[max-ostapenko]

BigQuery's bytes-processed estimates now account for cluster pruning on rank, so the guide's warning about the estimate bug is out of date.

Checked on the guide's own example (crawl.pages, 2023-05-01, desktop, rank <= 1000): the dry run estimated 92.9 MB and the run processed 91.2 MB. On the 2026-09 crawl, estimates matched processed bytes exactly for rank filters on both crawl.pages and crawl.requests, with and without client/is_root_page.

Changes:

• Clustering caution: keep the "no guarantee" link, say estimates now reflect rank.
• New tip: check the console estimate or bq query --dry_run (both free) instead of relying on figures in the guide.
• "Use RANK": replace the "leap of faith" paragraph with advice to widen rank step by step, plus a note about the old bug.
• sample_data: reposition as a small fixed set for developing query logic; note it's rebuilt each crawl.
• Comparison table: rank now ✅ for accurate estimates; closing line recommends rank, with TABLESAMPLE for random samples.
• Remove hardcoded scan sizes (329 MB, 10 MB, 6.56 TB, 680 MB). They no longer matched even for the fixed 2023-05-01 month (e.g. a11y is now ~3 MB, the full-month query ~0.7 TiB), so the text now compares relatively.