Mix since the last report: research (figure backfill, swe-orgs batches), venture features, process verify runs. Nothing contradicts the owner's quotes in goals.md.
Instruction lines unchanged except research/PROJECTS.md, whose one 10.5k-character progress line was the single costliest read (68 reports); cut.
Open: process/1 success of 72% reflects verify-only runs that CLAUDE.md says do not count as useful work; no owner proposal added, as process/prompt-1.md already routes them to Solve mode.
Follow-up on 2026-10-17b: E30 gh/MCP wording; too few runs to judge.