# What changed in the city scanner. Source: [Home](https://www.nfw.run/en)[State of the web](https://www.nfw.run/state-of-the-web)[Method](https://www.nfw.run/state-of-the-web/method) // changelog Every city report names the scanner version that measured it, and its address ends with that version: only the reports from before the scanner had version numbers have none. Here is every version, newest first, with what it changed in the measurements: the figures of two reports compare well only if the measuring did not change in between. Version 1.2.0 is live today. ## What improved, in brief Verona was scanned twice in September 2026, once before the scanner had version numbers and once on version 1.0.0, over the same 7,377 businesses with a website: the most direct comparison between versions. Here is what changed. - More sites get past bot protection The scanner presents itself consistently as the Chrome it runs, browses in Italian and, when a homepage refuses a plain request, tries again in a real browser. In Verona the sites turned away fell from 529 to 407, and the report now says which protection stopped them (SiteGround, Cloudflare, Vercel and others). No CAPTCHA solvers and no disguises: a site that still says no is counted as blocked. - More sites measured Working sites analysed went from 4,640 to 4,739, and sites measured by Lighthouse from 4,314 to 4,481. - Several machines measure together Lighthouse no longer runs inside the scan but in a shared queue that more machines can join. Each one is calibrated on PageSpeed Insights before it measures, so they agree: on the same Verona sites three machines averaged 47.8, 46.9 and 46.9. - Steadier scores Lighthouse has reserved cores, measures every homepage in rounds of 3 runs, up to 9, and we publish the median run. Runs made while the CPU was under load are set aside. Before, the same site measured twice 10 minutes apart scored 86 and 93. - Scores closer to PageSpeed Insights From version 1.1.0 calibration matches the blocking time, the measure that was furthest off: the average score error against PageSpeed, on pages never seen while tuning, fell from 11 to about 8 points. That is also why Verona's median speed score went from 41 to 53 between the first scan and the one on version 1.0.0: the sites did not get faster in a week, the measurement got closer to Google's. - Faster scans that compare The network and browser parts of a scan have separate slots, and Lighthouse a stage of its own: Verona's second scan finished in under 23 hours, the first took three days, with a restart halfway. And a new scan counts the same businesses as the one before, so two scans of a city compare. 1. ## v1.2.0 currentdeployed Oct 2, 2026 - Every site is also checked for an ARD (Agentic Resource Discovery) manifest, with the same conformance test Lighthouse ships: the reports' AI score counts it under rules 2.0.0. - Lighthouse no longer runs its Agentic Browsing checks: llms.txt and the ARD manifest are checked once per site, not on every run, with 3 fewer requests per run. Lighthouse scores do not change, because those checks ran after the measuring. 1. ## v1.1.0 deployed Oct 1, 2026 - The machines' calibration on PageSpeed Insights matches Total Blocking Time, no longer the score. On the same Verona sites the machines agreed with each other, but not with PageSpeed: the blocking time was 4 to 5 times its own, and a page's score was 11 points off on average. With the new calibration, on pages never seen while tuning, that fell to about 8, close to PageSpeed's own variability. - Lighthouse 13.5.0, the same version as PageSpeed Insights. Reports measured with this version - [Pescantina, October 2026](https://www.nfw.run/state-of-the-web/pescantina-city-october-2026-v1-1-0) 1. ## v1.0.0 deployed Sep 30, 2026 - Lighthouse has reserved cores, in a queue several machines share: before, it shared the CPU with the next sites being scanned, and the same site could move 10 points between two measurements. - Every homepage is measured in rounds of 3 runs, up to 9, and the report uses the median run, taken whole, instead of the average. Runs made while the CPU was under load are set aside. - Every machine is calibrated on PageSpeed Insights: it picks the CPU slowdown that makes its scores match PageSpeed's, instead of a fixed 1.2x slowdown. Reports measured with this version - [Verona, September 2026](https://www.nfw.run/state-of-the-web/verona-city-september-2026-v1-0-0) 1. ## before v1.0.0 deployed Sep 29, 2026 - Lighthouse measures every homepage 3 times and the report uses the average, with PageSpeed Insights' CPU slowdown (1.2x) instead of Lighthouse's default (4x). - Speed Index and Total Blocking Time as Lighthouse measures them: before, the blocking time was an estimate. - Sites behind a bot protection (SiteGround, Cloudflare and the like) are tried again in the browser, and the report counts the ones turned away even so, protection by protection. - The scan browses in Italian: multilingual sites are audited in their Italian version. 1. ## before v1.0.0 deployed Jul 29, 2026 - The version of the first city scans: every Overture Maps business that lists a website is visited, and the homepage of every working site goes through Lighthouse, structured data, the AI score and spam blocklists. - Lighthouse measures the homepage once, with the default CPU slowdown (4x) and on the same machine that is scanning other sites meanwhile: speed and blocking time come out slower, and scores lower, than PageSpeed Insights'. - Speed Index and Total Blocking Time are not recorded: the blocking time is an estimate. - Sites behind a bot protection are not tried again in the browser, and multilingual sites are read in the browser's default language. Reports measured with this version - [Verona, September 2026](https://www.nfw.run/state-of-the-web/verona-city-september-2026)outdated [Back to the method](https://www.nfw.run/state-of-the-web/method) [The same scanner runs the free SEO audit: its changelog](https://www.nfw.run/seo-audit/changelog) [The AI score rules keep a changelog of their own](https://www.nfw.run/state-of-the-web/ai-score/changelog)