Every figure has a source.

The Overview page opened with "1 in 7 streets has tree shade". That figure was the tree density index of 14.8 converted into a fraction (1/7 = 14.3%) and then presented as a statement about streets. Nothing here is measured per street: the index counts registered trees per hectare. Replaced with exactly that, in its own unit - 19.6 registered trees per hectare, with the tree count and the land area alongside it so the arithmetic can be checked.

At the same time one tree register was being published under three different totals: 324,718 typed into two pages, 324,709 used by the data layer, and 324,526 returned by the API itself on 10 September. The register rises and falls daily. There is now a single dated value (324,709, August 2026) read from one source everywhere rather than typed per page. Deliberately not live: the same count is the denominator the per-neighbourhood tree density index is rescaled against, so a live value would shift all 33 figures every night by an amount no reader could see.

Thirteen of the 33 neighbourhoods read 0% natural green. The question was whether that is a real figure or an empty field being printed as a measurement. All 21 underlying district codes were queried against the BBGA API: all 21 return a published 0.0 for 2024 - no empty values, no missing records. So it is a real measurement. ORNTRGROEN counts "area of natural greenery with a high ecological value", measured on the BGT topographic map, and the thirteen are precisely the dense inner city and the redeveloped docklands: the green space there is landscaped.

The figure was right; the explanation around it was not. The code stated that a 0 meant "missing coverage" - that was false and has been corrected. And this page described the percentage as natural green relative to landscaped green; it is in fact divided by the neighbourhood's total land area (ORNTRGROEN ÷ ORLAND). Both are fixed above, and a neighbourhood page showing 0% now says why that is a measurement rather than a missing figure.

The pest section showed a "chance of outbreak" curve per pest, twelve months ahead. That percentage was a fixed number per pest multiplied by a bell curve - not a measurement, not a forecast, just a shape. Replaced with what does exist: the real, dated oak processionary caterpillar treatments from the city's open data, counted per month by treatment date (April to October: 0, 0, 2,052, 3,386, 1,232, 0, 0). April, May, September and October sit at zero because no treatment happens then, not because the measurement is missing. It is explicitly not a forecast: the dataset starts at the running season, and one season carries no prognosis. The section is therefore no longer called "Predictive Tree Health Forecaster".

Two source lines named datasets that are never fetched. The pest section cited Public Space Reports; the leaf-fall section cited the Waste guide, Waste containers and Public Space Reports - none of the three is queried by the data layer. Both lines are cut back to what actually sits underneath. The leaf-fall section also loses its "estimated composted tonnage" figure: that was tree density times residents divided by 8,000, an invented formula with an answer in tonnes. The per-species peak-week table stays, now stating that it is compiled arboricultural reference and not a municipal measurement.

Not a single figure on this dashboard changed - only the labels above them. What we called "tree canopy cover (%)" was never a measured canopy: it is the actual number of registered trees per hectare, rescaled so the citywide average lands on 14.8. That is a density measure, not a percentage of neighborhood area, so it is now called a tree density index and the percent sign is gone everywhere. The methodology line stating it was "tree canopy area as a percentage of neighborhood area" was false and has been replaced.

Two smaller corrections in the same pass. The heat risk score called population density a "BAG/WOZ proxy" when it is the real BBGA figure (residents BEVTOTAAL divided by land area ORLAND), and its source line named datasets that were never queried. And the park connectivity score said "physical distance" when it is the distance between park positions on this site's own map - not a walked route and not a distance in metres.

The previous gas calculator hashed the typed address to a random neighborhood and the "green roof savings" figure to a random number. Further research also found the underlying premise doesn't hold up: an extensive green roof mainly insulates in summer, barely in winter, because its vegetation layer is usually wet by then (Vakblad de Hovenier; Milieu Centraal). The 17 August rule that "a model using a real measurement as input" is exempt from the deletion policy is withdrawn for this section - no real measurement went in, and the model itself wasn't built on a real effect. Replaced with a section that looks up a neighborhood's real average gas use: address → coordinates (OpenStreetMap Nominatim) → official Amsterdam area (the municipality's own area boundaries) → CBS Kerncijfers wijken en buurten 2024. See the table above.

Also removed: the "nearest drop-off point" leaf-waste finder. None of the five collection points it showed appeared in either of Amsterdam's two official sources (the Recyclepunt register, the neighborhood-compost dataset), and the result was hashed rather than based on the chosen neighborhood anyway.

