Xamfrà: open methods to detect and appraise urban opportunities in the Barcelona Metropolitan Area

Blanco Quintanilla, Humbert

1. Introduction

1.1 The problem: a polycentric metropolis that has to choose

The AMB covers 63,829 ha and groups 36 municipalities of very different sizes: Barcelona holds 1,068 of the 2,151 census sections, while el Papiol, Sant Climent de Llobregat and la Palma de Cervelló have two each. Economic activity is not a single centre: the continuous tertiary fabric of the Eixample coexists with industrial and logistics corridors along the Llobregat and the Besòs, university campuses in Bellaterra and on the Diagonal, and ageing industrial estates surrounded by housing. This polycentric structure is both an opportunity (centralities that can be reinforced) and a risk (transforming productive land that is in use, or neighbourhoods with vulnerable residents).

The PDU metropolità, at its second initial approval (Metropolitan Council, 10-02-2026) [1], frames this choice with 75 metropolitan opportunity areas (AOM) scheduled across four six-year periods (sexennia), 28 economic-activity regeneration areas (ARUAE), 6 residential regeneration areas (ARURE) and 16 metropolitan project spaces (EPM). Its objectives speak of a «polycentric city», a network of differentiated centralities and 50 % protected housing on new residential floor area in the AOM. Tools that support such a decision must say where, why and with what risks, and do it so that a municipal planner can verify each step.

1.2 The users: public administration

Xamfrà's first users are public administrations and service companies: planning, housing, mobility and economic-development staff of the AMB and the municipalities. This audience has three needs that shape the design:

Traceability. A figure that cannot be explained in a council committee is useless; the source, year, threshold and rule are needed.
Separation of facts and judgements. The tool must show where conditions coincide; trading objectives off (housing against activity, density against green space) is a political decision that must not be hidden inside a coefficient.
Care with people. Many questions touch social vulnerability; the presentation must not turn section-level statistical context into a judgement about a building or a household.

1.3 Contribution

This paper documents, in one place and in academic form, the methods Xamfrà developed and published between 30-09 and 06-10-2026. The contributions are:

a framework of counted criteria with three-valued logic (yes / no / unknown), applied uniformly to some twenty modules, as an explicit alternative to composite indices (section 2);
the documented adaptation to the AMB of international methods: Getis-Ord Gi*, London's SHLAA, New Zealand's HBA, RAPTOR with Hansen accessibility, GLA industrial-land demand, Bates' displacement typology, CEQR ratios, Munich's SoBoN, Copenhagen's proximity principle and economic complexity (section 4);
reproducible results for the AMB with manifests that fix inputs (SHA-256), parameters, sensitivity and limits (sections 5 and 6);
a discussion of validation, uncertainty and ethics, including back-tests the method does not pass well (sections 6 and 8).

2. Design principles

2.1 Counted criteria, not weighted indices

The OECD–JRC Handbook on composite indicators [2] acknowledges that an index summarises well but that every step (selection, normalisation, weighting, aggregation) introduces judgements the end user does not see, and that results can be very sensitive to the choice of weights and aggregation. The review by Greco et al. [3] reaches the same conclusion: weighting is the most contested and least justifiable part of composite indices. For administrative use there are three concrete problems:

Compensation. In a weighted sum, much opportunity offsets much risk. An area with excellent transit and a very vulnerable population can score the same as a mediocre area with no risk. Planning needs to see both at once.
Weights without legitimacy. Who decides that transit is worth twice as much as vacant land? In an index, the analyst who builds it; in a public decision, the political body.
Opacity about the unknown. An index needs a value for every component; missing data are imputed or treated as zero.

Xamfrà therefore uses explicit counted criteria. Each criterion is a question with a written rule and a visible numeric threshold (e.g. «metro, rail or tram station ≤ 800 m from the edge»). Criteria within a family are counted; they are never multiplied by a weight. Different families (opportunity, knowledge, risk) are counted separately and shown in a matrix, not as a subtraction. The number of criteria met is a size, not a quality: the viewer sorts by «most criteria met, fewest risks, most known answers» and says so.

2.2 Unknown ≠ no ≠ zero

Sources have systematic gaps: the ground-floor premises census exists only for Barcelona; tourist dwellings are geolocated only in Barcelona; the SIPAE business-park register does not declare occupancy for 42 of 154 parks. Treating such gaps as «no» would bias every result against the municipalities with less data. All rules therefore use Kleene's three-valued logic [4]:

Box 1. Three-valued counting rule. For a unit with criteria $c_k \in \{1, 0, \varnothing\}$ (yes, no, unknown), an optional gate $g$ and a minimum $N$ :

m = \sum_k \mathbf{1}[c_k = 1], \qquad u = \sum_k \mathbf{1}[c_k = \varnothing]

\text{meets} = \begin{cases} \text{yes} & \text{if } g \neq 0,\ g \neq \varnothing \text{ and } m \geq N \\ \text{no} & \text{if } g = 0 \text{ or } m + u < N \\ \text{undetermined} & \text{otherwise} \end{cases}

For conjunctive rules (A and B and C): any false → no; otherwise any unknown → unknown; otherwise yes.

The rule is the same in every module; the viewer's parsers re-derive counts from the published answers and reject an inconsistent file. The four data states we always keep apart are: observed value, measured zero, unknown (with its reason) and out of coverage. Example: a Catalan urban-map (MUC) sector with no pending-homes figure does not count as 0 homes; it stays outside the funnel and is reported separately (59 sectors with 15,290 planned homes).

2.3 Whole census sections

The main statistical unit is the 2025 census section (2,151 in the AMB), versioned by year. When a method needs the population «around» a place, it assigns whole sections by their representative point (within a straight-line radius or inside a polygon); it never apportions population by area or interpolates it within a clipped polygon. This avoids manufacturing precision the data lack and keeps totals auditable, at the price of an edge effect we document. Correspondences between section versions (2015, 2021, 2023, 2025) use the code when the geometry is unchanged and, in the ex-post evaluation, harmonised intersection groups (section 4.8). Codes keep their leading zeros (11-digit sections, 6-digit INE municipalities).

2.4 A building's colour is context

Xamfrà tints the original buildings of the base map (footprints in 2D, volumes with unchanged height in 3D) with the thematic colour. It never draws a duplicate footprint nor extrudes buildings by a variable. A census colour on a building is section context, not a measure of that building's residents, income or price; the legend always says so, with the section year. This rule avoids the ecological fallacy [5] in its most visual form.

2.5 Provenance and licences

Every published dataset has a manifest.json with: generation date, URL and fetch date of each source, SHA-256 and size of every input and output, parameters, rules, licences, caveats (in Catalan, English and Spanish) and, where relevant, the sensitivity table. Files are built in a staging folder, validated (counts, joins, geometries, arithmetic identities) and only then swapped in. The viewer's loaders check size and hash against the manifest. When a source declares no licence, the project publishes it with citation (decision of 06-10-2026); an explicit non-commercial licence is flagged (section 8).

2.6 Sensitivity is part of the result

Every threshold is a convention. Each module with decisive thresholds therefore publishes a sensitivity table (variant → result) in its manifest, and the viewer shows it. Our reading is qualitative: which figures are robust (the land of the poles under the z threshold) and which are fragile (the number of poles under the valley split), and should not be quoted without the parameter (section 6).

2.7 Local AI with finite, validated plans

Xamfrà includes a small local language model (Qwen3 1.7B served with Ollama 0.34.4 on loopback) that turns a Catalan or English question into a finite query plan validated against a closed JSON schema [6]. The model receives no datasets, runs no SQL and cannot return any figure to the interface; counts, work area and sources are resolved by deterministic tools. An out-of-coverage question returns unavailable or clarification. We do not present it as a general agent.

2.8 The work area is not the camera

All figures are computed on the chosen work area (AMB, municipality, district, neighbourhood or drawn area), never on what is visible on screen. Asynchronous answers are bound to the key of the work area and filters that requested them and are discarded if the user changes the question. A drawn area does not establish census sections; modules that depend on them say so rather than inventing population.

