Site finders and a building map

Why these maps

Report 4 (June 2026) scored every square of a grid over the Tallinn area from 0 to 100 on three indices. Fair-Share, the EKA team's own (on the Tallinn maps, "Report 4 regreening index"), weighs homes, children's places, everyday services and, as a need, nature places (the fewer, the higher the score). Team 1 (the Active city) and Team 2 (the Everyday-life city) were proposed by teams in an EDITUA workshop. Report 4 calls the squares "parcels" or "cells"; here a parcel is a cadastral parcel, a plot in the national land cadastre.

A high score says that new green space would reach many people there, not that there is land to green. Report 4 named vacant land, ownership and surrounding density as next layers, its "feasibility axis", without computing them; the site finders add candidate land, owner types and density layers.

The site finders were made for the EDITUA × Street Studio workshop "From Maps to Streets: Testing Greening Potential at Urban Junctions". Besides candidate land they list street junctions.

Report 4's squares are mostly 125 m across, some up to 2 km, and everything in a square shares one score. The building map tests how much scores change when the walk starts from each building instead of one point per square: a sensitivity test, not a validation against independent data.

Two findings made after Report 4 matter for both maps; the first is already in its map story:

• Fair-Share's top band (scores above 80) is driven by homes, children's places and everyday services, not by missing nature. Low bands mostly mean few people.
• "Nature places" are mostly outdoor gyms, courts, skate parks and fire pits (298 of 395 places in Tallinn and 3 km around it). The list has no parks or forests; both maps now say so, and in Tallinn both let you add tree cover to the index.

The site finders, local and regional

There are 16 site finders, one for Tallinn ("Tallinn focus") and one for each county, reached from a picker at damianocerrone.com/dashboard/editua/map/ (figure 1, table 1). All 16 work the same way.

Candidates are of three sorts: unbuilt cadastral parcels; surface car parks of 1,000 m² or more; and, in Tallinn only, pieces inside built parcels (parcels with buildings), cut from the city's land-cover map as open ground, lawns, strips along streets or paved forecourts.

How to use them, step by step

1. Pick a region. A welcome pop-up offers a 15-step tutorial.
2. Pick or compose the index. Tab "1 · Index" (figures 2 and 4) offers Fair-Share, Team 1, Team 2 and "Your own". Each is a weighted mix of ingredients (homes, a group of places or one kind of place within a walk, or a map layer such as tree cover); any ingredient can be reweighted, removed, added or flipped into a need.
3. Filter the score. Set the range with the two sliders or a button: Top 20%, Top 10%, 60–100 or All. A first visit opens at Fair-Share 60–100 with the start filters (the kinds on at the start in table 2, likely gardens hidden, minimum size 250 m²): the start view.
4. Look at the candidates. Tab "2 · Sites" (headed "Land that may be vacant or under-used" on screen) lists the candidate sites scoring inside the range, coloured by owner type, with filters for owner, kind, detail plan, size and, in Tallinn, general-plan zone. For more sites, press Top 20% in "1 · Index", switch on more kinds, or switch off "Only show what scores inside the filter range" (figure 3).
5. Check a site from the desk. The card (figure 3) gives size, owner type, use, ground cover, plans, any contamination and each ingredient's level, with oblique aerial photos inside the page and links to the cadastre map, a 3D view and Street View.
6. Go and see it. Tab "3 · My list" makes walking routes and holds field notes per site: "Does what you see support the map?" (yes, partly, no); "Need" (for greening), "Likely impact" and "Transformability", each rated 1 to 5; and text boxes by theme, such as "Sealed surfaces" and "Practical constraints". The About tab gives plain visit rules (not legal advice): look from public ground first, enter a yard only with the owner's permission (ref. 4), keep out of building sites and ruins, and leave people, especially children, out of photos.
7. Share and hand in. Links carry the index, filters and list; each region keeps its own list. Notes stay in the browser, so download them (CSV or GeoJSON) before handing in or changing device.

Street junctions

In "2 · Sites", "Street junctions" lists the named junctions in the score range instead of the sites, highest score first. Each takes the score of the grid square it falls in, so many tie; ties are listed by number of arms, the streets meeting there (figure 6). The candidate sites stay on the map, so a junction can be read against the land near it. Junctions go into "My list" like sites, and a revised index under "Your own" re-ranks them.

Table 1. The 16 site finders

Start view: the sites a first visit lists (step 3). Counties in the picker's order. County scores are relative to the county, and county squares are larger (in most counties the 1 km square is the commonest size), which is partly why county start views hold more sites than Tallinn's. The Tallinn grid and the Harju grid overlap in the suburbs, so grid squares are not summed over all 16. Routes from: where "3 · My list" starts its walking routes when the phone's location is off (in a county, the centre of the town named).

Kinds and owners

Each site has one kind (table 2). The kinds off at the start are car parks, industrial or commercial yards and paved forecourts, which were mostly in use on the photo; land that is green already; building sites; and narrow strips along streets.

Table 2. Kinds of candidate land, number of sites

The five kinds inside built parcels exist only in Tallinn, where the city's own land-cover map allows them. The full rules are in Part 2.

Owner types come from the cadastre. About three quarters of the candidates are on private land; by area the private share is smaller, as private sites are mostly small (Part 2). Cards name the body that manages state land and, in Tallinn, the city's land; for all other land they show only the owner type, never a name.

How far to trust the candidates

Every candidate is desk-made from open data and needs a visit. AI models judged samples on the May 2024 aerial photo of Tallinn; the judges knew what the map calls each site. The checks show where the rules went wrong; they cannot show what is on the ground today (numbers in Part 2).

• Tallinn parcels and car parks. Three review rounds judged about 1,100 sites of versions 1 to 3. In round 3, of the 25 top sites that each index listed at the start, 60–64% looked vacant and 84–100% vacant or partly used. The rules were then changed (version 4) and the new lists were not judged again; until version 8.3 the finder said "about three quarters were empty on the photo", a figure from re-applying the version 4 rules to the round-3 verdicts, not from a new check; it now gives the round-3 figure.
• Pieces inside built parcels. Of 40 pieces of open ground, two thirds looked vacant; of the 25 the final rules keep, about three quarters, an optimistic figure as some rules were tuned on them. The check of strips, mown lawns and paved forecourts is unfinished.
• Not checked. The current Tallinn start view as a whole; the county candidates on the photo; how much vacant land the rules miss; anything on site.

Known weaknesses

• A garden split off as its own parcel looks vacant. The likely-gardens rule hides such parcels when it can tell; some get through.
• "Paved or gravel ground" is the weakest kind on at the start: in round 3, 21% of the paved sites looked vacant, and 43% of the 21 that the version 4 rules would keep (re-applying the rules to the round-3 verdicts), against about 80% for all kinds on at the start taken together.
• Many candidates may already have building rights: 34 of the 69 sites in Tallinn's start view and 4,070 of the 9,754 in the county start views lie under a detail plan in force. A site can look empty on the photo and still have building rights, so "vacant" in the checks means empty on the photo, not free to green.

Tallinn by building and street block

The building map ("Tallinn, building by building", damianocerrone.com/dashboard/editua/buildings/) measures walking distances along OpenStreetMap's streets and paths from a point inside each of Tallinn's 48,754 buildings, gives each building its own scores, and averages those per cadastral parcel, street block and Report 4 square. Home buildings are those with at least one dwelling, registered or filled in as Part 2 describes (the map's "Only buildings with homes"); homes and dwellings are the same unit in this report.

1. Pick a mode and a level. Index or Places in reach; under "Show", Buildings, Parcels, Blocks or Grid (figure 9). At grid level, a switch chooses between the square's own values ("Report 4 scores", or "Raul's values" in Places in reach) and the mean of its buildings ("Building mean").
2. Set the index or the measure. In Index mode, the Homes row in "Compose it" ("In its grid square" or "Around the building") and the score filter; in Places in reach, a measure and a walk class (figure 10). Click a building, parcel or block for its card: a building's card names its Report 4 square and that square's three scores; "Copy a link to this view" saves the view and its settings as a link.
3. Read the result. In Index mode, "Highest scores" lists the 15 highest-scoring buildings, parcels, blocks or squares in the range and "Hide what is outside the range" hides the rest; in Places in reach, a building's card lists the service targets it misses.

a · Grid squares

b · Street blocks

c · Buildings

What the finer scale changes

With homes counted "In its grid square", as in Report 4, scoring each building changes little. On Fair-Share, home buildings differ from their square's published score by a median of 1.65 points, and 89% are in their square's 20-point band (table 7). For 59% of home buildings, more of the difference comes from the building map's walking network than from where the building stands.

In Index mode, with that setting (the default), every building takes its square's homes count and places are still counted per grid square, so only the walk's starting point is finer. Places in reach counts each place at its own position. The finer scale matters in four cases:

• Short walks. For a 5-minute walk (W5), 48% of home buildings reach about as many places (within 10%) as their square's own point on the same network; for a 15-minute walk (W15), 87% do (table 8).
• Pass-or-fail targets. For 482 home buildings (4,075 dwellings), the answer to "are all 21 service targets met?" differs from the answer at their square's own point.
• Large squares. In 12 of the 1,568 squares with five or more home buildings, the buildings' Fair-Share scores span more than 20 points; 8 of them are 500 m squares, where homes lie far apart.
• Homes counted around each building. The circle used has the area of a 125 m square. In 125 m squares scores barely change; in 250 m and 500 m squares, where Report 4 counts the homes of the whole square, Fair-Share falls by a median of 6 and 15 points. Overall, 70% of home buildings stay in their square's band (Fair-Share).

Parcels and blocks are other units, not finer ones. A parcel usually holds one home building, so its mean is close to that building's score. Street blocks, the land between streets, cover a median of 1.9 ha when they hold homes, a little more than one 125 m square (1.6 ha), but 112 very large blocks hold 22% of the dwellings.

A block's value is the plain mean of its buildings, not weighted by dwellings; the figures here use home buildings only, while the map opens with all buildings. In Index mode a block's mean also blends the homes counts of the squares the block spans (a median of three), so a block can fall in a band that none of its squares has (figure 9b). For 76% of home buildings the block mean is in the band of the building's square, but for only 60% of their dwellings (Fair-Share). Of the 57,000 dwellings in top-band squares (above 80), 28,900 are in blocks whose mean is 80 or less.

