RFC-026: Human Feedback Signal — 10. Feedback Aggregation for Dashboards

AIGP Specification › RFC-026: Human Feedback Signal › 10. Feedback Aggregation for Dashboards

← 9. Rate Limiting and Abuse Prevention · Section index · 11. Compatibility →

10. Feedback Aggregation for Dashboards

10.1 Aggregation Buckets

The governance server SHALL compute rolling aggregates from FEEDBACK messages:

Aggregate	Granularity	Key Dimensions
Satisfaction rate	Hourly, Daily	app_id, use_case, model_id
Feedback volume	Hourly, Daily	app_id
Negative feedback trend	Daily (7-day rolling)	app_id, use_case
Top negative topics	Daily	app_id (from nlp.topics)
Expectation gap rate	Daily	app_id, use_case

10.2 Satisfaction Score

satisfaction_score = thumbs_up_count / (thumbs_up_count + thumbs_down_count)

Computed per app_id, use_case, model_id at hourly and daily granularity. Exposed on the STATUS response when FEEDBACK data is available:

{
  "feedback_summary": {
    "period": "24h",
    "satisfaction_score": 0.82,
    "total_feedback": 156,
    "thumbs_up": 128,
    "thumbs_down": 28,
    "top_negative_topics": ["iam", "timeout", "formatting"]
  }
}

10.3 Alert Integration

Feedback aggregates integrate with the existing Alert entity:

Condition	Alert Severity	Threshold (configurable)
Satisfaction score drops below threshold	WARNING	< 0.70 (default)
Satisfaction score drops below critical	CRITICAL	< 0.50 (default)
Negative feedback spike (>3x hourly average)	WARNING	3x multiplier
Expectation gap rate exceeds threshold	WARNING	> 0.30 (default)

10.4 Dashboard Panels

The governance server dashboard SHALL include a Feedback panel showing:

• Satisfaction score trend (7-day line chart)
• Feedback volume by rating (stacked bar, hourly)
• Top negative topics (word cloud or ranked list)
• Per-use-case breakdown (table with score, volume, trend direction)

---

← 9. Rate Limiting and Abuse Prevention · Section index · 11. Compatibility →