Skip to content

docs(research): interaction borderline perf diagnostic (real over-budget on pretable filter-text)#134

Open
blove wants to merge 3 commits into
mainfrom
interaction-borderline-perf-diag
Open

docs(research): interaction borderline perf diagnostic (real over-budget on pretable filter-text)#134
blove wants to merge 3 commits into
mainfrom
interaction-borderline-perf-diag

Conversation

@blove
Copy link
Copy Markdown
Contributor

@blove blove commented May 11, 2026

Summary

Tightens PR #131's two borderline numbers via n=20 re-run. No auto-merge — real-over-budget verdict on pretable filter-text requires editorial review.

Slice Verdict Detail
pretable filter-text over budget? real-over-budget 16.79 ± 0.31 ms at n=20; mean−2σ = 16.17 > 16 ms budget. Reliably ~0.8 ms over.
tanstack vs pretable filter-metadata parity noise-tied 1.6 ms mean diff vs 23.2 ms 2σ noise floor (tanstack σ=11.6 ms at n=8). PR #131's tie was sample noise.

Incidental finding: pretable filter-metadata is also over budget at the mean (17.51 ± 2.44 ms at n=20). PR #131's 16.0 ms n=3 reading was a low-end sample.

Full data

(adapter, script) n mean (ms) σ (ms) min median max
pretable, filter-text 20 16.79 0.31 16.10 16.70 17.50
pretable, filter-metadata 20 17.51 2.44 15.80 16.75 24.80
tanstack, filter-metadata 8 19.11 11.60 8.20 17.25 41.70
tanstack, filter-text 6 27.97 17.23 8.30 29.45 49.60

Recommendations (in the memo)

  1. Update /bench page prose: move filter-text + filter-metadata out of the "clears the budget" set.
  2. Update ComparisonTable.tsx interaction rows: decide whether to keep the "≤ 16" budget column for interaction (would show pretable failing it).
  3. Update TanStack trail-marker label: drop the "filter-metadata ties pretable" parenthetical (the high-repeat data shows no real tie).
  4. Scope a perf-fix investigation for pretable filter-metadata + filter-text. Both reliably over budget; likely shared root cause in the wrapped-text filter pipeline.

All four are editorial / scoping decisions the user should weigh in on before any homepage change ships.

Matrix wall-clock note

Original matrix flaked at ~22% (PR #133 had the same pattern); ran a follow-up tanstack-only matrix that also flaked early. Final sample counts: pretable 20+20, tanstack 8+6. Tight enough for unambiguous verdicts.

Test plan

  • `pnpm -w typecheck` passes (no source changes)
  • `pnpm -w lint` 0 errors
  • `pnpm format` clean
  • Milestone JSON + memo committed under `status/milestones/` and `docs/research/`

🤖 Generated with Claude Code

blove and others added 3 commits May 11, 2026 14:10
Tightens PR #131's two borderline numbers (pretable filter-text 17.7 ms,
tanstack vs pretable filter-metadata 15.7/16.0 ms) via n=20 re-run.
Mirrors PR #124 / PR #133's pattern.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Verdict: pretable filter-text is real-over-budget (16.79 ± 0.31 ms at
n=20); tanstack vs pretable filter-metadata is noise-tied (1.6 ms
mean diff vs 23 ms 2σ noise floor — tanstack's σ at n=8 is 11.6 ms).

Incidental finding: pretable filter-metadata is also over budget at
the mean (17.51 ± 2.44 ms). PR #131's 16.0 ms n=3 reading was a
low-end sample. Homepage prose claims filter-metadata is "clear of"
the single-frame budget; that's no longer accurate.

Three recommendations queued (editorial cleanup + perf-fix
investigation); see memo for details.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
@vercel
Copy link
Copy Markdown

vercel Bot commented May 11, 2026

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
pretable Ready Ready Preview, Comment May 11, 2026 9:16pm

@github-actions
Copy link
Copy Markdown
Contributor

Vercel preview ready

Preview: https://pretable-9q0cp73pq-cacheplane.vercel.app
Commit: b2474b6531adf95b14f295c5638367a3dc602373

Updated automatically by the deploy-preview job.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant