Skip to content

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

Speed

Speed answers one question: how much real work did you finish in this window? A finished unit is a sub-session that the LLM judge marked goal_complete and that was not rolled back as rework.

Crushed Sub-sessions Per Window. Counts only sub-sessions that pass the substance floor and that the judge marked complete. Scaled to the selected window (daily, weekly, or monthly).

CSPW = count(sub-sessions where substance_floor_passed AND goal_complete AND NOT rework)
MetricWhat it tells you
Δ-CSPWPercent change vs the previous window of the same length — are you speeding up or slowing down?
TTC (median)Time To Crush — median wall-clock from a sub-session’s first prompt to its goal_complete signal. Lower means tight loops.

Sub-sessions below the substance floor (under 3 turns, fewer than 10 lines changed, no mutating tool call) are filtered out. They surface separately on the Trivia page so nothing disappears silently.

Stalled · Light · Steady · Crushing. Bands are scaled to match the selected window. Thresholds are pre-calibration and will be refined after 4+ weeks of data.

Hours-spent metrics reward grinding. CSPW rewards finishing. Pair it with Prompt Grade to make sure the throughput is healthy and not coming from rushed, low-quality prompts.