Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

Widespread drop in brand field coverage for Brazil between releases ~90% of a random sample of well-known chains lost matched POIs (median -21%, some down to zero

Aperta
  • Qualcosa non va?
#546 2 commenti 0 reazioni 2 assegnatari Vedi su GitHub

@ejdanforth ci sta già lavorando.

Dal 18/8/2026.

Valutazione

Questa issue non è ancora stata valutata.

Descrizione

external places

Summary

We track POI counts for a catalog of ~100 known Brazilian retail/service chains, matched against the Overture places theme via the brand field (with a name-based fallback for gaps). Comparing a snapshot from early July 2026 (read directly from the Overture places dataset) against the current data (accessed via the CARTO Databricks Marketplace mirror, release 2026-06-17.0 / schema v1.17.0, table last committed ~2026-06-18, behavior still present as of 2026-07-30), we see a broad, consistent drop in matched POI volume across almost all chains — not isolated to one brand or sector.

  • 94 of 100 chains sampled show a drop in matched POI count.
  • Median drop: ~-21%.
  • Several large, well-known chains show much steeper drops, down to -80%+.
  • A handful of chains (5 of 100) actually increased, so this isn't a uniform scaling issue — it looks more like a matching/coverage regression affecting most, but not all, brands unevenly.

We are not 100% able to isolate whether this originates from the places release itself or from the CARTO mirror's processing, since we only have access to the two snapshots via these two different access paths (direct Overture read months ago vs. CARTO mirror now) — flagging here first since OvertureMaps/data seemed like the right place to check whether this is expected/known.

Dataset / environment

  • Old snapshot: read directly from Overture places, ~early July 2026.
  • Current snapshot: overture_maps_places.carto.place (CARTO Databricks Marketplace mirror), delta.lastCommitTimestamp ≈ 2026-06-18 (shortly after the 2026-06-17.0 / schema v1.17.0 release).
  • Country: Brazil (addresses[0].country = 'BR').
  • Note: our current-side counts pass through our own name-normalization/chain-matching layer (which includes a confidence >= 0.20 filter and a name-based fallback for missing brand), so the exact numbers below aren't a byte-for-byte comparison of the raw brand field alone — but the fallback logic didn't change between the two snapshots, so the relative drop should still be attributable to the underlying data, not our pipeline.

Evidence — sample of the steepest drops (approximate old vs. new POI counts matched to the chain)

Chain Sector ~Before ~Now Change
Smart Fit Gym ~1,390 ~250 -82%
Instituto Embelleze Beauty school ~275 ~65 -76%
Drogarias Ultra Popular Pharmacy ~320 ~120 -62%
Burger King Fast food ~2,450 ~1,080 -56%
Espaço Laser Beauty/aesthetics ~465 ~250 -46%
Chevrolet Auto dealer ~495 ~290 -42%
McDonald's Fast food ~5,570 ~3,260 -41%
Carrefour Supermarket ~750 ~450 -40%
Outback Restaurant ~380 ~235 -38%

For context, more "typical" cases in the middle of the distribution (median): chains like Óticas Carol, Crefisa, Americanas, Farmácia São João, Cacau Show — all landed around -19% to -21%.

A few chains bucked the trend and increased (Bob's +54%, Oggi +266%), suggesting this is not a uniform re-scaling — it looks like uneven matching/coverage changes per brand.

Broader context: overall brand coverage in Brazil is much lower than in the US in the current release

Country Total POIs POIs with brand.names.primary %
US 14,894,884 1,522,538 10.22%
BR 5,528,005 262,406 4.75%
MX 3,072,777 144,153 4.69%
AR 896,794 29,573 3.30%
PT 398,362 21,215 5.33%

Reproduction query (Databricks SQL / Spark SQL, works the same against raw Overture Parquet)

SELECT brand.names.primary AS brand, COUNT(*) AS matched_pois
FROM <places_table>
WHERE addresses[0].country = 'BR'
  AND UPPER(brand.names.primary) = 'SMART FIT'
GROUP BY brand.names.primary

Expected behavior

Brand coverage for well-established, high-confidence, correctly-categorized chains should be stable (or only decrease slightly) between releases — not drop by double digits (up to -80%+) for the majority of a random sample of ~100 chains.

Additional context

Happy to share our full comparison list (100 chains), or run more diagnostics (e.g. checking whether the drop correlates with a specific sources[] composition change, specific regions within Brazil, or POI age) if useful for triage.

Lingua principale
Python
Stelle
1.2k
Fork
48
Merge medio
10m
PR unite (30g)
2

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di OvertureMaps/data

Tutte le issue di OvertureMaps/data

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.