3. Data infrastructure

3.1 Sources

Table 1 summarises the main sources. The full list, with URL, fetch date and hash, is in each dataset's manifest (Annex A).

Table 1 Main sources, versions and licences.

Domain	Source	Version / period	Licence (declared)
Population	INE register by section; ICGC/Idescat section geometry	1-1-2025	CC BY 4.0
Income	INE Household Income Distribution Atlas (ADRH) [7]	2015–2023	CC BY 4.0
Housing (tenure, non-main dwellings)	INE 2021 Census by section	2021	CC BY 4.0
Buildings and floor area	Spanish Directorate General for Cadastre, INSPIRE Buildings [8]	October 2026	INSPIRE licence; transformed product (commercial use pending review)
Activity points and walking network	OpenStreetMap, Geofabrik extract 29-09-2026	2026	ODbL 1.0
Metropolitan planning	AMB service `ide.amb.cat` PDU/O99_AIM (AOM, ARUAE, ARURE, EPM); normative sheets of annex 2e	initial approval 10-02-2026	not declared (published with citation)
Planning sectors	AMB `Planejament/Sectors`; Catalan urban map (MUC) development sectors (Socrata `6wpu-9y2e`)	rows 02-2024	Generalitat open data
Activity zoning	Harmonised MUC A1–A3	1-1-2024	Generalitat open data
Business parks	SIPAE (Generalitat)	2026	Generalitat open data
Public transport	GTFS of AMB, TMB, FGC, Renfe and TRAM (weekday)	September–October 2026	per operator
Mobility	MITMA district OD matrix from mobile phones [9]	13–17-10-2025	«Basado en datos abiertos MITMA»
Employment	Idescat and TGSS, social-security affiliations by contribution account	2020-12 to 2026-08	reuse with citation
Climate	Landsat 8-9 summer LST; SNCZI flood zones T10–T500; Urban Atlas STL 2018	2018–2025	per source
Protected housing	HPO applicants register; Housing Agency (stock in force, qualifications)	2012–2025	not declared (published with citation)
Prices	MITMA appraisal value; Incasòl rental deposits	2016–2026	per source

3.2 Reference systems and geometry

All metric computations (distances, areas, intersections, grids) use EPSG:25831 (ETRS89 / UTM 31N). Public outputs are GeoJSON in EPSG:4326 with 5 decimals and geometry simplified for drawing (4–8 m); simplification is never used for counting. Distances are straight-line unless stated «walking»: then they use the AMB pedestrian graph derived from OSM (45 partitions, 820,907 merged nodes and 914,007 edges; station-proximity-rule manifest), with multi-source Dijkstra at 1.33 m/s.

3.3 Versions, staging and tests

Each dataset is generated by a script in scripts/ that (i) reads only cached or published files, (ii) writes to a staging folder, (iii) validates and (iv) swaps. Each module has Python unit tests (tests/test_<module>.py) and pure TypeScript tests that read the published files. Generation dates of the datasets cited range from 05-10-2026 22:22 UTC (activity-poles) to 06-10-2026 19:41 UTC (economic-complexity).

4. Methods

4.1 Centralities and activity poles

Question. Where are the economic activity poles and of what type? Which centralities does the PDU propose and which have the ingredients to consolidate?

Activity mass. For each non-residential Cadastre building, gross floor area is split among the cells of a 150 m grid covered by its footprint. A fixed weight is added for each OSM activity point not already inside a non-residential building (shops on the ground floor of apartment blocks, which the Cadastre classifies as residential).

Box 2. Activity mass and hot spots. For cell $i$ :

M_i = \sum_{b} \mathrm{GFA}_b \cdot \frac{A(b \cap i)}{A(b)} + \mathrm{EPS} \cdot n^{\mathrm{OSM}}_i, \qquad \mathrm{EPS} = 70\ \mathrm{m^2}

Getis-Ord statistic [10, 11] with binary weights $w_{ij} = 1$ if $d_{ij} \leq 300$ m:

G^*_i = \frac{\sum_j w_{ij} M_j - \bar{M} \sum_j w_{ij}}{S \sqrt{\dfrac{n \sum_j w_{ij}^2 - \left(\sum_j w_{ij}\right)^2}{n-1}}}

A cell belongs to a pole if $G^*_i \geq 2.58$ and the mean neighbourhood density is ≥ 4,000 m²/ha.

The statistical population is every cell of the 36 municipalities, so the threshold is relative to the AMB; z is used as a threshold, not a p-value (no correction for spatial autocorrelation). Pole cells are closed, connected (queen contiguity) and split at density valleys (a region is split where density drops below 0.8 of the lower peak); basins under 12 ha or 60,000 m² are absorbed or dropped. The type is the family (offices and knowledge, industry and logistics, retail and leisure, public services) with ≥ 50 % of the mass; if none reaches it, the pole is mixed.

Campuses and surroundings. Campuses are built from OSM university or research areas (plus hospitals with ≥ 2 catalogued research or teaching centres within 100 m), grouped when closer than 150 m and of ≥ 1 ha. The zone of influence is the whole sections whose representative point lies ≤ 1.5 km from the edge. A descriptive typology adapts Katz and Wagner [12]: knowledge actors on campus × industrial share of the zone's non-residential floor, each split at its median across the 29 campuses (2 actors; 8.4 %).

Emerging centralities. The 103 PDU proposals (75 AOM and 28 ARUAE) are related to the 65 poles by overlap (a proposal reinforces a pole if ≥ 25 % of its area lies inside). Each unit gets six yes / no / unknown ingredients:

Ingredient	Rule	Reference
transit	metro, rail or tram stations ≤ 800 m from the edge with ≥ 2 lines	ITDP TOD Standard [13]
knowledge	campus ≤ 1 km or catalogued research centre ≤ 500 m	Katz and Wagner [12]
land	vacant parcels ≥ 10 % and ≥ 0.25 ha, or MUC sectors with homes still to build	Brownfield Land Register [14]
people	≥ 20,000 residents in sections within 1 km	own criterion
activity	an activity pole ≤ 500 m	Arzaghi and Henderson [15]
timing	AOM scheduled for sexennium 1 or 2; ARUAE without sexennium = unknown	PDU

The final typology (consolidated and reinforced, existing without proposal, proposal reinforcing a pole, new with ≥ 4 of 6 ingredients, partial 2–3, lacking 0–1) is a rule-based classification, not a ranking.

Trip attraction. For each centrality, morning arrivals are estimated from the MITMA district OD matrix: $\hat{T}_u = \sum_z w_z T_z$ with $w_z = A(u \cap z)/A(z)$ . Because it assumes trips are uniform within a district, the figure is always labelled «estimate» and carries a confidence (high if $\bar{w} \geq 0.5$ ); in the AMB, 157 of 168 units have low confidence (centrality-attraction/manifest.json, 06-10-2026).

Early signals. Five local signals (transformative works licences 2022–2026, new buildings since 2020, ground-floor premises census changes 2022→2024, office and tourism floor area 2021→2026, neighbourhood rent) are counted per section (≥ 2 = «several signals»). Two municipal signals (housing starts, DIRCE firms) are shown but not counted, because they would give the same answer to every section of a municipality (counting them would move sections with ≥ 3 signals from 20 to 78; manifest sensitivity).

4.2 Transformation areas and opportunity detection

Units. Land that could change is organised into 1,626 units in three families (transformation-areas/manifest.json, 06-10-2026): 109 PDU proposals (4,066 ha), 471 AMB planning sectors (2,847 ha) and 1,046 MUC A1–A3 activity-zoning pieces ≥ 0.5 ha (3,061 ha). The EPM (16 corridors of 100–450 ha) are context, not units: at that size every criterion holds. Units overlap (a sector inside an AOM) and their totals are not additive.

