Nice work on the viz, and the fixes around it (redirect, sidebar exposing taxonomy.mdx, release ordering, the taxonomy.mdx rewrite) are solid on their own. Agree with Eric Godwin (@ericgodwin) that the explorer itself probably wants a different home. A few architectural points behind that, then some smaller things.

question: what does "canonical" mean here in practice? scripts/build-taxonomy.mjs reads a local checkout of tf-data-platform plus a hand-exported Athena counts CSV, and the output is committed. That's a manually synced snapshot with no provenance (source commit, checksum), and every release needs someone with pipeline access to re-run it and open a several-thousand-line PR. Who owns that per release, and where's the checklist?

suggestion: publish the taxonomy artifacts upstream with the release (S3 next to theme=places, or a public taxonomy repo with tagged releases), then have both the explorer and the download links point at those URLs. taxonomy.mdx already says the taxonomy is "published as a set of files with each release"; right now that's only true because the files were copied into static/. Once there's a real published source, dataUrl fetching makes sense and the docs repo stops being the distribution channel.

suggestion (non-blocking): the source of truth is already overture_taxonomy.json. I'd publish JSON (or Parquet for parity with the release) as the primary artifact and treat CSV as a derivative. A flat CSV with a " > "-joined path column is a lossy form for a 2,302-node tree. Related: taxonomy.json carries the counts but isn't downloadable, while the CSVs without counts are.

issue: _README.md and the script header point contributors at ../tf-data-platform/..., which is private. That's an internal runbook in a public repo and unusable by anyone outside.

issue: docs/guides/places/csv/2026-08-19-counts.csv (1,988 lines) isn't referenced anywhere on the branch. Also has the Athena _col0 header leaked in.

issue: taxonomy.mdx "at a glance" hardcodes 2,302 / 298 / 70,345,858 in prose while the generator already emits stats. Those will drift next release.

issue: taxonomy.mdx drops the "Provider Category Mappings" section (July 2026 BrightQuery/NAICS content) with no replacement. Seems outside the stated scope; was that intentional?

thought: on scope, this adds a 686-line d3 component, two runtime deps, and inlines ~2 MB of legacy CSV into one page via raw-loader. A standalone labs.overturemaps.org page that fetches from published URLs, linked or embedded from docs, keeps the docs build lean and lets the explorer iterate on its own cadence.

Suggested split:

• merge now: redirect, sidebar, ordering fix, taxonomy.mdx prose (minus hardcoded stats and downloads table, provider-mappings section restored)
• upstream: get taxonomy artifacts published with the release
• labs: explorer + viz fetching those URLs, docs links to it

Happy to help with the upstream issue or the labs scaffold.