Productionize direct AllThePlaces enrichment after a successful pilot #400

Provider / source name

Direct AllThePlaces (ATP) business extract; conditional production follow-up to pilot #397.

Provider website

https://www.alltheplaces.xyz/

Domain

POI / places; Knowledge / enrichment

API / data documentation

ATP upstream project, its published run format and data/software license. Pin the exact chosen run/schema/checksum during the pilot; a floating latest URL is not a release contract.

Access requirements

Fully open — no key required

Geographic coverage

Source-dependent business coverage; launch in the pilot region only. This issue does not claim uniform global coverage or guaranteed operating-hours freshness.

Data license & attribution terms

ATP output data is released under CC0; spider software is MIT. Preserve canonical source/run/observation metadata and donor provenance even where attribution is optional. Merged OSM/Overture content retains its own source-specific license/notice requirements. Do not infer permission to make automatic OSM edits from possession of a business fact.

Maps to an existing service?

Extend data-manager/PostGIS ingestion and the existing place/conflation/resolver pipeline. Decide whether the current generic POI-ingest acquisition/staging mechanisms can host this source; a new daemon is not assumed.

Would you like to build it?

Maybe, with some guidance. Conditional backlog; implementation has not started.

Verified need and scope

Rechecked at a8e03bc3efc49207d91f7c1855402c6cf149b8fd; existing issues/discussions contain only pilot #397 for direct ATP. Existing Overture source normalization already credits ATP-derived records, but no direct ATP importer/job is present. Reuse data-manager ingestion contracts, conflation safeguards and Overture fact enrichment, adapting contracts where necessary rather than weakening identity rules.

Productionize only after #397 demonstrates useful correct delta. Add source-aware normalization, staged/atomic generations, repeatable update jobs, field-level donation/conflict policy and rollback. Feed the place identities established by #399. Out of scope: rerunning ATP spiders in the shared API, replacing OSM facts blindly, scraping proprietary providers or expanding to a planet-scale deployment.

Acceptance criteria

• The pilot's go/no-go decision and selected region/schema are explicit; a no-go leaves production deferred.
• Acquisition verifies run/checksum/coverage and handles corrupt/changed/missing sources without replacing the previous generation.
• Updates are idempotent, resumable/cancellable and observable, with bounded resource preflight, retention and rollback.
• Canonical donor IDs and observation times survive into per-field facts; contradictory/missing/stale facts follow a documented precedence policy.
• Same-brand branches, tenants/floors, closures and coordinate ambiguity have negative conflation tests and sampled quality thresholds.
• API/search/details and ambient product resolve the same entity; incremental source changes invalidate affected caches/generations coherently.
• OSM contribution eligibility remains separate from display enrichment; attribution/provenance and operator docs are updated.
• Repeat the pilot's useful-place/hour/contact metrics after publication, with no false improvement from raw row counts.

Dependencies and verification

Blocked by positive ATP pilot #397 and the place identity/publication contract in #399. Related #301 (Overture scaling) and #303 (OSM replication) are separate work, not required planet-scale prerequisites. Resolve donor precedence and closure/freshness policy before implementation.

Use staging failure/retry/rollback and real migrated PostGIS tests, provider/conflation negative fixtures, source notices and resolver/cache tests. Run pnpm lint, pnpm check-types, pnpm test, applicable pnpm test:database, pnpm check:policy, and docs build; publish a bounded-region rollout/rollback report.