Criteria. Six opportunity, two knowledge and three risk criteria, with thresholds in the manifest:

Family	Criterion	Yes when
Opportunity	`pdu_proposal`	is a PDU proposal or ≥ 25 % inside an AOM/ARUAE/ARURE
	`aged_industry`	≥ 50 % of dated industrial buildings predate 1980 (NBE-CT-79)
	`vacant_land`	unbuilt urban parcels ≥ 10 % of the area and ≥ 0.25 ha
	`transit`	rail, metro or tram station ≤ 800 m
	`plan_in_motion`	RPUC plans approved since 2015 (or in process in Barcelona) cover ≥ 10 %
	`low_occupancy`	SIPAE declared occupancy < 70 %
Knowledge	`campus_zone`, `activity_pole`	≥ 25 % inside a campus zone or a pole
Risk	`productive_in_use`	industry or logistics pole ≥ 25 %, or SIPAE occupancy ≥ 85 % (London «no net loss» logic [16])
	`low_income_nearby`	≥ half of the sections within 400 m in the AMB bottom income quartile (ADRH 2023, €14,031)
	`residents_nearby`	≥ 5,000 residents in those sections

Risks are not subtracted from opportunity: Figure 1 shows the opportunity × risk matrix as counts.

Planned housing. The 606 MUC development sectors in the AMB provide potential homes (num_habitg), pending homes (hab_vacant) and free land. We detected and corrected a unit inconsistency: the pending-floor field, described in m², is consistent with hectares (49 of 76 checked rows within 5 %), and is converted while flagged as an inference.

PDU sheets. The normative sheets have no text layer: values are drawn. We extract them by OCR (Tesseract 5, Catalan and English models) at 300 dpi, locating crop boxes from template words and reading the IFI (gross buildability) and IFU (residential share) classes from filled dots. Every OCR'd area matches the service geometry (ratio 0.9999–1.0000), and each document's SHA-1 matches the fingerprint in the AMB's signed diligence (pdu-area-sheets/manifest.json). The sheets do not state a number of homes; when needed, it is derived as a range (area × IFI × IFU ÷ 90 m²/home) and labelled «derived, not official».

Opportunity detector. The detector gathers 1,932 candidates (1,626 units, 65 poles and 241 station areas of 800 m) with 1,804 overlap links, and assesses them through two separate lenses: 8 opportunity signals (planning in motion, housing capacity, land capacity, aged fabric, transit, knowledge, activity, near-term schedule) and 5 constraints (productive land in use, heat, flood, service gaps for new residents and displacement). Planning questions are expressed as editable presets: required criteria (only yes passes), excluded constraints (only no passes, or unknown if the user allows it, counted separately), candidate kinds and minimum signals. For instance, «Affordable housing near transit with services» requires housing capacity and transit and excludes service gaps: 124 strict candidates, 170 if unknown is allowed to pass (docs/OPPORTUNITY-DETECTOR.md, computed in the viewer from the published files). Next-step recommendations are deterministic templates (R1–R12) triggered by a published yes; no free text is generated.

Real-projects register and annual ledger. A curated register of 35 operations (128 sources, 62 official or bulletin) shows where a transformation is under way, with phase and evidence date (8 in planning, 9 in management, 7 in urbanisation, 10 in building, 1 completed). An annual ledger (opportunity-ledger/2026.json, taken 06-10-2026) copies 241 station areas, 1,626 units, 168 centralities and 36 municipalities; snapshots are never rewritten, so each year it will be possible to compare what planning foresaw with what happened.

4.3 Housing capacity: a SHLAA / HBA funnel with a back-test

Question. Of the homes the plans enable, how many survive known constraints, and how many could be completed in 10, 15 or 20 years at a past pace?

We adapt New Zealand's capacity assessment (NPS-UD: plan-enabled → infrastructure-ready → feasible → reasonably expected) [17] and London's SHLAA (site capacity discounted by constraint probabilities) [18].

Constraints are flood (high), heat (medium), productive land in use (high), service gaps (medium), displacement pressure (medium) and no station within 800 m (medium). They are treated as independent; the probabilities are SHLAA conventions, not measured AMB delivery rates. The «feasible» step is not evaluated: neither 0 % nor 100 % is assumed, and the viewer says so (Figure 2).

Back-test. Because the 2015 sector pipeline cannot be rebuilt (the MUC is a 2024 snapshot), the pace rule is tested on the windows available (Table 4, section 6.1). The result is clear: the rule misses by tens of per cent when the housing cycle turns, and horizon figures must be read as orders of magnitude.

4.4 Job accessibility by public transport

Question. From each section, how many jobs can be reached by public transport and walking in 30 and 45 minutes in the morning peak?

Job surface (modelled). Only municipal totals by activity family are official. We split the Cadastre non-residential floor area by family with floor-per-job ratios from the Employment Density Guide [19] (offices 20 m², industry and logistics 55, retail and leisure 30, public services 40, own assumption) and add OSM points (EPS 70 m²). Each section × family cell is then fitted to the official municipality × family total (Idescat and TGSS affiliations, 2026-06) by iterative proportional fitting [20], which here converges in one pass because each cell belongs to a single margin. AMB total: 1,975,385 jobs.

Travel times. We use the viewer's own RAPTOR core [21] on a weekday GTFS timetable: 25 departures from 07:00 to 09:00 every 5 minutes, ≤ 2 transfers, stops ≤ 500 m, straight-line walking × 1.3 at 4.5 km/h, 75 min maximum. Waiting at the first stop counts, following Conway et al. [22]. The result is a 25 × 2,151 × 2,151 matrix.

Box 4. Cumulative and Hansen accessibility. For origin $i$ , departure $\tau$ and jobs $J_j$ :

A^{T}_{i,\tau} = \sum_j J_j \, \mathbf{1}[t_{ij\tau} \leq T], \quad T \in \{30, 45\}; \qquad A^{H}_{i} = \operatorname{med}_\tau \sum_j J_j \, e^{-\beta t_{ij\tau}}

P25, median and P75 over the 25 departures are published. $\beta$ is fitted by trip-weighted least squares over 5-minute bins ( $t \leq 60$ ):

\ln\!\left(\frac{V_b}{\sum_{(z,z') \in b} P_z J_{z'}}\right) = a - \beta \, t_b \quad\Rightarrow\quad \beta = 0.0654\ \mathrm{min}^{-1}\ (\text{half-life } 10.6\ \mathrm{min}),\ R^2 = 0.81

with $V_b$ the MITMA home → work/study trips departing 07–10 h between AMB districts (5,798 pairs, 272,546 trips).

Because MITMA trips cover all modes, β is a deterrence parameter of the model, not a behavioural estimate. A Δ scenario adds the jobs of pending activity floor area (sectors and productive-land gains) and removes those of approved losses: +78,639 net jobs, with the same travel times.

Validation. The surface is validated against three independent sources (section 6.2): MITMA arrivals by district (Spearman ρ 0.91 against 0.78 for population), Barcelona premises floor area by section (ρ 0.89) and a leave-one-municipality-out test.

4.5 Productive land: demand against supply

Question. Will there be enough activity land (MUC A1–A3) for the industrial and logistics demand expected in 2035 and 2045 if the PDU conversions go ahead?

We follow London's industrial-land demand study [23] and the HCA densities [19]:

Supply is today's A1–A3 land, minus approved losses, plus gains, without the PDU and with the PDU (minus proposed conversions). The «no net loss» signal fires when supply with the PDU falls below the range and supply without it does not. An alternative rule projects the historical pace of SIPAE park creation (75.9 ha/year in 1996–2025, 3.3 ha/year in 2011–2025).

4.6 Impacts and safeguards

Displacement pressure. Six criteria are counted per section (renters ≥ AMB Q75; income ≤ Q25; new-lease rent growth 2020→2025 ≥ AMB; tourist housing; residents aged 65+ ≥ Q75; official vulnerability) and a section is flagged when it meets ≥ 3; judicial-district evictions are context and not counted (one figure would cover up to nine municipalities). The criteria describe conditions associated with risk; they do not measure displacement or predict who moves.