What the maps show so far

• Where Fair-Share is highest, greening more likely means taking up paving (depaving) on a forecourt, car park or yard, or changing an existing green, than planting an empty plot. Tallinn's start view lists 69 sites (12.3 ha), mostly small lawns and yards.
• In the counties the start views list 9,754 sites, almost half of them meadows or fields and 77% private; 59–72% of the sites in each county's top 20% of squares pass the start filters (two thirds overall), against a tenth in Tallinn.
• The targets missed most often are "9 kinds of shop" (of 18), "4 kinds of culture venue" (of 9) and "Civil-defence shelter" (table 6). By district, all 21 are met for 77% of Kesklinn's dwellings, 26–41% in Mustamäe, Lasnamäe and Põhja-Tallinn, 11% in Kristiine, 4% in Nõmme and none in Haabersti or Pirita (figure 10).

Using the maps together, and what they cannot tell you

In Tallinn the two maps share the presets, the arithmetic and the grid's percentiles, but not the walking network: at a square's own point, building-map scores differ from the published ones by 2.6 points on average (Fair-Share). To see whether the homes near a site account for its score, open the building map at the same place with Homes "Around the building". Near a junction, the building map's Places in reach mode shows which of the 21 service targets each building misses. The maps do not link yet. To open the same place in the building map, rewrite the finder's address: v=15.0/59.44080/24.78220 becomes buildings/#v-1500-5944080-2478220 (zoom × 100, coordinates × 100,000).

Neither map can tell whether a site is in use today, whether its owner would agree, whether greening is wanted there, or what the soil, pipes and cables allow. A visit answers the first; the others need the owner, the city and the utility networks' records. The index counts places within a walk, not trees or the quality of green, unless tree cover is added. Scores are relative to their area, the Tallinn area or one county.

Data and licences

Table 3. Data and licences

The site finder credits the grid as "the Estonian 15-minute walk accessibility grid AGRD_EE v6B" without an author; the building map and the outputs page credit Raul Kalvo.

Where Report 4's text differs from its script and data

Report 4's text describes each term as divided by its maximum (x / max x); its published scores come from its QGIS script, which divides each term by its 95th percentile, caps it at 1 and stretches the total from P1 to P99 (Index arithmetic, below). This report and both maps follow the published scores, as Report 4's map story does.

Site finder method

Grids

Tallinn uses the file behind Report 4: 5,966 squares over the city and a ring of suburbs (3,989 of 125 m, 688 of 250 m, 1,082 of 500 m, 196 of 1 km, 11 of 2 km) with the published scores and the 80 non-empty amenity values. It adds three of Raul Kalvo's density layers (homes per hectare within 500 m, floor area per hectare within 250 m, jobs within 30 minutes by public transport) and Copernicus 2018 sealed surface and tree cover. The counties use the national AGRD_EE v6B grid (46,213 squares); a square belongs to the county holding its representative point, giving 676 (Hiiu) to 5,766 (Harju) squares of 125 m to 8 km. In Tallinn eight amenity fields are empty in every square (basic schools, three indoor-sport and four outdoor-sport kinds); in each county grid 10–37 fields are empty (51–78 kinds remain).

Index arithmetic

The arithmetic is that of the QGIS script behind Report 4's published scores, re-implemented in Python for the map story and in the browser for both maps. For each square and each ingredient b (a "basket" in the code):

Fair-Share is homes 25%, children's places 20%, nature places 35% (as a need) and everyday services 20%. Team 1 is homes 40% (its "home streets" and "dense housing" both use the homes count), education 20%, food 20% and services 20%; its "parking lots" category has no data in the grid and is left out. Team 2 is homes, food, places to move, interests and culture, and education, 20% each. Over the 5,966 squares Fair-Share rises with its homes term (correlation r = 0.78) and falls slightly where nature places are missing (r = −0.20). In Tallinn the three presets show the stored, published scores; any change recomputes in the browser, which reproduces the stored scores within 0.6 points. Fair-Share's Top 20% starts at 54 in Tallinn (P80 = 54.6): 1,224 squares, against 1,194 in Report 4's own shortlists.

In the counties, P95, P1, P99 and P80 come from the county's own squares. There 56–82% of squares (20–57% of their homes) have no nature place in reach (Tallinn: 16% of squares, 5% of homes), and 45–74% (14–42% of homes) have no children's place, nature place or everyday service (Tallinn: 7% of squares). In those squares the 35% need term is at its maximum and Fair-Share depends on homes alone: the "deficit saturates in empty land" of Report 4's limitations. The term is also coarse: its county P95 runs from 1.0 square (Hiiu) to 7.8 (Tartu; Tallinn 13.3), so one more square holding a nature place at full weight moves a county score by 9 points (Tartu) to 47 (Hiiu; Tallinn 5).

