com/gozonerd/shadow-ai-assessment --skill comparative-ranking-methodology-gozonerd
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: Comparative Ranking Methodology
Source: https://github.com/gozonerd/shadow-ai-assessment/tree/main/.claude/skills/dmis/methodology/Comparative_Ranking
Command: npx skills add https://github.com/gozonerd/shadow-ai-assessment --skill comparative-ranking-methodology-gozonerd

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When multiple valid prompting approaches exist for the same task, choosing one arbitrarily risks suboptimal output quality. This methodology replaces guesswork with a structured competitive selection process that ranks candidate outputs against predefined quality criteria. ## Core Features & Use Cases - Parallel Prompt Variants: Designs N meaningfully different prompt variants (P1=15, P2=25, P3=5, P4=5 per pipeline) executed as parallel subagent invocations. - Criteria-Based Ranking: Scores every output against predefined quality criteria from the task-type SKILL.md and ranks them by composite score. - Patch Contribution Extraction: Identifies unique strengths in lower-ranked outputs that can improve the selected primary thread. - Use Case: During a thread generation phase, run 15 prompt variants for pipeline P1, rank all outputs against the documented quality rubric, select the top performer for synthesis, and record the ranking rationale in the convergence audit input. ## Quick Start Apply the comparative ranking methodology to generate and rank prompt variant outputs for the current thread generation task.

Frequently Asked Questions about Comparative Ranking Methodology

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I choose the best output from multiple AI generations?▼

Generate outputs from meaningfully different prompt variants, validate each against the output schema, then score them against predefined quality criteria. Rank by composite score and select the top performer, documenting the rationale for audit purposes.

How many prompt variants should I generate for comparative ranking?▼

The variant count equals the thread count per pipeline: P1 uses 15 variants, P2 uses 25, P3 uses 5, and P4 uses 5. Each variant must differ meaningfully rather than being a trivial rephrasing.

When should I use comparative ranking instead of a single prompt?▼

Use it during thread generation phases where multiple valid approaches exist and quality criteria are well-defined enough for comparative scoring. It is not suitable when only one viable approach exists or criteria cannot be specified in advance.

Can ranking criteria be adjusted after seeing the outputs?▼

No. Ranking criteria must be predefined before execution. Post-hoc adjustment to favor a preferred output is an explicit anti-pattern that undermines the integrity of the selection process.

What are the limitations of prompt-variant-based diversity?▼

Diversity comes only from prompt variation, not model diversity, since the architecture is Kimi-native. Near-identical variants produce redundant outputs, so each variant must embody a genuinely different strategy.

Related Skills

Explore More Product & Management Skills→
selfagencyselfagency

prompt-builder

Build structured GitHub Copilot prompt files with YAML frontmatter and validation criteria.

Community
Intermediate
mohamedameen-iomohamedameen-io
158

sk-niche

Score and rank business niche ideas using Taki Moore 3Q and Hormozi 4-criteria.

Community
Intermediate
bmsull560bmsull560

structured-outputs

Validate LLM outputs against Pydantic schemas with function calling.

Community
Advanced
jshsakurajshsakura
22

prompt-engineer

Refine prompts to enforce task boundaries, output contracts, and safety constraints.

Community
Advanced