Neighborhoods now shows a real leaderboard of which neighborhoods gained the most green space per resident - calculated from two real BBGA annual figures per neighborhood (current year vs. one real year prior), not modeled. Since BBGA publishes annually, not quarterly, the leaderboard honestly says "since <year>", never "this quarter". While building this, it turned out the Lambda's BBGA calls had no timeout - a single stalled connection could hang the entire daily refresh up to the full Lambda timeout (this actually happened twice in production, at 180s and again at 300s). Every BBGA call now has a hard 10s timeout so one slow request can never block the whole run again.

Pedestrian activity per area (added earlier the same day) was removed again the same day: checking the full crowdmonitor/passanten dataset (not just the Vondelpark sensors) showed the single most recent reading, across every sensor, is from 28 March 2024 - the entire dataset has been frozen for over two years, not just occasionally stale. That's no longer a "real but sometimes dated" case, it's an effectively discontinued source, and doesn't fit how this dashboard defines "real".

More new sections on real data, after another research pass. The Overview page now shows "How green is Amsterdam compared to other cities?" - independent CBS figures (not from Amsterdam's own systems) for Amsterdam, Rotterdam, Utrecht, Den Haag and Eindhoven. The "By district" view now also shows real housing development pressure (project count and planned homes per district, 958 real projects citywide). Infrastructure & Heritage has a new Energy Labels section: the citywide distribution of ~330,000 real EPBD energy labels, as a measured reference point alongside the illustrative green-roof savings figure. (A fourth section, Pedestrian activity per area on Neighborhoods, was removed again the same day - see the entry above.)

After researching additional real data sources for wildlife, trees, water and light pollution, four new sections were added, each individually verified with a live test request before anything was built. Trees & Ecology now also shows the City of Amsterdam's real wildlife crossings (639 fauna passages, fish passages, etc., point-location joined to neighborhoods the same way as the nest boxes) and real tree safety inspection figures (148,000+ in the past year) and storm damage reports. The "By district" view now shows real ecological core areas and corridor zones per district. Climate & Equity has a new Light Pollution layer with a real sky-brightness reading per neighborhood from Province of Noord-Holland. A real water-quality source (swimming water, pollution) was also researched - no usable public API was found at Waternet/AGV, so that section wasn't added rather than building something incomplete.

The Waarnemingen (sightings) feed - 28 fictional wildlife sightings - was removed earlier this week because Observation.org's public search page explicitly blocks AI crawlers (robots.txt) and their official API is manual-approval-only. As a real, verifiable replacement, Trees & Ecology now shows a new "Newly planted" layer: real recently planted trees per neighborhood (rolling 2-year window), with a real species breakdown, straight from the same official tree register the Tree summary feature already uses. Less "lively" than a sightings feed, but fully traceable - every neighborhood, count and tree species is checkable in the source data.

The "cleanest vs. fastest route" comparison on the Climate & Equity map showed traffic density and canopy buffer per street segment, entirely invented. Amsterdam publishes no street-level air-quality data, but RIVM Luchtmeetnet does publish real, hourly PM2.5 and NO2 readings from the city's 11 real monitoring stations (GGD Amsterdam, including Vondelpark, Westerpark, Stadhouderskade, Van Diemenstraat). The map now shows these 11 real stations with their most recent reading; not every station measures both pollutants - that's shown honestly as missing, not filled in. The route comparison itself has been removed, not replaced, since no street-level data source exists for it.

The Winter sun/grit layer on the Climate & Equity map showed sun exposure and gritting policy per street segment, entirely randomly generated. After research, no public dataset exists with street-level sun/shade data for Amsterdam (Amsterdam's only open solar data is rooftop-panel suitability, not a street-level shadow map), and the catalog entry for a municipal gritting-route file turned out to be a dead link. Under the policy above - illustrative content with no real data source gets removed, not labeled - the entire layer has been removed. Winter black ice itself does now run on real data, see the Water & Paving section above.

Two sections that were previously fully illustrative now run on real measurements. The Winter Black Ice Prevention Index uses PDOK's official AHN elevation model (0.5m resolution, meters NAP) and real black ice reports from Public Space Reports ("Gladheid winterdienst", "Gladheid door blad", "Gladheid door olie op de weg" - 7,792 real reports citywide), both sampled per neighborhood; the fictional "risk intersections" (randomly composed street names) have been removed, not replaced. In Find a quiet green spot, the "Quiet" score now comes from RIVM's official Lden noise map (traffic, rail and industry, dB) instead of a randomly generated figure; shade and seating remain our own estimate, now explicitly labeled as such in the detail panel. Both the elevation and noise measurements come from the same new Lambda integration with the PDOK and RIVM WMS services.