A site's score is the mean of the scores of the squares it overlaps, weighted by the share of its area in each, over the part of the site inside the grid (the area-weighted sum divided by that share). A site with less than half of its area inside grid squares gets no score (9 in Tallinn, 2,558 in the counties) and never appears while "Only show what scores inside the filter range" is on. The unscored county sites lie at the edge of the grid: 2,522 of the 2,558 within 1 km of the county outline; 2,323 have their centre in no square of any finder, most of them in coastal counties, and 198 in a square that belongs to a neighbouring region. Up to a tenth of a coastal county's sites are affected (Saare 378, Hiiu 177).

Candidate rules

Unbuilt parcels (Tallinn and counties)

• Size and shape. Parcels of 100 m² or more that hold a circle 10 m across.
• Left out by land use. Defence, water, waste, mining, peat-extraction (since version 8.3) and protected land; transport land (30% or more) and street-like addresses, except city reserve land at least 10 m wide.
• Also left out when water and wetland ≥ 15%; sand or gravel ≥ 50% of the land (beaches, dunes); ETAK special areas (sports complex, port, cemetery, quarry, airfield) ≥ 30%; sports grounds and OpenStreetMap playgrounds ≥ 30%; OpenStreetMap cemetery ≥ 30%, school ≥ 50%, military ≥ 10%, park ≥ 50% (except City Property Department plots), mapped uses together ≥ 50%; multi-storey or underground parking ≥ 30%; pedestrian area ≥ 30%, or roads and pedestrian area together ≥ 30% on public open space; ETAK roadway ≥ 30% (kept as a car park where the city base map maps it as a drivable plaza); agricultural or forest land that is ≥ 50% green, bare or field.
• Building rule. Buildings (ETAK, and in Tallinn also the city base map, whichever maps more) cover under 5% of the parcel. A parcel with 150 m² or more of buildings is kept only when it is 5 ha or less, is not a yard and holds no sports building; its card says so (14 sites in Tallinn, 1,363 in the counties). No residential or public building of 40 m² or more; no home in the building register; no underground garage, canopy, greenhouse or tank above small thresholds. The finder's About tab compares this with Raum+ (ref. 1: building cover under 6% and under 30 m² of buildings); this map allows more sheds, and shows the Urban Atlas class "land without current use" as a second opinion (ref. 2).
• Kind. The largest land-cover group if it reaches 40% of the land, otherwise "mixed cover"; ruins, buildings being built, car parks and woodland have their own rules. In Tallinn, Copernicus tree cover of 50% or more turns parcels under 1 ha into "trees and shrubs", and parcels of 1 ha or more that ETAK maps at least a fifth as trees into woodland, except where ETAK maps half or more as lawn, meadow or field, or open ground as mostly bare.

Hidden at the start (kept, one switch away): likely gardens of the house next door (part of the same registered property as a house parcel next door and under 3,000 m²; a yard parcel under 1,000 m²; or under 700 m² next to a built parcel with the same street and house-number base), and sites where OpenStreetMap maps a building of 100 m² or more that ETAK lacks, covering 5% or more of the site. Tallinn hides 421 sites this way, the counties 6,069.

Pieces inside built parcels (Tallinn only)

From the city's wall-to-wall land cover (TAR): meadow, wasteland, bare ground, shrubs, trees and lawn, minus every mapped use (roads, car parks, pedestrian areas, sports grounds, playgrounds, schools, cemeteries, allotments, utility yards, structures and more) and minus 5 m around buildings; cut by parcel. An earlier EKA mapping of Tallinn's tühermaad (vacant or derelict land) also took everything with a set use off the city map (ref. 3). A piece is kept when it is 300 m² or more, at least 15 m across and keeps 200 m² after a morphological opening of 7.5 m (shrunk by 7.5 m and grown back), on a built parcel (buildings cover 5% or more, or one building of 40 m² or more) that is not transport, agricultural or forest, defence, water, landfill, mining or protected land. Dropped after the photo checks: house gardens, residential parcels under 3,000 m², grounds of public buildings, apartment courtyard greens, the shore, pieces wrapping a building, woodland of 1 ha or more, public open space, pieces of 5 ha or more, pieces in green, cemetery and small-house zones, and building sites. Lawn on business or industrial land is kept as "lawn in a built parcel (check)". Three kinds added in version 6, all off at the start: narrow strips (open ground along a street or pavement, 2 m clear of buildings, 100 m² or more, at least 5 m across, under 3,000 m²); mown lawns (the lawns the rules above drop as courtyard, public-building or non-business lawn); paved forecourts or yards (asphalt, concrete or paving that is not a road, square or mapped car park; 300 m² or more, at least 10 m across, 2 m clear of buildings).

Surface car parks (Tallinn and counties)

ETAK car parks and OpenStreetMap surface parking, minus decks (60% or more under roofs), buildings, canopies, parcels already in the list as sites, and excluded land; 1,000 m² or more holding a circle 10 m across, at least half inside the cadastre; dropped when green covers 50% or more, military land 10% or more, or street-side parking 70% or more. The owner is the one holding 70% or more of the area, otherwise "mixed". "Courtyard car park" when yard or residential land covers half or more.

