Migrate direct raw-S3 bbop-sqlite consumers to the semanticsql CDN

[bot] Opened by a Claude Code agent on behalf of @kltm. Body updated 2026-08-18 (2nd revision): every entry below is now verified against fetched default-branch file content (literal-match), not code-search hits — search tokenizes on hyphens and produced several false positives, now removed. Verdicts also cross-checked against raw-bucket access logs (2026-07-08 → 2026-08-18) and per-file git history.

Companion to #112 (remove raw-bucket access). These projects hardcode the raw bucket URL directly, so they will not auto-migrate via oaklib/ODK bumps; they need to repoint to https://semanticsql.berkeleybop.io (drop-in, same paths) before raw access is removed.

Gotcha RETIRED 2026-08-18: the CDN no longer 403s Python-urllib/* User-Agents (host-scoped Browser Integrity Check exemption, verified). All client types work; the swap is a plain one-line change.

Verified still-present (literal raw URL in current default-branch runtime code)

• monarch-initiative/monarch-ingest — src/monarch_ingest/download.yaml L28 — outreach monarch-initiative/monarch-ingest#711 (open, no engagement) — phenio pulls also confirmed in access logs
• Knowledge-Graph-Hub/kg-alzheimers — src/kg_alzheimers/download.yaml L186/L255 — outreach Knowledge-Graph-Hub/kg-alzheimers#78 (open, no engagement) — traffic-confirmed
• monarch-initiative/dictybase-ingest — download.yaml L10; ⚠️ src/versions.py filters on contains=["bbop-sqlite"], so the URL fix must also update that selector or version-tracking silently breaks — outreach monarch-initiative/dictybase-ingest#3 (open, no engagement) — traffic-confirmed
• biobricks-ai/semsql — stages/01_download.sh (URL generator); ⚠️ stages/00_invalidate.sh does anonymous aws s3 ls (bucket-listing dependency, see below); .bb/source.jsonld metadata — outreach biobricks-ai/semsql#1 (open, no engagement)
• Knowledge-Graph-Hub/universalizer — universalizer/oak_utils.py L48 (note: fetches uncompressed .db objects) — outreach Knowledge-Graph-Hub/universalizer#41 (open, no engagement)
• microbiomedata/ontology-loader — src/ontology_loader/ontology_processor.py L100 — outreach microbiomedata/ontology-loader#59 (open, no engagement). Covers microbiomedata/nmdc-runtime too (its notebook just calls this library).
• ccb-hms/NHANES-metadata — code/generate_ontology_tables.py L30 — outreach ccb-hms/NHANES-metadata#37 (open, no engagement)
• waldronlab/bioanalyzer-backend — app/normalization/grounding/seed.py L57 (SEMANTIC_SQL_BASE_URL) — outreach waldronlab/bioanalyzer-backend#124
• seandavi/metacurator — src/metacurator/grounding/local_duckdb.py L26 (single base-URL constant; cleanest possible migration) — outreach seandavi/metacurator#29 — hancestro pulls confirmed in access logs
• seandavi/cdsci-lake — src/cdsci/lake/config.py L141 + sources/ontology/ingest.py; ⚠️ its registry IS the bucket listing (paginated ListObjectsV2) — outreach via comment on seandavi/cdsci-lake#40
• CultureBotAI/MediaIngredientMech — scripts/check_chebi_currency.py L53 — outreach via comment on CultureBotAI/MediaIngredientMech#206
• microbiomedata/external-metadata-awareness — notebooks/environmental_context_value_sets/generate_voting_sheet.ipynb (source cell) + ⚠️ notebooks/multi-lexmatch/interleave_s3_catalog_yaml_registry_bioportal_obo.ipynb calls the S3 ListBucket XML API; 2 docs files — outreach microbiomedata/external-metadata-awareness#553
• Knowledge-Graph-Hub/kg-registry — resource/reactome/reactome*.md product_url: frontmatter — machine-consumable registry metadata, not prose; directs downstream users to the raw URL — outreach Knowledge-Graph-Hub/kg-registry#701

⚠️ Bucket-LISTING dependents (URL swap alone does not migrate these)

Four verified consumers enumerate the bucket rather than (only) fetching objects: ontoProc2 (post-migration code parses ListBucket XML from the CDN root), cdsci-lake (ListObjectsV2 registry), biobricks 00_invalidate.sh (aws s3 ls), external-metadata-awareness (ListBucket XML notebook). Live probes 2026-08-18: the CDN root currently proxies the bucket's V1 listing (works only while the bucket stays public), and CloudFront strips query strings, so V2/pagination params are silently ignored (fine at ~332 keys; breaks at 1,000). #112 must decide a listing strategy: grant s3:ListBucket to the CDN origin access at lockdown, or publish a manifest file and migrate these four to it.

Migrated (verified in current code — done or nearly)

• Knowledge-Graph-Hub/kg-microbe — repointed to the CDN 2026-07-21 (9c8ddcad, via #595). Residual raw-log traffic through 2026-08-18 attributed to stale checkouts/deployments, not master.
• vjcitn/ontoProc2 — runtime repointed 2026-07-29 (1d4df64d, via vjcitn/ontoProc2#11, still open): remaining items are the README aws s3 ls s3://bbop-sqlite/ example and the listing caveat above. vjcitn/op2workshop README carries the same example.
• monarch-initiative/dismech, ai4curation/ai-gene-review — oaklib lock bumps landed 2026-08-07 / 2026-08-12; verified in traffic.

Docs/examples only (verified literals, prose/log context; fix opportunistically)

• cthoyt/pystow — docstring examples (api.py, impl.py)
• berkeleybop/metpo — 3 docs files (its script uses sqlite:obo: selectors — migrates with oaklib); turbomam/metpo-attic — 4 docs files incl. a copy-pasteable curl example
• brad-usredoxlabs/computable-lab — 1 doc
• monarch-initiative/ontogpt-experiments — committed stdout log of an old run (evidence, not config)

Removed after verification (false positives of hyphen-tokenized code search)

monarch-initiative/rare-disease-identification (docstring prose only; runtime uses a local OBO), monarch-initiative/mondo notebook (URL only in saved output cells; source uses sqlite:obo: selectors), several name-only doc mentions elsewhere. Selector-based (sqlite:obo:) consumers are out of scope here — they migrate via oaklib ≥0.7.2, with the standing caveat that pinned lockfiles do not auto-migrate (three CI consumers to date required manual lock bumps).

Coordination notes

• Outreach-by-issue has had near-zero engagement: of the 2026-07-14 wave, only kg-microbe (silent fix) and ontoProc2 (same-day author response) acted; the other six issues have zero comments after five weeks. Minimal PRs are likely the higher-yield instrument for the remaining repos.
• #114 remains the blocker for semsql download / library-default users: semsql on PyPI is still 0.4.0 (2025-02-05).
• The other ecosystem-wide unlock is an ODK release: the bundled-oaklib bump (INCATools/ontology-development-kit#1354) is merged but unreleased — the newest stable image (odkfull:v1.6.1) still bundles oaklib 0.6.23, so every ODK-based ontology repo's CI pulls from the raw bucket and cannot migrate by its own action until ODK ships (verified 2026-08-18: only odkfull:dev carries the fix). Release request filed: INCATools/ontology-development-kit#1368.
• Populations with no discoverable repo (cloud notebooks, one-shot users, private code) remain reachable only via release notes / announcements before #112.

— Posted by Claude Code agent on behalf of @kltm.