Crowdsourced Network Measurement

Every measurement begins as an app-initiated speed test on a subscriber's handset: geolocated result, technology generation and operator. nPerf aggregates those individual results into map tiles with per-operator and per-technology layers, covering dozens of countries with city-level drill-downs - India's page alone splits Vi, BSNL, Jio and Airtel.

Two clocks govern freshness. Coverage layers refresh hourly; speed maps refresh every 15 minutes. Underneath both sits a rolling two-year window of app-test data that is purged monthly, so the corpus always reflects current conditions rather than accumulating history. Raw data is not self-served: map viewing is free while bulk CSV, PRO dashboards and Analytics are quote-based products, and terms forbid extracting or reusing substantial parts of the database or commercializing test results without written consent.

App-test panels measure the tester population, not the network - and nPerf's own India page shows the unevenness, splitting one market across 4 operators (Vi, BSNL, Jio and Airtel) with sample depth varying per layer.

• Sample geography skews urban. Tests come from where people run apps, so rural blackspots are under-represented relative to drive-test alternatives.
• Device mix contaminates comparisons. Handset capability differs across operators' user bases; a coverage gap may be a device gap.
• Retention limits trend analysis. With 2 years retained and purged monthly, multi-year capacity planning cannot be reconstructed.
• Commercial reuse is contractually closed. French sui generis database rights plus explicit prohibitions mean the free maps support due diligence, not redistribution.

1. Ask what the panel actually sampled. nPerf derives maps from its own app's speed tests - request panel composition and test volumes per market before treating tiles as ground truth.

1. Verify refresh cadence per layer. Coverage refreshes hourly, speed maps every 15 minutes; any vendor claim slower than this describes a different product.

1. Check retention against your analysis horizon. A rolling two-year window purged monthly supports current-state views, not longitudinal studies.

1. Read extraction terms before building a dashboard. Terms forbid systematic extraction of minor parts and prohibit commercializing test results without prior written consent - quote-based licensing exists precisely for that.

1. Price the official export early. Bulk CSV data, PRO dashboards and Analytics are quote-based; get the quotation before committing architecture to scraped tiles.

Entries that sit next to Crowdsourced Network Measurement in this glossary: