After #487, 50k filter settle sits at ~104ms vs TanStack's ~50–58ms same-run. The traced remaining window (2026-08-25 re-attribution, docs/superpowers/specs/2026-08-24-columnar-verdicts-results.md): rebuild body ~19% (old-tree merge walk, flip-in sort, bulk tree build), residual HAMT ~12% (derivedById handling and rows-map ops), render/commit ~15%, layout dense refilter ~7%. No dominant lever remains — closing the gap needs most of these. This issue tracks the two row-model shares if the arc resumes.
🤖 Generated with Claude Code
After #487, 50k filter settle sits at ~104ms vs TanStack's ~50–58ms same-run. The traced remaining window (2026-08-25 re-attribution, docs/superpowers/specs/2026-08-24-columnar-verdicts-results.md): rebuild body ~19% (old-tree merge walk, flip-in sort, bulk tree build), residual HAMT ~12% (derivedById handling and rows-map ops), render/commit ~15%, layout dense refilter ~7%. No dominant lever remains — closing the gap needs most of these. This issue tracks the two row-model shares if the arc resumes.
🤖 Generated with Claude Code