Hand corrections and stable ids

Eleven hand corrections from the photo, all in Tallinn: one site changed to "Being built", one to "Public green, not a park", nine dropped (among them the zoo and the water treatment works). Parcel sites keep the cadastral code as their id; car parks and pieces inside built parcels get an id from their centroid, carried over to a new version's shape that overlaps it at an intersection over union (IoU) of 0.5 or more (an overlap of 0.2 to 0.5 keeps the old id as a second, alias id; 781 in Tallinn), so shared links and saved lists keep working.

Owners, plans, zones, contaminated land and junctions

• Owners. Six types from the cadastre's ownership form: City of Tallinn (counties: Municipality), State, Private, Public body, Mixed, Unreformed or unsettled (ownership not yet settled; mostly state land not yet through the land reform, reformimata maa). Tallinn adds the city's land manager and whether the city lets, offers or sells the land (gis.tallinn.ee); both add the state land manager (KATRI). No owner names. Private land holds 75% of Tallinn's 6,324 sites and 76% of the counties' 72,425, but 45.5% of Tallinn's candidate area (city land 39.5%) and 60.0% of the counties' (municipalities 23.3%, the state 13.5%).
• Detail plans. Tallinn: the planning register. "In preparation" = applied for, initiated or adopted; "in force" = in force or partly in force. A plan counts when it covers at least a quarter of the site; "in preparation" wins over "in force" and the card names both. Sites: 2,935 in force, 693 in preparation, 2,696 none. Counties: the national detail-plan index; "in preparation" only when initiated in the last five years, because the index keeps many ended plans in that status (matched with Tallinn's register, only 105 of 240 plans in that status were still active). County sites: 29,932 in force, 544 in preparation.
• General-plan zones (Tallinn). The base zone covering most of the site (an overlay zone only where no base zone covers a quarter), grouped as Green, Public, Building, Other or none, from the district general plans in force and, in Kesklinn and Põhja-Tallinn, the 2001 general plan; the card shows the drafts separately. Sites: Building 4,901, Green 547, Public 359, none 313, Other 204; 314 lie in a green-network development area of a district plan in force.
• Contaminated land. Tallinn: the city's layer, risk high, medium, low or not assessed; a site is flagged when 5% or 50 m² of it is inside (338 sites: 164 high, 85 medium, 19 low, 70 not assessed). Counties: the national register of residual pollution sites (EELIS) without the records assessed as not hazardous (these only add a note to the card): polygons as drawn, points as circles of 25 m radius, and the same 5% or 50 m² rule (746 sites). Of these, 727 overlap an EELIS polygon (598 lie at least half inside it), and 196 overlap one of the three polygons of 1,000 ha or more (Tapa airfield, the Dejevo rocket base and the phosphorite mining areas, all marked as cleaned up or largely cleaned up), so in the counties a flag often means "inside a large registered former military or industrial area". The card gives the record's category and clean-up status, and says not to dig or touch soil or water.
• Street junctions. OpenStreetMap streets from trunk roads to living streets, plus pedestrian streets; a node shared by three or more street segments, clustered within about 15 m (Tallinn) or 24 m (counties), kept where at least two differently named streets meet, scored by the square it falls in: 2,751 in Tallinn, 25,524 in the counties.

Table 4. Versions of the site finders, October 2026

Building map method

Buildings and dwellings

Of the 50,651 ETAK building footprints with building-register fields inside Tallinn, 48,754 are used (planned buildings, ruins, foundations and underground buildings left out). Walks start from a point inside each building's footprint; the building belongs to the parcel, the block and the grid square (for the grid-level means) that hold that point; for Report 4's terms (homes, map layers) it takes the grid square that holds most of its footprint. Of the buildings used, 68 lie outside Report 4's grid (61 on Aegna, 7 at Paljassaare and Järvevana tee) and have no index score.

Dwellings come from the building register's open-data extract of 27 January 2025: residential parts per building, existing or being built, counted as parts minus one when there are two or more, an adjustment that brings the register close to Raul Kalvo's address-based counts (the same count for 66% of single apartment buildings and 88% of houses). They are placed on the footprints by register code, then by shape overlap, nearby shape or address. Apartment buildings registered with two dwellings or fewer but 300 m² or more of footprint times floors get their square's shortfall (its DUs_Sum minus the register's dwellings in it), at most one dwelling per 60 m²; houses missing from the 2025 register get one or two. Result: 233,600 dwellings in 23,976 home buildings; 1.5% of the dwellings come from this fill.

Walking network and engine

OpenStreetMap Estonia (Geofabrik extract, replication of 14 October 2026) with a foot profile: roads marked foot=no are kept where the road carries a sidewalk tag (a pavement); roads whose pavements are drawn as separate ways are kept only where no walkable way runs alongside for 100 m or more; private ways are used only to walk out to the public network, never through. The graph has 345,212 nodes and 7,571 km (7,369 km public); the engine uses its three connected parts of 500 nodes or more (the main network, Aegna, Naissaar; 332,657 public nodes). A point attaches to the nearest edge at the foot of the perpendicular, and the length of that perpendicular is added to the walk. One shortest-path search runs from each place of interest on the undirected walking network; the attachment legs are one-way (places only leave, buildings only arrive), so no route passes through another point's leg. The engine equals a brute-force computation to within 3.2 × 10⁻¹⁴. Six places more than 500 m from any way are left out.