Staged typology. At the coordinator's request we adapt the typology of Bates [24] and the Urban Displacement Project [25] with AMB thresholds (US thresholds do not transfer): three blocks (vulnerable: ≥ 2 of 4 signals; demographic change: the AEM's IVU fell ≥ 1 decile 2015→2022; market: accelerating, appreciated, adjacent or stable) are combined through a published lookup table. Unknown values are expanded over every possible combination: a stage is assigned only if all lead to the same result.

Rings of influence. For each unit, pole or station we compute residents, residents in pressure sections, renting households and residents aged 65+ in whole sections within 400 and 800 m. Rings of different units are never summed; they are aggregated by the union of section codes.

Proposal appraisal. The 580 official proposals (109 PDU and 471 sectors, plus 2 activity pieces with a real project inside) go through a battery of checks in three groups (strengths, gaps and safeguards), inspired by the Green Book strategic case [26], APUR's site diagnostic [27] and New York's CEQR thresholds [28]:

Recommendations are fixed templates filled with evidence figures («Reserve land for a school: +{pupils} estimated primary pupils»), with no language model. Contradictions between official sources (e.g. announced homes > MUC planned homes) are flagged for reconciliation, not resolved.

Station-proximity rule. We adapt Copenhagen's proximity principle (~500–600 m) [30]: what share of planned growth lies within a 600 m walk of a station with ≥ 20 departures/h (07–09 h), compared with today's stock?

\text{share}_{600} = \frac{\sum_s E_s \, \mathbf{1}[d^{\mathrm{network}}_s \leq 600]}{\sum_s E_s}

On the walking graph; when the origin cannot be snapped to the graph (new sectors without streets), the straight line is used and labelled: 57 % of pending homes are measured on the network.

4.7 Value capture

Question. For each unit with planned or derived homes, how large is the land-value uplift of the transformation, and how much affordable housing could the legal mechanisms fund?

This is a documented model, not a valuation (not an ECO/805 appraisal). We apply the static residual method to the residential part:

Box 7. Residual method and legal capture. With $F$ residential floor area, $V$ market value of a new home (€/m²), $K$ profit-and-overheads coefficient, $V_c$ construction cost, $U$ urbanisation charges, $\mathrm{EUV}$ existing-use value of land, $r$ protected reserve, $c$ cession and $P$ maximum HPO price:

\mathrm{Res} = F\left(\frac{V}{K} - V_c - U\right), \qquad \mathrm{Uplift} = \mathrm{Res} - \mathrm{EUV} \cdot A \cdot \nu \cdot \phi

\mathrm{Reserve} = F\, r \left[\left(\frac{V}{K} - V_c\right) - \min\!\left(\frac{P}{K} - V_c,\ 0.2P\right)\right]

\mathrm{Cession} = c\, F \left[(1 - r)\left(\frac{V}{K} - V_c\right) + r \cdot \mathrm{land}_{\mathrm{HPO}}\right], \qquad \mathrm{SoBoN} = \tfrac{2}{3}\,\mathrm{Uplift}

( $\nu$ : net private land share; $\phi$ : residential share; HPO land capped at 20 % of $P$ under Law 11/2025.)

Parameters take three values (low / central / high) anchored in sources: $V$ from the MITMA appraisal value 2026-Q2 (with a rent-calibrated proxy where MITMA does not publish, 15 municipalities); $K$ between 1.45 and 1.30 (RD 1492/2011); $V_c$ between €2,100 and €1,700/m² (APCE 2025 + ~15 %); $U$ and the net private land share are declared assumptions. The reserve is 50 % in PDU units (art. 154 of the plan's rules) and the legal 30 % elsewhere (Law 11/2025 exempts plans initially approved before 2026); the new law's shares (40 / 50 %) are shown as sensitivity. The SoBoN comparison (the owner keeps one third of the uplift) [29] is not Catalan law; it serves to locate the legal requirement.

4.8 Descriptive ex-post evaluation

Question. Where land was already transformed (since ~2000), how did the population, income, age, nationality and rent of the surroundings change compared with their municipality?

Fifteen cases (22@, Diagonal Mar, Fòrum, la Mina, Badalona seafront, Plaça Europa, Can Batlló, Finestrelles, la Marina del Prat Vermell, Glòries, Volpelleres, Llevant Viladecans, Fontsanta, Can Sant Joan and Parc de l'Alba) take their perimeter from the instruments in the AMB planning register.

The 2015 and 2023 sections are not identical: each ADRH year matches the ICGC partition of the following 1 January (verified: all 2,137 codes of 2015 and 2,149 of 2023 are present). Sections overlapping ≥ 10 % are connected into 2,060 harmonised units compared as one territory, without interpolation. The area is the set of groups with ≥ 25 % inside the perimeter; the ring, the other groups with ≥ 25 % within 500 m. Flags: small_numbers (< 1,000 residents), low_section_coverage, coarse_groups and window_after_main_phase (plan approved before 2010, so the window shows consolidation).

This is not a causal estimate. The structure is that of a difference in differences [31] but without a constructed counterfactual, verified parallel trends or a significance test; the word we use is «descriptive». An analogue rule assigns each current unit the two most similar past cases (same type +3, context +2, size +1).

4.9 Legal and physical constraints

Question. Which easements, protections and physical constraints touch each transformation unit?

Eight constraints are computed in EPSG:25831 against each unit's published polygon: heritage (AMB Refós catalogue, Generalitat BCIN/BCIL inventory, Diputació protected points), aircraft noise (EEA END round 4, ≥ 55 dB Lden on ≥ 3 pixels), aeronautical easements (always unknown: no open vector source was found), ENAIRE UAS zones (context, never a gap), coast (proxy by distance to the coastline), natural areas (PEIN, Natura 2000, ENPE ≥ 0.1 ha), active contaminated soils (ARC) and OSM power lines ≥ 110 kV (own indicative corridor of 15–30 m half-width, not the legal easement). It is a screening: «no» means «not in the stated open source», not «free of constraints». Answers are copied into the proposal appraisal as gaps (except UAS and easements) and into the detector dossier as context, without redefining its constraint counts.

Public ownership. Only ownership by public bodies is identified (Generalitat and public-sector inventories, IMPSOL promotions), geocoded by street number; the Cadastre owner register is not used. Absence means «not in the open inventories», never «private».

4.10 Criteria crossing views

Ten views answer social and service questions per census section with the rule of Box 1: displacement pressure (6 criteria, ≥ 3), older people alone and heat (5, ≥ 3), well-connected vacant land (gate: vacant land ≥ 0.5 ha; 4, ≥ 3), non-main dwellings with demand (gate: ≥ Q75 of non-main dwellings; 4, ≥ 3), tourism pressure (4, ≥ 3), nursery places (5, ≥ 3), primary-care pressure (5, ≥ 3), energy retrofit (5, ≥ 3), children and play spaces (gate: no play space within a 300 m walk; 3, ≥ 2) and early signals (5, ≥ 2). Thresholds are mostly AMB quartiles computed over sections with a known value, or fixed numbers chosen against the observed spread and written into the rule.

Three common decisions deserve mention. First, distances to services are computed on foot on the OSM graph from Cadastre parcels with dwellings, weighted by dwellings; if the origin does not snap to the graph or the nearest service may lie outside the AMB (edge effect), the answer is unknown, never a circle. Second, municipal-level criteria (hotel places, visitors per resident) give the same answer to every section of a municipality and are labelled as such. Third, service capacity (places, staff) is not openly published: nursery enrolment or the population assigned to a primary-care area are use or load, not capacity, and do not enter any criterion as if they were.