The City of Amsterdam publishes a real, live structural register of quay walls (civieleconstructies/kademuur, ~1,579 individually registered sections) - including material, foundation type, and whether a section is in scope for the real Programma Bruggen en Kademuren (PBK), the reinforcement program launched after the 2020 Grimburgwal quay wall collapse. Via a point-location spatial join (each section's midpoint), the map and detail panel now show the real count, material and PBK scope per neighborhood. There is no public numeric risk figure in this open dataset - the red/orange/green classification the municipality uses internally exists only in PDFs, not the API - so the earlier fictional "vulnerability score" and its accompanying risk-matrix chart (which also blended real tree canopy with random noise into a "tree biomass" figure) have been removed, not replaced with a new estimate.

Until now, sections with no real data source got a clearer "illustrative" label when no alternative could be found. From now on a stricter rule applies: if something is illustrative and isn't one of our own suggestions (like the recommended interventions), it needs a real data source or the section gets removed - not just more honestly labeled. A model that uses a real measurement as input and applies a published, citable formula (like Tree Summary's CO₂/water/shade calculation or the green-roof gas savings) doesn't count as "illustrative" under this rule - that stays.

Two sections removed under this rule: the Sightings feed ("recently spotted") - no real, legally usable data source available (see below); and "Plantable volume vs. greening target" - no open dataset on underground planting space or utility lines was found (the only "soil" dataset in Amsterdam's open data catalog turned out to be soil contamination surveys, not underground space).

For the Sightings feed, we also considered whether Observation.org's public search page (not their gated API) could be read directly via its search-result URL parameters. We didn't do this: their robots.txt explicitly excludes AI crawlers by name (including ClaudeBot/ Claude-Web/anthropic-ai) with a site-wide Disallow: /. That's a clear signal that automated access - even via the public pages - isn't wanted, so we respect that instead of looking for a workaround.

Tree species names came straight from the real Dutch tree register and were shown untranslated on the English site too (e.g. "Iep" instead of "Elm"). Added a translation table for all 20 real genera, applied everywhere a species is shown (Tree summary, the map view, leaf-fall prediction, the special-trees register, and the host-species field in the pest forecast).

12 of the 20 real tree species (including Poplar and Cherry) turned out to have no matching photo, showing a broken image. Added a real, individually verified Wikimedia Commons photo for each missing species (via the Commons search API, not guessed - guessed filenames turned out not to exist). An image that still fails to load now disappears cleanly instead of showing a broken icon.

Investigated whether Observation.org/waarneming.nl could be a real data source for the Sightings feed: their API requires manual approval (no self-service signup) and individual sightings can carry different licenses, including non-commercial ones - not a feasible quick fix for this dashboard. The feed stays illustrative, but now shows a clear note about that directly next to the feed itself, not only on this page.

Both per-neighborhood and citywide, the full list of real facility types and counts is now shown (for example "Nest box × 32 · Insect hotel × 8"), not just the most common type. The separate "Ecological facilities per neighborhood" card has been removed - the same information (including the citywide type tally) now lives more clearly in the map's own Facilities tab. A link to report a problem with a facility to the municipality has also been added - we looked into whether a direct connection to Signalen (SIA), the municipal reporting system, was possible, but there's no public reporting API for it (only a read-only API for existing reports, and a login-gated API for municipal staff) - the link redirects instead to the official report form at meldingen.amsterdam.nl, with a tooltip that says so honestly.

Trees & Ecology - "Ecological facilities" (previously "Wildlife Facility & Corridor Health Monitor") has been rebuilt on real data. Amsterdam publishes an official register of ~4,588 individually registered nest boxes, bat boxes, insect hotels and more (ecologie/faunaverblijfplaatsen). Via a spatial join (point location within neighborhood boundary), the map and list now show the real count and type per neighborhood; neighborhoods with no match honestly show "no known facility" instead of a made-up example. There is no live condition/status field in this source (empty for all 4,588 objects), so the earlier "Active/Reported/Under repair" status and the "Report damage" button - which didn't drive anything real - have been removed.