Raul Kalvo's method (Places in reach)

P and H are from his service list (sheet "Liikumisklassid"). The logistic form was inferred from his EE(v6B) decay tables (Checks); he has not confirmed it. In his tables each square's origin is its Location point (the dwelling-weighted centre of its homes, its centre when it has none), each place counts at the Location point of its square, and d comes from his square-to-square table ODM(v6B)_ActiveMobility(16km). The building map keeps f, P and H but walks from each building to each place's own position on the 2026 OpenStreetMap network.

Table 5. Raul Kalvo's walk classes, walking distances in metres

Places: POI list v12 in Tallinn and 3 km around it, 11,962 points in 88 kinds (11,956 used). Compared with his own tables, v12 has about a third of the indoor play rooms they used, and lacks deposit-return points and regional centres.

Report 4's variant (Index)

Report 4's walk values are not Raul Kalvo's W15. For each kind of place they count the grid squares holding at least one such place, from Location point to Location point, weighted by walking time. The decay is the same logistic one, with P = 830.6 m (10 minutes) and H = 415.3 m (5 minutes) at 83.05648 m per minute: full to about 830 m, half at 1,246 m, none from 1,661 m, close to his W10. This rule was inferred from Report 4's grid file (Checks); Raul Kalvo has not confirmed it. The building map uses the same: 6,914 square-and-kind destinations in 80 kinds, each at its square's Location point.

Building scores against the grid's percentiles

The building map uses the site finder's arithmetic, with P95, P1 and P99 taken from Report 4's 5,966 squares, so a building's 72 means the same as a square's 72. The stretch (P1 to P99 of the raw score) is 22.74–73.36 for Fair-Share, 0.00–93.79 for Team 1 and 0.00–87.95 for Team 2. The homes ingredient is either the dwellings of the building's grid square (Report 4's own term, the default) or the dwellings within 70.5 m of the building (a circle with the area of a 125 m square): those of the buildings whose inside point lies within 70.5 m, in a straight line. Both homes terms are divided by the grid's P95 of DUs_Sum (210). In 125 m squares the circle counts a median of 0.97 times the square's homes and Fair-Share does not move (median change 0.0 points). In 250 m and 500 m squares, whose count covers 4 and 16 times the circle's area, it counts a median of 0.30 and 0.14 times as many, and Fair-Share falls by a median of 6.4 and 14.8 points (over all home buildings, a median change of −2.3 points against the published square). Scaling the circle counts so that their P95 over home buildings (180) matches 210 changes the median by about a point. Map layers come from the building's square. Top 20% starts at 54 (Fair-Share), 46 (Team 1) and 49 (Team 2), cut at the grid's published scores.

Service targets

Raul Kalvo's service list, column "walking, urban", read as his June 2025 slides show it: 21 targets (table 6), pooled where one cell spans several rows. A kind of place counts when its value, rounded to 2 decimal places, reaches 1: one place within about 920 m at W10 (460 m at W5, 1,440 m at W15), or several farther away whose weights add up to 1, such as two at 1,200 m at W10. A target is met when at least the set number of its kinds count; care homes are left out. Four further targets have no places in POI v12 and are left out (private and special-needs schools, non-stationary learning, park paths). Of 12 readings tried (5.2–18.9% of Estonian dwellings at the urban walking level with his tables), this one was chosen because it gives the 15% of his slide 41; leaving out care homes (marked "–" in the slides) matters most (6.0% with them), and 4 decimal places instead of 2 give 13.6%. The Tallinn shares depend on this reading, which he has not yet confirmed.

Table 6. The 21 service targets (walking, urban) and the share of Tallinn's home buildings meeting each

Home buildings: the 23,976 with dwellings; all 21 are met for 30.0% of their dwellings.

Parcels, blocks and squares

A unit's value is the plain mean of the buildings whose inside point lies in it, all buildings or home buildings only; a unit without buildings has no value. Parcels: Tallinn cadastre of 15 October 2026, 38,642 parcels, 28,449 with a building. Street blocks: Raul Kalvo's blocks with their inside point in Tallinn, 2,310 (his types S 2,198, XL-NotOptimized 86, XL-Optimized 26; the 112 XL blocks a median of 28 ha), of which 2,135 hold a building (a median of 17); 412 buildings lie in no block. Of the blocks with homes, 81% span two or more squares. Inside blocks with five or more home buildings, the buildings' Fair-Share scores span a median of 12 points. Grid: the map draws 4,038 of Report 4's 5,966 squares, those in Tallinn city or holding a Tallinn building.