4.11 Economy: specialisation, dynamism and complexity

Specialisation and change. With affiliations by CCAE section and municipality (Idescat, 12-2020 and 12-2025):

Three crossed rules (growing sector with little land; under-used, well-connected business park; dynamic pole with transformation land nearby) are resolved with Kleene logic.

Economic complexity. At municipal scale and over the 295 Catalan municipalities with ≥ 1,000 workers (of 947), with 78 CNAE-2025 divisions (TGSS, 08-2026):

Box 10. Revealed advantage, fitness-complexity and relatedness density.

\mathrm{RCA}_{cs} = \frac{x_{cs}/x_c}{x_s/X}, \quad M_{cs} = \mathbf{1}[\mathrm{RCA}_{cs} \geq 1]

\tilde{F}_c^{(n)} = \sum_s M_{cs} Q_s^{(n-1)}, \qquad \tilde{Q}_s^{(n)} = \frac{1}{\sum_c M_{cs} / F_c^{(n-1)}}

(normalised at each iteration; 60 iterations; Tacchella et al. [33]). Proximity $\varphi_{ss'} = \sum_c M_{cs}M_{cs'} / \max(k_s, k_{s'})$ [34]; density $\omega_{cs} = \sum_{s'} M_{cs'}\varphi_{ss'} / \sum_{s'} \varphi_{ss'}$ [35, 36].

We also compute the Hidalgo–Hausmann ECI [37], but at this resolution it measures the main specialisation axis rather than sophistication (as Mealy et al. warn [38]; correlation with diversity only ρ = 0.11), and we do not rank by it. A diversification candidate is an explicit rule, not a score: no RCA, density ≥ the municipality's own P75, sector complexity > 0 and ubiquity ≥ 10. Stability is measured with 189 bootstrap resamples of the estimation municipalities.

5. Results for the AMB

5.1 Polycentric structure

The hot-spot method identifies 65 activity poles (6,175 ha, 9.7 % of AMB land) that hold 59.5 % of the measured activity mass (45.3 of 76.1 million m²; activity-poles/manifest.json, 05-10-2026). By type: 23 industry and logistics (2,283 ha), 18 public services and facilities (842 ha), 11 mixed (2,323 ha), 8 offices and knowledge (498 ha) and 5 retail and leisure (230 ha). Thirteen municipalities have no pole: no area exceeds the AMB-relative threshold, which does not mean «zero activity».

The 29 campuses (311.6 ha) hold 110 of the 132 catalogued research centres; their zones of influence cover 1,261 of the 2,151 sections and 2,013,713 residents (campus-influence/manifest.json, 06-10-2026). The descriptive typology splits them into four groups of 7, 7, 7 and 8 campuses.

Among the 168 centrality units (emerging-centralities/manifest.json, 06-10-2026), 21 poles are consolidated centralities reinforced by the PDU, 44 poles have no proposal, 37 proposals reinforce an existing pole, 30 new proposals have ≥ 4 of the 6 ingredients (1,632 ha), 22 have 2–3 and 14 have 0–1. The «room and momentum» block (≥ 2 capacity and ≥ 2 planning signals) groups 47 units.

5.2 Opportunity and risk

Of the 1,626 transformation candidates, 191 meet ≥ 4 of 6 opportunity criteria; of these, 173 carry at least one risk flag and only 18 none (Figure 1). Coincidence of opportunity and risk is the rule, not the exception: precisely what a net index would hide.

Figure 1 Opportunity × risk matrix of the 1,626 AMB transformation units. Each cell counts units; risks are not subtracted from opportunity. Source: public/data/transformation-areas/areas.geojson (06-10-2026).

With all lenses loaded, the detector places 155 candidates with ≥ 4 signals and no constraint met, 545 with ≥ 4 signals and some constraint, 299 with < 4 signals and no constraint and 933 with < 4 and some; of the 454 with no constraint met, 321 have some constraint unknown (docs/OPPORTUNITY-DETECTOR.md). The most frequent legal and physical constraints on units are heritage (425 yes / 1,047 no / 154 unknown), power lines (175 / 1,423 / 28) and natural areas (31 / 1,595 / 0) (planning-constraints/manifest.json, 06-10-2026).

5.3 Housing capacity and proximity to transit

Table 2 AMB capacity funnel (default parameters, 15-year horizon, 2015–2025 pace). Source: capacity-funnel/manifest.json (06-10-2026).

Step	Homes
1. Plan-enabled (MUC pending)	70,914
2. Constrained (central / unknowns applied)	56,346 / 51,500
3. Feasible	not evaluated
4. Expected within 10 / 15 / 20 years	30,028 / 37,125 / 41,977
AMB pace, 2015–2025 / 2021–2025 (homes/year)	4,343 / 5,595

Figure 2 AMB housing capacity funnel. The feasibility step is not evaluated and is shown empty. Source: public/data/capacity-funnel/manifest.json (06-10-2026).

The pace, not the land, limits most suburban municipalities (Cerdanyola, Gavà, Montcada, el Papiol, Santa Coloma de Cervelló); in Barcelona, l'Hospitalet and Badalona the land is the limit. As a cross-check, the Observatori del Territori residual homes (2024) equal the MUC pending sum in 33 of the 36 municipalities (182 homes of absolute difference in total).

The proximity rule shows a structural mismatch: only 25 % of pending homes (17,953 of 70,914) and 22 % of AOM and ARUAE proposal hectares lie within a 600 m walk of a well-served station, against 60 % of today's dwellings and 59 % of residents (station-proximity-rule/manifest.json, 06-10-2026). Eighty sectors with 40,468 pending homes lie more than 1 km from a well-served station (Parc de l'Alba, Pla de Ponent, ARE Eixample Sud in el Prat, Llevant in Sant Vicenç dels Horts…). The municipal contrast is sharp: 93 % in Cornellà and 82 % in l'Hospitalet, 0 % in Cerdanyola, Gavà, Viladecans and Barberà.

5.4 Job accessibility

The typical AMB resident (population-weighted median) reaches 485,959 jobs within 30 min and 1,169,969 within 45 min (job-accessibility/manifest.json, 06-10-2026). The distribution is bimodal (Figure 3): a large group of residents in sections with little 30-minute access (periphery and municipalities whose local bus is missing from the data) and another around 800,000–950,000 jobs (Barcelona and the first ring). Departure minute matters: the median section has a P75 − P25 gap of 55,538 jobs at 30 min. 142 sections in 9 municipalities are flagged as a lower bound (local bus outside the published timetable).

Figure 3 AMB residents by jobs reachable within 30 min by public transport and walking (MODELLED). Source: public/data/job-accessibility/ (06-10-2026).

5.5 Productive land

Activity-land supply is 3,172 ha today; with approved losses and gains, 3,677 ha without the PDU and 3,168 ha with the conversions it proposes (industrial-land-demand/amb.json, 06-10-2026). Central demand is 3,109–3,257 ha in 2035 and 3,317–3,475 ha in 2045 (high: 3,561–3,731 ha). Without the PDU, supply stays above every range except the 2045 high one; with the PDU it falls within the 2035 central range and below the 2045 central one. The «no net loss» signal fires in 12 municipalities in 2035 (central) and 10 in 2045. The AMB net balance is almost neutral (−4.5 ha), but 23 municipalities are negative and 385 of the 509 ha of proposed loss are land in use (productive-land-balance/manifest.json).

Displacement pressure. 469 sections (739,689 residents) meet ≥ 3 of 6 criteria; 1,641 do not and 41 are undetermined (impact-lens/manifest.json, 06-10-2026). In relative terms l'Hospitalet stands out (155 of 197 sections). The criteria are correlated: 83 % of low-income sections are also officially vulnerable, so a count of 3 can partly repeat one dimension. The staged typology classifies 422 sections as early market risk, 90 as vulnerable without pressure, 47 as dynamic and 18 as susceptible; late stages are 0 because municipal or neighbourhood market resolution does not detect «appreciated».

