Throttle Detection

The Throttle Detection page identifies when API rate limits are slowing you down.

What it detects

Prism runs 5 throttle detectors:

Detector	What it finds
Rate limit responses	HTTP 429 responses from the API
Elevated latency	Response times significantly above baseline
Retry headers	Retry-After headers in API responses
Request gaps	Unusual pauses between requests suggesting backoff
Error bursts	Clusters of errors indicating capacity issues

Hours lost

The page calculates total developer hours lost to throttling:

Hours lost = Σ (throttle_duration × sessions_affected)

This is presented as:

• Total hours lost in the selected time period
• Per-week average — trend line showing if throttling is improving
• Peak times — when throttling is most common (helps with scheduling)

ROI calculation

The dashboard shows the ROI of addressing throttling:

Developer hourly rate × hours lost = cost of throttling

This helps justify infrastructure changes or plan tier upgrades.

Mitigation strategies

Based on detected patterns, the page suggests:

Pattern	Suggestion
Frequent rate limits	Upgrade API tier or distribute requests
Peak-time throttling	Shift heavy work to off-peak hours
Burst errors	Implement request queuing or backoff
Model-specific limits	Use different models to distribute load