Checks and the scale analysis

Checks of the building map

• Raul Kalvo's tables. With his own square-to-square network distances, the decay above reproduces all 561,150 values per walk class of his Tallinn-area tables (6,450 squares × 87 kinds) to 4 decimal places when indoor play rooms are left out; with them, 99.3–99.8%. As the decay was inferred from these tables, this confirms the inference, not the method independently.
• Report 4's walk values. With his network durations (distance at 83.05648 m per minute), the rule under Report 4's variant reproduces all 520,960 values of Report 4's grid file (largest error 8.5 × 10⁻¹⁴). The rule was inferred from that file.
• His grid recomputed on the new network (his squares' Location points and destinations): r = 0.997 (W15), 0.993 (W10), 0.975 (W5); median ratio new/his 1.004, 0.993, 0.963. Building means per square against his squares: r = 0.996, 0.991, 0.969 (3,558 squares).
• Report 4 recomputed on the new network: r = 0.994 for the places (80 kinds, 3,967 Tallinn squares); Fair-Share, Team 1 and Team 2 against the published scores r = 0.987, 0.996, 0.991, mean absolute difference 2.60, 1.32 and 2.05 points. Building means per square against the published scores: r = 0.984, 0.993, 0.991.
• Page against Python. The page's index scores equal the Python computation within 0.0031 points for the three presets with both homes ingredients, and for custom mixes at every level.

Checks of the site finders

All photo judges were AI models (Claude agents) looking at the May 2024 orthophoto: one per site in rounds 1–3 (round 1: four judges on separate samples), two per piece in the inside-parcel checks. They knew each site's kind, and in rounds 1–3 also judged whether it was right; only the second judge of the prototype check was blind to it. With 25 sites a share has a 95% confidence interval of about ±19 percentage points; with 10 per kind, about ±30. Verdicts: vacant, partly used, kept green, in use, unclear. Shares are vacant / vacant or partly used.

• Round 1 (version 1, 4,266 Tallinn sites): four AI judges, 583 verdicts. Fair-Share's top 25 was mostly car parks (11) and paved or gravel ground (9); version 2 switched car parks off at the start.
• Round 2 (version 2, 4,195 sites): the top 25 each index showed at the start: Fair-Share 64% / 84%, Team 1 56% / 80%, Team 2 60% / 80%. A random sample of the kinds on at the start (199 verdicts), weighted by kind: about 66% / 81%.
• Round 3 (version 3, 4,025 sites), measured: top 25 Fair-Share 60% / 84%, Team 1 64% / 100%, Team 2 64% / 100%. Ten random sites per kind on at the start, weighted by the 1,623 sites then shown: about 80% / 90%. All 68 "paved" sites then shown: 21% / 53%. Simulated: applying the version 4 rule change to the round-3 verdicts gives 76% / 92%, 72% / 96% and 76% / 100% for the three top 25s, and 43% / 86% for the 21 sites it would leave paved; nobody judged the published version 4 lists.
• Pieces inside built parcels, prototype: 98 pieces, two AI judges; the same verdict for 80 (Cohen's kappa 0.74), the same vacant-or-not answer for 90. As both were Claude agents, this agreement shows consistency, not accuracy. Version 5: 40 pieces of open ground 66% / 81% (mean of the two judges), the 25 the final rules keep 74% / 86% (in-sample: some final rules came from these pieces); 20 lawns on business land 28% vacant.
• Version 6, unfinished: 30 pieces sampled per new kind, 90 in all; Maa- ja Ruumiamet's tiles went blank during the check and only 50 had a usable photo judged by both. Means of the two judges: narrow strips (16) 59% / 78%; mown lawns (18) 58% kept greens, 31% vacant; paved forecourts (16) 72% in use, 6% vacant.
• Data audits: owner codes matched the cadastre for all 4,195 sites of version 2 and all 3,120 parcels of version 3; no duplicate car-park ids; no personal data in any file; the 600 pieces inside built parcels of a pre-release build of version 5 matched their parcel in the cadastre on every field checked.
• Regression tests: version 7 gave the same scores, counts and lists as a copy of version 6 on six test links. In version 8 every county page loaded, drew its map inside the county, and showed its detail plans (as a filter and as a map layer) and a site card; all 15 passed on the local preview before publishing (Rapla on a second run). The picker's counts equal the data for all 16 finders.

Scale analysis: definitions

The 23,969 home buildings inside Report 4's grid (233,593 dwellings) against their squares, with the building map's own code and data.

Report 4 square

The Report 4 square that holds most of the building's footprint: the square whose homes count the building takes, and the one its card names.

Published score

Report 4's stored score of that square.

Square point

The square's own point, its Location point (see Raul Kalvo's method), attached to the building map's network like a building and computed with the same settings: for the index the same destinations, homes ingredient and P95, P1 and P99; for Places in reach exact place positions (table 8). As that point already lies among the square's homes, the position part below stays small. The square point splits each building's difference in two: the network part (square point minus published score: a different network and engine from Report 4's) and the position part (building minus square point: where the building stands in the square).

Bands

The maps' 20-point bands; the top band is scores above 80. Medians and 90th percentiles (P90) are of the absolute difference.

Unit means

Parcel and block means of home buildings only (on the map, tick "Only buildings with homes").

Table 7. Home buildings against their Report 4 square

The medians of the two parts do not add up to the total. The network part is the larger of the two for 58–60% of home buildings, depending on the index. By square size, the Fair-Share median difference is 1.54 points in 125 m squares, 1.66 in 250 m and 2.30 in 500 m squares; squares of 250 m and 500 m hold 35% of the home buildings but 6.4% of the dwellings.

Table 8. Places in reach: total of the 88 kinds per home building

The square point is his method computed at the square's own point on the new network with exact place positions, so the second pair of rows leaves only where the building stands.