Proposal appraisal. The most repeated gaps among the 582 units are tree canopy (313), heritage (230), heat (217), no station within 800 m (209), a protected share < 40 % (150), a park within 300 m (138), low income around (132) and nursery (120); 18 units carry contradictions between official sources (proposal-appraisal/manifest.json, 06-10-2026).

Value capture. For 345 counted units with homes (104,363 / 129,278 / 154,196 homes by scenario), modelled uplift is €−2.7 / 4.8 / 15.8 bn and central legal capture €5.1 bn (reserve 4.75; cession 0.37) (value-capture/manifest.json, 06-10-2026). In 132 of 211 valued units, legal obligations exceed the uplift in the central scenario; 64 have negative uplift before any obligation. The main driver is that HPO at the maximum price does not cover $V_c \cdot K$ : HPO land has a negative value and the reserve cross-subsidises it. Where uplift can be valued, the legal rule funds 62,342 affordable homes (central), against 28,296 under the SoBoN-like rule. Figure 4 shows the range by municipality.

Figure 4 Modelled uplift by municipality: central scenario (dot) and low–high range (line). Documented model, not a valuation. Source: public/data/value-capture/by-municipality.json (06-10-2026).

Protected housing. The AMB had 80,384 registrations in force in the HPO applicants register on 31-12-2025 (62.2 per 1,000 households) against 41,141 HPO homes in force (31.9 ‰) (protected-housing-balance/amb.json, 06-10-2026). At the recent pace of definitive qualifications (1,129/year in 2016–2025; 1,372/year in 2021–2025), covering current registrations would take 59–71 years; the planning pipeline (9,777–19,176 protected homes expected within 15 years) would cover 12–24 % of them. Registrations are not need; the pipeline is not construction (Figure 5).

Figure 5 HPO register applicants and HPO homes in force per 1,000 households, by municipality. Source: public/data/protected-housing-balance/ (06-10-2026).

5.7 Ex-post lessons

Between 2015 and 2023, income per person grew faster than in the municipality in the Barcelona and first-ring transformations (la Marina +26.1 pp, Diagonal Mar +12.4, la Mina +10.3, Plaça Europa +8.7, Can Batlló +8.4, 22@ +7.5), while outer-AMB residential growth (Volpelleres, Finestrelles, Fontsanta, Llevant) tracked its municipality within ±3 pp (ex-post-transformations/cases.json, 06-10-2026; Figure 6). In the 500 m rings, income mostly moved within ±3 pp of the municipality. In the Barcelona register 2002–2025, 22@ grows 85 % against 11 % for the city. Diagonal Mar, the earliest enclave, ages (+4.8 pp more residents aged 65+ than Barcelona) and loses population. These results are consistent with reading 22@ as land-rent capture [39], but do not prove it.

Figure 6 Income-per-person growth 2015→2023 of each transformed area against its municipality. Descriptive, no counterfactual. Source: public/data/ex-post-transformations/cases.json (06-10-2026).

5.8 Services and living conditions (crossing views)

Figure 7 summarises the ten criteria views. Headline figures (manifests of 06-10-2026): 333 sections meet ≥ 3 nursery-pressure criteria, with about 16,820 estimated children aged 0–2; 138 primary-care pressure sections, with 45,186 residents aged 65+; 413 energy-retrofit priority sections (625,818 residents); 67 sections of older people alone and heat, with 7,712 residents aged 80+; 271 sections with 751 ha of well-connected vacant land (of 6,113 ha of core vacant land in the AMB); 344 sections with 74,497 non-main dwellings where there is demand (38 % of the AMB's 197,989); 222 tourism-pressure sections, all in Barcelona; and 65 priority sections without a play space within a 300 m walk (76 % of children aged 0–12 have one within 300 m).

Figure 7 Result of the ten criteria crossing views: sections that meet, undetermined and that do not meet. Source: manifests in public/data/<view>/ (06-10-2026).

5.9 Economy

AMB affiliations grew 15.3 % between 12-2020 and 12-2025 (from 1,692,585 to 1,951,305), led by information and communications (+36.9 %), hospitality (+28.2 %) and professional and scientific activities (+26.4 %), with industry at +3.3 % (docs/ECONOMIC-DYNAMISM.md; economic-dynamism/, 06-10-2026). The rule «growing sector with little land» is true for 15 municipality × sector combinations (2 unknown, 55 no); «under-used, well-connected park» for 3 parks (13 unknown for lack of declared occupancy); «dynamic pole with transformation land nearby» for 6 poles. In economic complexity (fitness-complexity over Catalonia), Barcelona ranks 1st, Sant Cugat del Vallès 2nd and Esplugues de Llobregat 3rd in Catalonia; Viladecans is last of the 32 reliable AMB municipalities (economic-complexity/municipalities.json, 06-10-2026).

6. Validation and uncertainty

6.1 Back-tests

Table 4 Back-test of the capacity funnel's pace rule (capacity-funnel/manifest.json, backtest).

Test	Predicted	Actual	Bias	WAPE
pace 2005–2014 → 2015–2025	77,790	47,775	+62.8 %	69.7 %
pace 2010–2014 → 2015–2025	35,435	47,775	−25.8 %	33.3 %
pace 2015–2019 → 2020–2025	17,839	32,909	−45.8 %	48.4 %
min(2020 potential, 2015–2019 pace × 6) → 2020–2025	17,542	32,909	−46.7 %	49.0 %

The rule over-predicts when trained on an expansion (2005–2014, including the bubble) and under-predicts when trained on a recession. We publish this because it is the most useful information for anyone using the horizon figures: they are orders of magnitude, not forecasts.

6.2 External validation of modelled surfaces

Activity poles against SIPAE. SIPAE parks are not an input to detection. Of the 94 parks of ≥ 10 ha, 38 have at least half inside a pole and 32 touch none; 42.1 % of SIPAE area lies inside poles and 42.7 % of pole area lies inside SIPAE parks. This is a plausibility check, not accuracy.
Modelled jobs. Against MITMA 07–10 h arrivals for work/study in 102 districts: Spearman ρ 0.906 (population, as a naive baseline, 0.776); in Barcelona's 10 districts 0.697 against 0.394. Against Barcelona non-residential premises floor area by section (a source independent of Cadastre INSPIRE): ρ 0.888 over 1,068 sections. Leaving out each municipality, AMB factors fitted on 35 municipalities predict the 36th's total with a 39 % weighted absolute error (median 60 %; ρ 0.96): the uncalibrated model is not transferable (Castellbisbal +178 %, el Prat +104 %, logistics land and airport), which is why it is fitted to official totals.
Pending homes. MUC against the Observatori del Territori: identical in 33 of 36 municipalities.
PDU sheets. OCR area against GIS geometry: 75 of 75 at ratio 0.9999–1.0000; IFI/IFU text ranges against dots: all consistent.

6.3 Sensitivity

Activity poles. The detected land is stable under the z threshold and the EPS (Jaccard ≥ 0.96) and under area minimums, but changes mainly with the absolute minimum density (Jaccard 0.46 at 6,000 m²/ha) and with radius or grid (Figure 8). The number of poles depends on the valley split (39 to 75) without changing the land: it is the most fragile figure and should not be quoted without its parameter.

Figure 8 Activity-pole sensitivity: area overlap (Jaccard) of each variant with the baseline. Source: public/data/activity-poles/manifest.json (05-10-2026).

Other modules (manifests of 06-10-2026):

Emerging centralities: with the ingredient minimum at 3 or 5, new centralities with ingredients go from 30 to 41 or 18; with 1 transit line instead of 2, proposals lacking ingredients fall from 14 to 9.
Displacement typology: requiring ≥ 3 of 4 vulnerability signals, early market risk falls from 422 to 123 sections; requiring a 2-decile fall, «dynamic» drops to 0.
Value capture: ±10 % on market value moves the AMB central uplift between €2.24 and €7.46 bn (±54 %); construction cost, between 2.94 and 6.76. Prices and costs dominate.
Industrial demand 2045 (central, 3,159 ha need): GLA ratio 0.65 → 3,242; proportional to floor change → 3,607; uncalibrated → 3,367.
Proximity rule: at 500 m, 17 % of pending homes; at 800 m, 33 %; straight line at 600 m, 31 % (the network is stricter).
Economic complexity: Spearman correlation of AMB municipal complexity across bootstrap resamples ≥ 0.939 (P5); diversification candidates are less stable (median mean Jaccard 0.48), to be read as an indicative list.

