Measurement
The Measurement group is where the raw per-prompt and per-sub-session signals live. These are the inputs the PRISM Score composites are built from — one page per metric, so you can drill into the underlying behavior rather than reading a single aggregated number.
| Item | Route | What it shows | Read more |
|---|---|---|---|
| Prompt Efficiency | /prism/prompt-efficiency | PES — four-dimension prompt rubric (Context Leverage, Information Density, Turn Economy, Ambiguity Cost) with trend and worst prompts | PES |
| Session Efficiency | /prism/session-efficiency | SSE — outcome anchor per sub-session, broken down by tokens / turns / human time / latency axes | SSE |
| Iteration Efficiency | /insights/ie | Turns per active CLI hour, convergence pattern | IE |
| Context Reset Rate | /insights/crr | /compact / /clear cadence vs. the healthy-middle band | CRR |
| Response Latency | /insights/rlr | Latency distribution, regressions, and the ratio against baseline | — |
| Flow Continuity | /insights/fc | Median minutes of uninterrupted focused stretches | FC |
| Quality Retention | /insights/qr | Share of AI output that survived review — feeds the Speed penalty | — |
| Token Usage | /insights/tokens | Tokens broken out by model and by cache-read / cache-write / uncached / output | — |
Which ones feed Skill
Section titled “Which ones feed Skill”Five of these are weighted inputs into the Skill composite. The others are diagnostic: they explain why a Skill input moved.
| Skill input | Weight | Page |
|---|---|---|
| SSE | 45% | Session Efficiency |
| PES | 20% | Prompt Efficiency |
| IE | 15% | Iteration Efficiency |
| CRR | 10% | Context Reset Rate |
| FC | 10% | Flow Continuity |
Not in Skill but surfaced here:
- Quality Retention feeds the Speed score as a penalty (low QR discounts Speed).
- Response Latency is a raw operational metric — useful for spotting regressions but not a coaching signal.
- Token Usage is a cost-tracking breakdown, not a quality signal.
How often these update
Section titled “How often these update”All pages update within a few minutes of a sub-session closing. Sub-session boundaries are detected by the engine from explicit signals (/clear, /compact, session-id change), implicit signals (idle gap, intent shift), and completion signals. See Architecture for the full pipeline.