feedback-analysis

name	feedback-analysis
description	Use when processing accumulated feedback signals across PAS artifacts. Scans backlog directories, classifies signals, identifies patterns, and produces a prioritized report.

Feedback Analysis

Overview

Scan all feedback/backlog/ directories across the PAS project, read accumulated signals (PPU, OQI, GATE, STA), cluster them by target and theme, and produce a prioritized report that feeds into the Discovery phase.

When to Use

• At the start of a Discovery phase (feedback-driven mode)
• When the product owner asks "what feedback exists?"
• After multiple process runs have accumulated signals

Process

1. Scan for signals: Glob for all files in **/feedback/backlog/*.md across the project (excluding .gitkeep)
2. Parse each signal: Extract type (PPU/OQI/GATE/STA), target, priority, evidence, and the fix/preference
3. Cluster by target: Group signals that target the same artifact (skill, agent, or process)
4. Cluster by theme: Within each target, identify recurring themes (e.g., multiple OQIs about the same root cause)
5. Prioritize: Rank clusters by:
   • Signal count (more signals = stronger signal)
   • Priority level (HIGH > MEDIUM > LOW)
   • GATE signals always surface (they represent guardrails)
   • STA signals surface as constraints on proposed changes
6. Check for conflicts: Do any PPU signals contradict existing STA anchors? Flag these.
7. Produce report: Write the analysis to the workspace

Output Format

# Feedback Analysis Report

## Summary
- Total signals: {N}
- By type: {N} PPU, {N} OQI, {N} GATE, {N} STA
- Clusters identified: {N}

## Priority Clusters

### Cluster 1: {theme} ({N} signals, highest priority: {HIGH|MEDIUM|LOW})
**Target:** {artifact}
**Signals:** {list signal IDs}
**Pattern:** {what these signals collectively indicate}
**Suggested action:** {what addressing this would look like}

### Cluster 2: ...

## Conflicts
{Any PPU vs STA conflicts, or contradictory signals}

## Unclustered Signals
{One-off signals that don't form patterns, listed individually}

Quality Checks

• Every signal file was read (none skipped)
• Clusters are based on evidence, not inference
• GATE and STA signals are always surfaced, never buried
• Conflicts are explicitly flagged
• Report says "no signals found" when backlogs are empty — does not fabricate patterns

Common Mistakes

(Populated by feedback over time)