Skip to content

Search is only available in production builds. Try building and previewing the site to test it out locally.

PRISM Score v3.0

PRISM Score v3.0 replaces v2.1’s three-score model (Speed · Skill · Efficiency) with a single outcome metric:

PRISM Score = 100 × crushed_count / total_count

A session “crushes” when an LLM outcome judge confirms it landed its goal against per-intent criteria, with no later rework. Every session also gets an LLM-generated title and 1–2 sentence summary that appear in the dashboard.

For the full mechanics, see the dedicated section: PRISM Score v3.0.

v2.1’s Speed/Skill/Efficiency still runs in parallel during the calibration window — toggle between the two in the dashboard sidebar.