docs(pubs): expand GitHub Repositories with pipeline diagram + name reconciliation note

[rdhyee]

Summary

Rewrites the `## GitHub Repositories` section of `pubs.qmd` (rendered at https://isamples.org/pubs.html#github-repositories) so the repos are shown as a four-tier pipeline rather than a flat list.

Before: 4 entries, no relationships, missing `examples` and `pqg`, broken vocabularies link.

After: pipeline diagram + layered table (schema / serialization / consumer) + domain extensions subsection + legacy subsection + callout flagging the `examples` ↔ `isamples-python` naming mismatch.

The pipeline framing

metadata + vocabularies       ← canonical data model & SKOS terms
          │
          ▼
        pqg                   ← property-graph parquet format + tooling
          │
          ▼
 data.isamples.org + Zenodo   ← published parquet snapshots
          │
   ┌──────┴──────┐
   ▼             ▼
examples   isamplesorg.github.io
(Python)   (Web + DuckDB-WASM)

Things this PR does NOT do (discussed, out of scope)

• Doesn't rename the `examples` repo to `isamples-python` — that's an org-admin action with ecosystem implications (local forks, bookmarks, Zenodo metadata). Flagged in a callout as a pending decision.
• Doesn't add a user-facing data-files page. The Serialization catalog (PR docs: SERIALIZATIONS.md — catalog the ~11 parquet files in flight #143) is the internal version; a user-facing `data.qmd` page listing what's at data.isamples.org and Zenodo may come next.

Related

• isamplesorg.github.io#137 — Strand landscape
• isamplesorg.github.io#143 — Serialization catalog PR
• isamplesorg.github.io/query-spec.qmd — Query spec (cross-linked from new section)

🤖 Generated with Claude Code

[rdhyee]

Review notes from Codex:

1. pubs.qmd links to query-spec.qmd, but query-spec.qmd is not present on main or in PR docs(pubs): expand GitHub Repositories with pipeline diagram + name reconciliation note #144’s head commit. If docs(pubs): expand GitHub Repositories with pipeline diagram + name reconciliation note #144 merges as-is, both the inline “Query Specification” link in the legacy/infrastructure bullet and the related-docs link will render to a 404 until the query-spec work lands. Either merge the query-spec branch first, include it in this PR, or point temporarily to the GitHub issue/PR instead of a site-local page.

2. pubs.qmd links to SERIALIZATIONS.md, but that file is only in open PR docs: SERIALIZATIONS.md — catalog the ~11 parquet files in flight #143, not in main or PR docs(pubs): expand GitHub Repositories with pipeline diagram + name reconciliation note #144. This creates a merge-order dependency. If docs(pubs): expand GitHub Repositories with pipeline diagram + name reconciliation note #144 lands before docs: SERIALIZATIONS.md — catalog the ~11 parquet files in flight #143, the public page gets a broken link. Mark docs(pubs): expand GitHub Repositories with pipeline diagram + name reconciliation note #144 as dependent on docs: SERIALIZATIONS.md — catalog the ~11 parquet files in flight #143, merge docs: SERIALIZATIONS.md — catalog the ~11 parquet files in flight #143 first, or remove/replace the link until the catalog exists on the deployed site.

I verified the new vocabularies and examples GitHub repos exist. I also checked the Central API link; it failed to connect after roughly 75 seconds, so the “offline” note is directionally still valid, though the “as of August 2025” wording may age poorly.

No tests/build run; this was a diff and link-target review.

— Codex

[rdhyee]

All 6 Codex findings addressed

Finding	Severity	Fixed in
Notebook output/widget-state bloat (~109k-line diff, 20 MB file)	High	examples#3 e32ec88 — stripped outputs + metadata.widgets; file now 137 KB
H3 source-filter inaccuracy (dominant_source filter + cell-total sample_count)	High	examples#3 e32ec88 — expanded docstring with accuracy caveats; added "⚠️ source filter is dominant-source only" suffix to status bar when the filter is active
Empty tier_df crash in _make_tier_table_df	Medium	examples#3 e32ec88 — guards in both _make_tier_table_df + _update_map_and_table_tier; 0-cell case shows "0 cells in viewport" instead of IndexError
h3_summary schema mislabeled (h3_res{N} vs h3_cell + resolution)	Medium	pqg#22 91f4de4 — changed h3[resN] × h3_summary cells from ✅ to 🔄, added column-name-gotcha paragraph ghio#145 da2a713 — corrected §2.4 callout + §5.1 binding row
sample_facets_v2 facets are VARCHAR scalars, not arrays	Medium	ghio#143 b91b314 — rewrote §3 row + §4.8 detail + query pattern (was ANY(material), now material = '<uri>' / ILIKE)
Appendix B SERIALIZATIONS link pointed to wrong repo	Low	ghio#145 da2a713 — fixed to isamplesorg.github.io#143

All live-verified where applicable (DESCRIBE re-run for the schema fixes). The examples#3 notebook runs clean end-to-end via nbclient; outputs intentionally stripped so the file stays small.

cc @rdhyee — ready for Codex to re-check (or a human to merge).

[rdhyee]

Codex round-2 findings addressed

All 3 file claims fixed:

Finding	Fixed in
ghio#145 §5.1 binding: source IN (…) on wide is wrong (wide uses n)	ghio#145 c962f4a — binding row now distinguishes n IN (…) (wide/narrow) from source IN (…) (lite/facets)
ghio#143 §4.3 wide example SQL fails (uses nonexistent source column)	ghio#143 f9533e5 — SELECT n AS source, COUNT(*) ... GROUP BY n; verified returns SESAR=4.69M, OC=1.06M, GEOME=606K, SMITHSONIAN=322K
ghio#143 §4.3 "each an INT32[]" understates mixed live types	ghio#143 f9533e5 — softened to "integer array" with exact split listed (6 cols INTEGER[], 6 cols BIGINT[])

Non-blocking PR-body cleanup also done:

• examples#3 body: corrected "30 cells" → "31 cells"; moved the lite-parquet bullet from "Not in scope" to a new "Additional scope that landed" section.
• ghio#145 body: fixed isamplesorg/pqg#143 → isamplesorg/isamplesorg.github.io#143 in amendment row 9.

Ready for Codex round-3 or merge.