[rdhyee]

🤖 rbotyee (Claude, operated by @rdhyee) — Raymond's intent: fix a known stale figure in the public docs during the 2026-08 solidify pass, and get this PR to a CC×Codex LGTM before a human merges. What I did: replaced "106 M" / "~106M rows" with the verified count 101,387,180 in the four places it appears (commit 1); then, after an adversarial Codex review, fixed the figures in the same four files that had been derived from the wrong count or from an older, smaller file, and standardized their sizes to decimal MB (commit 2).

Evidence

• isamples_202512_narrow.parquet has 101,387,180 rows — scripts/isamples_facts.py (2026-07-18, 2026-08-25) and a live DuckDB count(*) against data.isamples.org on 2026-08-27.
• Live sizes (Content-Length, 2026-08-27): narrow 860,063,694 B; wide 291,818,634 B; wide_h3 306,553,295 B. The old "292 MB / 820 MB" figures were MiB labelled as MB; "691 MB" and "~9M edge rows" were from an earlier file.
• Edge rows = 101,387,180 − 20,729,358 = 80,657,822 → wide has ~80 % fewer rows; wide is ~66 % smaller than narrow (860 → 292 MB).

Files

• how-to-use.qmd (file table: rows, sizes)
• README.md (JavaScript snippet comments)
• tutorials/narrow_vs_wide_performance.qmd (format table; "~79 %" → ~80 %; "~9M edge rows (60 %)" → ~81M (~66 %); "691 MB" → 860 MB)
• DATA_PROVENANCE.md (pipeline diagram)

Numbers only; no prose or code changes. Draft, for a human to merge.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QJzwMsuFoSgTxX2p9paToy

[rdhyee]

🤖 rbotyee (Claude, operated by @rdhyee) — Raymond's intent: have Codex adversarially review this PR and iterate until both reviewers say LGTM before he merges. What I did: four Codex rounds (gpt-5.4 default model, reviewing a clean clone of this branch), fixing what each round found and pushing; then applied the CC+Codex+LGTM label and marked the PR ready for review.

Codex verdict at 9cd741b — LGTM

1. All four files: figures match established row/byte facts.
2. tutorials/narrow_vs_wide_performance.qmd:252,1062: arithmetic verified — 79.55 % → ~80 %; 66.07 % → ~66 %; 80,657,822 → ~81M; H3 adds 14.73 MB → ~15MB.
3. Table pipe counts consistent. Diff contains only intended numeric corrections; git diff --check clean.

What the rounds caught (all fixed): figures derived from the old 106 M (~9M edge rows, 60 %, ~79 %); sizes that were MiB labelled as MB (292/820 → 307/860 MB) or from an older file (691 MB, 278/280 MB); 20M for 20,729,358 rows → 20.7M; H3 delta 14 → 15 MB. Facts used: live DuckDB count(*) = 101,387,180 and live Content-Length for the three files, both checked 2026-08-27.

Round transcripts are local (/tmp/pr349_codex_r{2,3,4,5}.txt); nothing else was posted.