The unknown share is part of the result. Figure 7 shows the early-signals view is undetermined in 1,175 of 2,151 sections: four of its five signals exist only in Barcelona, and a section outside Barcelona can never be «no». Tourism pressure is undetermined in 163 sections (municipalities with both municipal criteria met and section criteria unknown). In the funnel, 213 sectors lack a pending figure; in value capture, 134 counted units have unknown uplift (existing housing fabric, where the residual method does not apply); in public ownership, only 59 of 183 inventory plots have an exact street-number position. In the detector, 321 of the 454 candidates with no constraint met have some constraint unknown. Showing these shares is a safeguard: a composite index would have turned them into zeros or imputed means.

7. Comparison with other tools

The comparison is factual and partial: it relies on public documentation reviewed on 04-10-2026 (Aretian) or on general public descriptions of each tool; we had no access to their internal models.

Tool	What it does	Main difference with Xamfrà
Aretian City Digital Twin [40, 41]	Metropolitan digital twin (2024 prototype; new generation with NVIDIA, Lenovo and other partners, 2025; private-sector SaaS since 29-05-2026) with census-section indicators, a 0–100 composite housing «desirability» score, area typologies, recommended parcels and section-level economic complexity. City Science methods [42].	Aretian publishes results but not the formulas of its indices; Xamfrà publishes every rule and uses no composite index nor applies section values to parcels. We compute economic complexity only at municipal scale, where worker counts make it stable.
UrbanFootprint [43]	US commercial platform for land-use data and scenarios, with impact indicators per scenario.	US scope and data; Xamfrà is AMB-specific, open and with published methods. We share the idea of comparing scenarios with explicit indicators (metropolitan simulator, section 4.6).
CityScope (MIT Media Lab, City Science) [44]	Tangible and digital platform for participatory decisions with real-time scenario feedback.	Xamfrà adopts its idea of «live deltas» when a lever moves, but on official data and with a visible hypothesis label.
Growth Lab (Harvard): Atlas and Metroverse [36, 45]	Economic complexity of countries and cities, product or industry space, diversification opportunities.	We use the same published formulas (RCA, proximity, density, opportunity gain) but with fitness-complexity and an explicit candidate rule, limited to municipalities with ≥ 1,000 workers.

The point of contrast is not data quality but semantics: where other products offer a score, Xamfrà offers the list of conditions that would justify it and leaves the weighting to the decision-maker.

8. Limitations and ethics

No causal claims. No module estimates causal effects. Ex-post comparisons are descriptive; validation correlations are plausibility checks; policy-lever ranges (impact-lens/levers.json) are literature with declared evidence strength, not estimates for the AMB.

Privacy. No personal data are used. Social views operate on section aggregates; registers (tourism, protected housing, firms) are queried only as aggregates; private owners are never identified (Cadastre ownership is protected data and was not requested); property-listing datasets without suitable licences (e.g. idealista) are not used. Real-project records are cited facts, with promoters identified only as bodies or firms that have made them public.

Ecological fallacy and edge effect. A building's colour is section context; a section meeting pressure criteria says nothing about a particular household. Whole sections at a ring's edge enter or leave whole; distances to services just outside the AMB are treated as unknown.

Licences. The Barcelonès and l'Hospitalet noise maps are CC BY-NC-SA and are used only as station-monograph context, flagged; commercial use of the transformed Cadastre product is pending legal review; the licence of the IVU urban-vulnerability index is undeclared; AMB services without a declared licence are published with citation by project decision (06-10-2026). OSM-derived data are ODbL. Any commercial use requires revisiting these points.

Data gaps. There are no open series of works licences or premises censuses outside Barcelona; no service capacity (places, staff); no geolocated municipal or state ownership; no open vector source for aeronautical easements; the PDU is at initial approval; MUC sectors are a 02-2024 snapshot; many modules use scheduled times of a single weekday and straight-line walking.

Risk of misuse. Detecting «opportunities» on cheap land can accelerate gentrification [46]. The detector therefore never shows an opportunity without its displacement and productive-land constraints, and the next-step templates trigger anti-displacement measures when social signals are present.

9. Future work

Annual series. The opportunity ledger (2026 is the first snapshot) will allow comparing foresight and outcome from 2027; this should extend to annual productive-land and housing-pipeline series.
Feasibility. Connect the value-capture viable_after_legal result to the funnel's «feasible» step, at sector scale.
Accessibility. Per-minute departures, walking on the graph instead of straight lines, competition for jobs (Shen's measure) and scenarios with new transit lines.
Displacement. Section series of income (ADRH 2015–2023) and education for the demographic-change block; older people living alone (2021 Census tables).
Missing sources. Aeronautical easements, coastal delimitation, Parc Agrari, state and municipal public ownership, service capacity.
Participatory validation. Review thresholds with AMB and municipal planners and publish the decisions as versioned parameter changes.
Economic complexity. Document the method in its own document (today it is only in the manifest) and integrate it into the viewer.

