Repository navigation
Measure what a direct AllThePlaces extract adds beyond OSM and Overture #397
Description
What needs to be done, and why?
Determine whether a current direct AllThePlaces (ATP) run adds correct useful businesses, hours and contacts beyond the exact OSM/Overture releases already imported. Overture already carries ATP-derived provenance; the pilot must measure the remaining delta rather than count ATP records as wholly new.
Scope and open questions
One reproducible bounded regional comparison using a pinned published ATP extract, checksums/run timestamps, matching OSM/Overture releases and a labeled sample of difficult businesses. Use current conflation/evaluation rules as a starting point, including distinct chain branches, tenants/floors, closures and contradictory identities. Deliver a go/no-go report and reproducible acquisition/normalization/evaluation instructions. Out of scope: scheduled production ingestion, worldwide rollout or automatic OSM edits. Decide pilot regions and minimum quality/value thresholds before acquisition.
Verified current state
Rechecked against main at a8e03bc3efc49207d91f7c1855402c6cf149b8fd on October 7, 2026.
- Existing ATP source normalization and Overture source notices show ATP is already represented through Overture.
- Conflation safeguards, existing evaluation harness and fact enrichment are reuse points.
- Repository-wide provider/job searches found source notices but no dedicated direct ATP importer.
- ATP upstream license distinguishes CC0 output data from MIT spider software. Preserve provenance even where attribution is not mandatory.
What would show that this is complete?
- Record ATP run/source/checksum/schema and the OSM/Overture generations used; the run is reproducible.
- Report added distinct places and added/corrected hours/contacts separately from duplicates, existing ATP-derived Overture fields and unverifiable facts.
- Human-reviewed stratified samples report match precision, branch/tenant false merges, coordinate quality and closure errors; publish sample method/denominators.
- Track source IDs and observation times for each donated field; ambiguous identities remain unresolved rather than forced merges.
- Measure download/storage/runtime costs and regional limitations; do not infer global improvement from one region.
- Document a go/no-go decision with thresholds, licenses/provenance, and evidence for any subsequent production integration.
Verification approach
Run the existing conflation metrics on controlled/labeled fixtures and validate counts against sample identities. A pilot script must have deterministic positive/negative cases and resource measurements. For any committed tooling: pnpm lint, pnpm check-types, pnpm test; use a migrated disposable PostGIS database for real data comparison, and pnpm -C docs build for the report.
Dependencies
#393 provides the visual baseline. Related #301 concerns Overture release scaling, not direct ATP evaluation. A production ATP issue will depend on a positive pilot result; ambient tiles are useful later but are not needed to run this comparison.
Keep implementation within this scope; settle the listed design decisions during triage.
Activity
Medformatik commentedon Oct 7, 2026
Overnight preflight, rechecked against
bf47e937a2457f4a2539bcf43d1bf043ccea7248and the public instance on October 7, 2026:https://alltheplaces.xyz/and returns HTTP 200. Upstream licensing still distinguishes CC0 output data from MIT spider software. No extract was downloaded or counted as a measured improvement.2026-10-03T20:20:50Z, sequence4927. The public integration registry has 104 entries and no Overture integration entry; it supplies no imported Overture release or matching OSM POI generation. That does not prove no dataset exists elsewhere, or that the TileJSON input is the exact POI/conflation snapshot. The committed provider validates database release state, which is different evidence.I am leaving the measured pilot open rather than substituting a fresh unrelated Overture release or the inherited conflation corpus for the required already-imported baseline and new human-reviewed labels. Next prerequisites are a pinned regional export of the actual OSM/Overture generations (with checksums and source IDs), a fixed candidate ATP run, and independently reviewed branch/tenant/closure cases with denominators. Then set value/quality thresholds before acquiring/evaluating the extract and use a disposable migrated database. The discovery evaluation tooling in #424 will support reproducibility, but does not supply those missing real-data judgments.
[-]Measure the incremental value of a direct AllThePlaces extract[/-][+]Measure what a direct AllThePlaces extract adds beyond OSM and Overture[/+]Medformatik commentedon Nov 1, 2026
Closing as not planned after a backlog review, together with its follow-up #400.
packages/core/src/utils/overtureSource.tsrecords the ATP donor, record ID and licence.Reopen with a concrete report of missing or wrong POI hours or contacts that Overture's ATP-derived data doesn't cover.