Release · 2026-08-26T04:50:12+02:00 · By Threadnote
Threadnote 4.3.6
Threadnote 4.3.6 lets foreground graph indexing begin without first reclaiming unreachable rows from a previously published build. Schema admission and all durability checks still run before foreground work; the same bounded cleanup now runs when the indexing session exits and remains included in end-to-end indexing time.
What's new
Threadnote 4.3.6 lets foreground graph indexing begin without first reclaiming unreachable rows from a previously published build. Schema admission and all durability checks still run before foreground work; the same bounded cleanup now runs when the indexing session exits and remains included in end-to-end indexing time.
Faster incremental graph admission
- Published build-only rows remain invisible to readers and are no longer an input to the next indexing admission.
- Threadnote reclaims one bounded page from each completed-build table after foreground work, then schedules the existing best-effort continuation when more rows remain.
- Cleanup failures keep their previous non-fatal contract, and worktree race checks, SQLite durability, graph parity, and transaction behavior are unchanged.
The retained exact v4.3.5 IntelliJ observation passed the cold, one-file total, post-scan, parity, proportionality, transaction, and failure-counter gates, but registration took 5,233.720 milliseconds and missed its strict five-second gate by 233.720 milliseconds. That failed observation was preserved rather than rerun for a favorable sample. The exact v4.3.6 release-scale observation will likewise be retained separately after publication.
Existing standalone installations upgrade with:
threadnote update