Skip to content

Token Usage

Token Usage tracks how many tokens a finished sub-session costs. A sub-session is “crushed” when the goal was reached and no rework was needed. Lower tokens per crushed sub-session means you reached the same outcome with less spend.

Median total tokens (input + output + cache reads + cache writes) per crushed sub-session in the selected range. Median is used so one giant sub-session can’t skew the number. It’s shown with a Lean / Healthy / Heavy / Bloated band and the crushed-vs-total count for the range, and a Crush Weight over time chart plots it per period so you can watch the trend.

MetricWhat it tells you
TETToken Efficiency Trend — this range ÷ prior range. Below 1.00 means cheaper than last time; above means more expensive.
TPTTokens Per Turn — input + output divided by total turns in the range. Cache reads and writes are excluded. It is a usage metric, not a reading of your current context size.
Estimated cost per crushed sub-sessionThe median cost of a crushed sub-session, resolved either entirely from the price catalog or from a raw Claude Code estimate. An estimate — not your billed cost.
Token mixWhere your tokens went across the scored sub-sessions in the range: input · output · cache read · cache write.

Sub-sessions below the substance floor (no turns, no code change, no mutating tool calls) are routed to Trivia and do not feed Crush Weight. Sub-sessions that failed the goal gate or hit rework still count toward TPT but are not part of the Crush Weight median.

Lean < 5M · Healthy 5–15M · Heavy 15–35M · Bloated 35M+ tokens per crushed sub-session (cache included).