fpf-characterize

name	fpf-characterize
description	Define the characteristic space — what could matter, how to measure it, how to compare. Think through what "better" means before generating or comparing anything.
argument-hint	[domain-or-comparison-context]

What this skill IS

This is defining what matters, not listing metrics. You are constructing the space in which problems and solutions live — deciding which dimensions exist, how they're measured, and what trade-offs are possible. Without this, NQD characterization and Pareto analysis have no basis.

Work through the template. The characteristic space IS the lens through which you see the problem. Different spaces → different solutions look good.

Output

• Passport: .fpf/characterizations/CHR-${CLAUDE_SESSION_ID}--<slug>.md
• Cards (optional): .fpf/characterizations/CHRC-${CLAUDE_SESSION_ID}--<slug>.md

Constraints (quality bar)

• C1: Every characteristic has scale type (ordinal/interval/ratio) and polarity (↑/↓/target)
• C2: 1-3 active indicators selected from full space (not all characteristics are indicators)
• C3: Each indicator has threshold/target, baseline, reproducible measurement method
• C4: Comparison rules unambiguous — another agent could apply them mechanically
• C5: Measurement methods are reproducible (specify exact commands/tools)
• C6: Characteristic cards for non-trivial indicators include validity window

Format

# Characterization Passport
- **ID:** CHR-...  **Context:** ...
- **valid_until:** YYYY-MM-DD

## Characteristic space
| # | Characteristic | Scale | Polarity | Unit |
(all dimensions that COULD matter)

## Active indicators (selected for this comparison)
| # | Indicator | Target | Baseline | Measurement method |
(1-3 selected — these drive NQD Q-dimension in SPORT-*)
Selection rule: (WHY these indicators and not others — explicit rationale tied to roles/viewpoints)

## Comparison rules
- Dominance policy: (e.g., "V1 dominates V2 if better on all indicators")
- Tie-breaking: (e.g., "if tied on indicators, prefer higher N then D_p")
- Normalization: (if indicators have different scales)
- Missing data: (how to handle "unknown"/"no data" — explicit rule, not silence)

## Acceptance criteria
(what "good enough" means — feeds into PROB-* acceptance spec)