[joseph-isaacs]

What changes

bench, in two parts.

SQL File Size group

A new SQL File Size group charts the on-disk size of the files each SQL benchmark writes. The data comes from the file_sizes table added in #221. Develop runs fill it since vortex-data/vortex#11443, and vortex-data/vortex#11447 adds every format, not only Vortex.

Charts and lines. The table has one row per file, which would be far too many lines; ClickBench alone writes about 100 files per format. Instead:

• One chart per benchmark and scale factor, about 16 today.
• One line per format, summed over the benchmark's files. That's vortex and vortex-compact today, plus Parquet, DuckDB and the rest once #11447 lands, so about 4 lines per chart.
• Chart order: clickbench, tpch, tpcds, then the other benchmarks by name, each by numeric scale factor.
• Names: TPC-H, TPC-DS and nested TPC-H always name their scale factor (tpch sf=1.0, tpch sf=10.0). Benchmarks without scale factors use the plain name (clickbench).

Summary card. It reuses the compression-size card: each format's total, plus its min, geomean and max ratio to vortex, each scope at its newest commit with a vortex size. The "vs Arrow" column shows "—", because file sizes record no Arrow size. A size summary with its own baseline gets no headline. Without that, this group's header read "same size as Parquet".

Implementation:

• file_sizes is the sixth family in families.ts, slug.ts, queries.ts (chart, batched group query, discovery) and summary.ts.
• Discovery and the summary use a recursive skip scan over idx_file_sizes_baseline instead of reading every file row.
• Docs: CLAUDE.md, CONTRACT.md and design-decisions.md now say six families.

Group order

The landing page now runs:

1. Compression
2. Compression Size
3. SQL File Size
4. Random Access (NVMe)
5. Random Access (S3)
6. The query groups, in their existing order
7. Groups outside the fixed list, by name
8. String Encoding
9. String Encoding Size

The string-bench groups now come after everything else, from a separate GROUP_ORDER_LAST list.

Cache

GROUP_PAYLOAD_CACHE_VERSION is now v14, so a deployment of this change doesn't serve group lists cached by older code, which lack the new group and order. The preview hit exactly this before the bump.

How it was checked

• bun run test:bench against a local PostgreSQL 16: 1004 passed, 9 skipped. The new tests cover:

  • per-format sums;
  • the group payload matching chartPayload byte for byte at ?n=1…4,all;
  • chart order and names;
  • the summary against real rows;
  • no headline for a summary with its own baseline;
  • the group order, with the string groups last.

• bun run check:ci passes, and tsc --noEmit reports no errors in bench.

• bun run dev against a seeded database: the group, card and charts render.

• Production has rows since develop's run for Store SQL benchmark file sizes in the benchmarks database vortex#11443.

• Not run locally: build, knip, Playwright and Lighthouse; CI runs them.

• Every commit is signed off (git commit -s), for the DCO check.

• bun run check:ci passes.

• Each new image or other asset has an annotation in REUSE.toml. (No new assets.)

🤖 Generated with Claude Code

https://claude.ai/code/session_01UeniB9quRiv5fCx8AhzH9s

[vercel]

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

Project	Deployment	Actions	Updated
vortex.dev	Ready	Preview	Nov 13, 2026 2:43pm UTC

[github-actions]

🔦 Lighthouse audit

URL	Perf	A11y	Best practices	SEO
/	🟢 100	🟢 100	🟢 100	🟢 100
/bench	🟢 100	🟢 100	🟢 100	🟢 100
/bench/chart/ct.eyJrIjoiQ29tcHJlc3Npb25UaW1lIiwiZGF0YXNldCI6ImFyYWRlIiwiZGF0YXNldF92YXJpYW50IjpudWxsfQ	🟢 100	🟢 100	🟢 100	🟢 100
/blog	🟢 100	🟢 100	🟢 100	🟢 100
/blog/btrblocks-compressor	🟢 100	🟢 100	🟢 100	🟢 100

Run 43445344911 · 2026-11-13 14:46 UTC