Andrew Fribush
fig. 2.4.4 · the built-in summary against the rules alone, at the compactions the hosts actually ran.
SUMMARY FREES 95%, LOSES MOST OF WHAT CAME NEXTONE PAIR OF BARS IS ONE SETUP; GREY IS CONTEXT FREED, ACCENT IS RESULTS THE SESSIONUSED LATER THAT ARE GONE.CLAUDE CODE · 29 COMPACTIONS0%50%100%BUILT-IN SUMMARYFREED95.0%LOST583 OF 822 · 71%RULES ALONEFREED21.4%LOST9 OF 822 · 1.1%CODEX · 27 COMPACTIONS0%50%100%BUILT-IN SUMMARYFREED95.3%LOST790 OF 805 · 98%RULES ALONEFREED13.5%LOST12 OF 805 · 1.5%

source: better-compaction docs/scoring.md:538-549 (642712c) · nativeCompaction.data.ts

View in context: I Built Compaction Without a Model: 13 Lost Results Against 514 → · All figures

Andrew Fribush, software and computer engineer in Austin. I build consequential systems where software meets institutions: finance, healthcare, commerce, government data, and the mechanisms that govern AI agents. Taking new engagements: hello@andrewfribush.com.