Trees & Ecology - Tree summary: species and average planting year per neighborhood now come from a real sample of the official tree register, no longer a name-hash-derived fabrication.

Also: fixed a real bug where the "Plantable volume" gauges didn't respond to the 1m/2m/3m depth toggle; added click-for-route-info to both the Clean air and Winter sun maps (previously hover-only); and added utm_source=hoegroenisams.nl to outbound links that didn't already have tracking.

The 8 districts (map, "Per District" list, comparison chart) never fetched live data - only the 33 neighborhoods already did. Now fetched from the same official BBGA API, linked to real district-level area codes (for example "A" for Centrum; Weesp composed from its 3 already-live neighborhoods, since Weesp has no district-level area code of its own). This changed some figures substantially - Noord in particular (was 38.4 m², now considerably higher) because the BBGA definition of public green space also counts agricultural green space, and Noord includes the rural area Landelijk Noord. See "Data status" on the Sources & Methodology tab for the full explanation.

A box labeled "Monitor Groen standard (City of Amsterdam)" cited specific figures (61% of public space green, a 40% priority threshold, an NDVI-trend criterion of >5%) that aren't actually computed anywhere in the app - no NDVI is measured. Replaced with a real, traceable figure: 44.6% of Amsterdam's land area is public green space (BBGA ORPUBGROEN ÷ ORLAND, area "STAD", 2025). The NDVI criterion was removed; the risk classification now uses only the real, already-existing tree density index terciles.

The district map's color gradient (heatColor) still used fixed breakpoints (12/22/26/40 m²) from the old, since-replaced scale. That capped most districts at the same dark-green color, regardless of how far they actually were above the top-third boundary. The gradient now scales with the real tercile boundary for the 8 districts (30.7 m²), consistent with the list, legend and comparison chart.

The tercile boundary computed from the 33 neighborhoods (which includes outliers) was also being applied to the 8 districts (which are averages) - so no district could ever reach "top third," making the whole city incorrectly look like it was falling behind everywhere. Fixed with a separate tercile pair computed specifically for the 8 districts.

The citywide figures previously shown on this dashboard (baseline 22.1 m², current 25.2 m², targets 26.0 m² for 2030 and 28.0 m² for 2050) could not be traced to a verified official source. The per-neighborhood figures were already live-linked to the real BBGA indicator ORPUBGROEN (including agricultural/natural green and private gardens) - but the citywide comparison figures they were measured against used a different, smaller scale that doesn't trace back to anything real.

The corrected citywide average (87.5 m² in 2025) is calculated with the same method as the live per-neighborhood figures (BBGA ORPUBGROEN ÷ BEVTOTAAL, area "STAD"), and matches the City of Amsterdam's own Monitor Groen report (8,240 hectares of public green space as of 1 January 2023, ÷ 918,194 residents ≈ 90 m²/resident - the same order of magnitude as our 89.5 m² for that year).

The 2030/2050 targets of 26.0/28.0 m² have been removed - we could not find an official City of Amsterdam source establishing an m²-per-resident target for the Groenvisie 2020-2050. As far as we could determine, the actual Groenvisie uses a walking-distance norm (10 minutes to parklike green) instead of a square-meter figure - Monitor Groen reports 95% of Amsterdammers met this as of 1 January 2023. Status badges on this dashboard ("top/middle/bottom third") are now based on an honest comparison between the 33 neighborhoods, not an unverifiable absolute target. Amsterdam's own Court of Audit (Rekenkamer) has since confirmed no binding green-space norm exists either (see the Overview page).

The rainwater pkpi (previously "4.5bn liters/year") has been recalculated.No official citywide measurement of how much rainwater Amsterdam's green space captures annually exists - not from Waternet, nor the Amsterdam Rainproof portal. A real, widely-cited per-tree figure does exist from published i-Tree research on Manhattan street trees: an average of 6,842 liters/year per tree (2,228 liters canopy interception + 6,842 liters including tree-pit infiltration) - in line with an independent USGS estimate of roughly 6,000 liters/tree/year in avoided runoff. Multiplied by Amsterdam's actual registered tree count (324,709, official tree register), that gives ≈2.2 billion liters/year - an estimate built from two real, cited figures, not an Amsterdam-specific measurement. The earlier 4.5 billion figure couldn't be traced to a source; we suspect it mistakenly applied the Hemelwaterverordening's 60 liter/m² norm (a storage requirement for new construction) to all existing public green space, which isn't methodologically valid.