Service targets (all 21 met): 295 home buildings (4,184 dwellings) meet them while his square does not and 520 (5,359) the other way round, agreement 96.6%; against the square point 222 (1,998) and 260 (2,077), 98.0%. By dwellings, 30.0% meet all 21 by building and 30.5% by his squares. Inside the 1,568 squares with five or more home buildings, the buildings differ in how many targets they meet in 64% of squares, by three or more in 16%. In Haabersti and Pirita the culture-venue target is met for 2.5% and 0% of dwellings.

Examples, chosen by fixed rules in the analysis script:

• Network. Mustamäe, E. Vilde tee: square published at 79.5, square point 83.5, its three home buildings (439 dwellings) 83.3–83.5.
• Threshold. Kristiine, Kännu, Lehe and Rästa streets: the square meets all 21 targets by his values and at its square point; 7 of its 15 home buildings do, 8 miss "9 kinds of shop".
• Barrier, not confirmed. Kristiine, Algi and Kaarna streets: the widest Fair-Share range among 125 m squares with eight or more home buildings, 61.9 to 86.4. The lowest and highest are 97 m apart in a straight line and 499 m on the network; whether something on the ground separates them, or OpenStreetMap misses a path, is unchecked.
• Blend. Haabersti, a block between Rannamõisa tee, Taludevahe and Vahepere streets: 18 home buildings in five squares published at 28.5 to 68.2; the mean of its home buildings, 51.1 (53.0 for all buildings, as the map opens), falls in a band that none of its squares has.

Reproducing the maps

Each finder's Index tab downloads the grid as GeoJSON (score, in range, size, homes, Fair-Share, Team 1, Team 2), and "3 · My list" downloads the list as CSV or GeoJSON (cadastral code, kind, owner, area, plan, field notes, coordinates). The finders are built by build_sites.py, build_inside.py, build_junctions.py and score_lib.py, the building map by run_all.sh; the scripts are kept by the EKA team. The building map has no download; its values per building, parcel, block and square are in the page's binary data files. The photo samples were drawn with fixed seeds.

Limitations and open items

Table 9. Limitations and open items

Credits, sources and references

• Site finders (EDITUA × Street Studio): "This map: Damiano Cerrone", as the map's own credits say.
• Tallinn, building by building: computed 16–17 October 2026 for EDITUA. As the outputs page credits it: "Places, grid and method: Raul Kalvo". The map's own line: "POI list, grid, decay and service list: Raul Kalvo, EKA."
• Accessibility method, places of interest, walking values, service list and street blocks: Raul Kalvo, including his slides "Teenuste ligipääsetavus aktiivsete liikumisviisidega" (June 2025).
• Report 4 indices (Fair-Share, Team 1, Team 2): Damiano Cerrone, Siim Tuksam, Heli Telgmaa.
• Data: as listed under Data and licences.
• Figures: screenshots of the live maps taken on 17 October 2026 (site finders version 8.2). Aerial photo in figures 2–8: Maa- ja Ruumiamet orthophoto, flights of 2–8 May 2024 over Tallinn (figures 2–6; the outer suburbs in figure 2 from 27 March – 30 April 2024) and those given in the captions of figures 7 and 8; Maa- ja Ruumiamet open data licence, geoportaal.maaruum.ee/opendata-licence. Boundaries in figure 1: Maa- ja Ruumiamet (EHAK). Streets in figures 9 and 10: © OpenStreetMap contributors (ODbL); buildings and districts: Maa- ja Ruumiamet.

The maps and the earlier reports

• Site finders: damianocerrone.com/dashboard/editua/map/ (picker), …/map/tallinn/, and …/map/<county>/, e.g. …/map/tartu/
• Tallinn, building by building: damianocerrone.com/dashboard/editua/buildings/
• Report 1, Urban nature in Estonia: damianocerrone.com/dashboard/editua/report-1/
• Report 2, Workshop in Poznań: damianocerrone.com/dashboard/editua/report-2/
• Report 3, EcoUrban Composer indicators: damianocerrone.com/dashboard/editua/report-3/. Its indicators score what an intervention does on one site or block; these maps come a step earlier: where it would reach people, and which land might take it.
• Report 4, Regreening priority index: damianocerrone.com/dashboard/editua/report-4/; map story: damianocerrone.com/dashboard/editua-report4/
• All EDITUA outputs of the EKA team: damianocerrone.com/dashboard/editua/

References

1. Raum+ (ETH Zürich), method for surveying settlement reserves; Kanton Schwyz, Merkblatt Raum+ (2025).
2. Copernicus Land Monitoring Service, Urban Atlas mapping guide, version 6.3.
3. Karro-Kalberg, M. (2011), "Plaan B, tühermaade regenereerimine", master's thesis, Estonian Academy of Arts; not online, method as described by Leppik, T. (2016), "Iseorganiseeruv maastik ja selle kujundamine, Tallinna kesklinna linnaosas asuvate tühermaade näitel", master's thesis, Tallinn University of Technology, Tartu College, p. 38.
4. Keskkonnaseadustiku üldosa seadus (General Part of the Environmental Code Act), § 32, cited in the site finders' visit rules.