References

Àrea Metropolitana de Barcelona (2026). Pla director urbanístic metropolità: document d'aprovació inicial (second initial approval, Metropolitan Council, 10-02-2026). https://urbanisme.amb.cat/informar-se/document-aprovacio-inicial · portal: https://urbanisme.amb.cat/descobrir/pdu-metropolita
OECD & European Commission Joint Research Centre (2008). Handbook on Constructing Composite Indicators: Methodology and User Guide. OECD Publishing, Paris. https://doi.org/10.1787/9789264043466-en
Greco, S., Ishizaka, A., Tasiou, M. & Torrisi, G. (2019). On the methodological framework of composite indices: a review of the issues of weighting, aggregation, and robustness. Social Indicators Research 141, 61–94. https://doi.org/10.1007/s11205-017-1832-9
Kleene, S. C. (1952). Introduction to Metamathematics. North-Holland, Amsterdam (§64, strong three-valued logic).
Robinson, W. S. (1950). Ecological correlations and the behavior of individuals. American Sociological Review 15(3), 351–357. https://doi.org/10.2307/2087176
Ollama (2026). Structured outputs (documentation). https://docs.ollama.com/capabilities/structured-outputs · Qwen3-1.7B model (Apache 2.0): https://huggingface.co/Qwen/Qwen3-1.7B
Instituto Nacional de Estadística (2025). Atlas de distribución de renta de los hogares (experimental statistics). https://www.ine.es/experimental/atlas/experimental_atlas.htm
Dirección General del Catastro. Servicios INSPIRE de Cartografía Catastral. https://www.catastro.hacienda.gob.es/webinspire/index.html
Ministerio de Transportes y Movilidad Sostenible. Estudio de la movilidad con Big Data (open data). https://www.transportes.gob.es/ministerio/proyectos-singulares/estudios-de-movilidad-con-big-data
Getis, A. & Ord, J. K. (1992). The analysis of spatial association by use of distance statistics. Geographical Analysis 24(3), 189–206. https://doi.org/10.1111/j.1538-4632.1992.tb00261.x
Ord, J. K. & Getis, A. (1995). Local spatial autocorrelation statistics: distributional issues and an application. Geographical Analysis 27(4), 286–306. https://doi.org/10.1111/j.1538-4632.1995.tb00912.x
Katz, B. & Wagner, J. (2014). The Rise of Innovation Districts: A New Geography of Innovation in America. Brookings Institution. https://www.brookings.edu/articles/rise-of-innovation-districts/
Institute for Transportation and Development Policy (2017). TOD Standard, 3rd ed. https://www.itdp.org/publication/tod-standard/
UK Government (2017). Brownfield land registers (planning guidance). https://www.gov.uk/guidance/brownfield-land-registers
Arzaghi, M. & Henderson, J. V. (2008). Networking off Madison Avenue. Review of Economic Studies 75(4), 1011–1038. https://doi.org/10.1111/j.1467-937X.2008.00499.x
Greater London Authority (2021). The London Plan 2021, chapter 6 «Economy» (industrial land policies E4–E7). https://www.london.gov.uk/programmes-strategies/planning/london-plan/the-london-plan-2021-online/chapter-6-economy
Ministry for the Environment (New Zealand) (2020). Guidance on Housing and Business Development Capacity Assessments (HBAs) under the National Policy Statement on Urban Development. https://environment.govt.nz/assets/publications/Files/mfe-Guidance-on-Housing-and-Business-Development-Capacity-Assessments-HBAs-under-the-NPS-2020.pdf
Greater London Authority (2017). Strategic Housing Land Availability Assessment: methodology. https://www.london.gov.uk/sites/default/files/shlaa_methodology_6_feb_2017.pdf
Homes and Communities Agency (2015). Employment Density Guide, 3rd ed. https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/484133/employment_density_guide_3rd_edition.pdf
Deming, W. E. & Stephan, F. F. (1940). On a least squares adjustment of a sampled frequency table when the expected marginal totals are known. Annals of Mathematical Statistics 11(4), 427–444. https://doi.org/10.1214/aoms/1177731829
Delling, D., Pajor, T. & Werneck, R. F. (2015). Round-based public transit routing. Transportation Science 49(3), 591–604. https://doi.org/10.1287/trsc.2014.0534
Conway, M. W., Byrd, A. & van der Linden, M. (2017). Evidence-based transit and land use sketch planning using interactive accessibility methods on combined schedule and headway-based networks. Transportation Research Record 2653, 45–53. https://doi.org/10.3141/2653-06
CAG Consultants for the Greater London Authority (2017). London Industrial Land Demand Study. https://www.london.gov.uk/sites/default/files/london_industrial_land_demand_study_2017_commissioned_by_the_gla.pdf
Bates, L. K. (2013). Gentrification and Displacement Study: Implementing an Equitable Inclusive Development Strategy in the Context of Gentrification. City of Portland, Bureau of Planning and Sustainability. https://doi.org/10.15760/report-01
Urban Displacement Project (UC Berkeley / University of Toronto). Displacement typologies: methodology. https://www.urbandisplacement.org/
HM Treasury (2022). The Green Book: Central Government Guidance on Appraisal and Evaluation. https://www.gov.uk/government/publications/the-green-book-appraisal-and-evaluation-in-central-government
Atelier parisien d'urbanisme (APUR). Observatoire des quartiers de gare du Grand Paris Express. https://www.apur.org/en/urban-planning/station-neighbourhood/observatory-station-neighbourhoods-grand-paris-express-network
NYC Mayor's Office of Environmental Coordination. CEQR Technical Manual (community facilities and open space chapters). https://www.nyc.gov/site/oec/environmental-quality-review/technical-manual.page
Housing2030 / Landeshauptstadt München. Sozialgerechte Bodennutzung (SoBoN): socially just land use in the city of Munich. https://www.housing2030.org/project/sozialgerechte-bodennutzung-socially-just-land-use-in-the-city-of-munich-germany/
Centre for Public Impact. Land use and transport in Denmark (Finger Plan and station-proximity principle). https://centreforpublicimpact.org/public-impact-fundamentals/land-use-and-transport-in-denmark/
Angrist, J. D. & Pischke, J.-S. (2009). Mostly Harmless Econometrics, ch. 5. Princeton University Press. https://press.princeton.edu/books/paperback/9780691120355/mostly-harmless-econometrics
Dunn, E. S. (1960). A statistical and analytical technique for regional analysis. Papers and Proceedings of the Regional Science Association 6, 97–112. https://doi.org/10.1007/BF01947767
Tacchella, A., Cristelli, M., Caldarelli, G., Gabrielli, A. & Pietronero, L. (2012). A new metrics for countries' fitness and products' complexity. Scientific Reports 2, 723. https://doi.org/10.1038/srep00723
Hidalgo, C. A., Klinger, B., Barabási, A.-L. & Hausmann, R. (2007). The product space conditions the development of nations. Science 317, 482–487. https://doi.org/10.1126/science.1144581
Neffke, F., Henning, M. & Boschma, R. (2011). How do regions diversify over time? Industry relatedness and the development of new growth paths in regions. Economic Geography 87(3), 237–265. https://doi.org/10.1111/j.1944-8287.2011.01121.x
Hausmann, R., Hidalgo, C. A. et al. (2014). The Atlas of Economic Complexity: Mapping Paths to Prosperity. MIT Press. https://doi.org/10.7551/mitpress/9647.001.0001 · https://atlas.cid.harvard.edu/
Hidalgo, C. A. & Hausmann, R. (2009). The building blocks of economic complexity. PNAS 106(26), 10570–10575. https://doi.org/10.1073/pnas.0900943106
Mealy, P., Farmer, J. D. & Teytelboym, A. (2019). Interpreting economic complexity. Science Advances 5(1), eaau1705. https://doi.org/10.1126/sciadv.aau1705
Charnock, G., Purcell, T. F. & Ribera-Fumaz, R. (2014). City of rents: the limits to the Barcelona model of urban competitiveness. International Journal of Urban and Regional Research 38(1), 198–217. https://doi.org/10.1111/1468-2427.12103
Aretian (2025). Aretian unveils the Barcelona Metropolitan Digital Twin. https://www.aretian.com/post/aretian-unveils-the-barcelona-metropolitan-digital-twin
Aretian (2026). Aretian officially launches its City Digital Twin for the private sector. https://www.aretian.com/post/aretian-officially-launches-its-city-digital-twin-for-the-private-sector
Gras, R. & Burke, J. (2024). City Science: Performance Follows Form. Actar, New York/Barcelona.
UrbanFootprint. Urban intelligence platform (product website). https://urbanfootprint.com/
MIT Media Lab, City Science Group. CityScope. https://www.media.mit.edu/projects/cityscope/overview/
Harvard Growth Lab. Metroverse: an urban economy navigator. https://metroverse.cid.harvard.edu/
Anguelovski, I., Connolly, J. J. T., Cole, H. et al. (2022). Green gentrification in European and North American cities. Nature Communications 13, 3816. https://doi.org/10.1038/s41467-022-31572-1
Saltelli, A., Ratto, M., Andres, T. et al. (2008). Global Sensitivity Analysis: The Primer. Wiley. https://doi.org/10.1002/9780470725184

Annex A. Dataset catalogue

All datasets live in public/data/<dataset>/ with a manifest.json holding inputs (URL, fetch date, SHA-256), parameters, rules, licences, caveats and output hashes. Date = the manifest's generated_at_utc. Document = docs/<DOC>.md.

Dataset	Generated (UTC)	Main units	Method document
`activity-poles`	2026-10-05 22:22	65 poles	ACTIVITY-POLES
`activity-surface`	2026-10-06 16:47	11,204 cells of 150 m	ACTIVITY-POLES
`campus-influence`	2026-10-06 16:06	29 campuses, 1,261 sections	CAMPUS-INFLUENCE
`transformation-areas`	2026-10-06 16:23	1,626 units, 606 MUC sectors	TRANSFORMATION-AREAS
`emerging-centralities`	2026-10-06 16:34	168 units, 16 EPM	EMERGING-CENTRALITIES
`centrality-attraction`	2026-10-06 18:08	168 units	CENTRALITY-ATTRACTION
`early-signals`	2026-10-06 19:01	2,151 sections, 1,691 units	EARLY-SIGNALS
`station-potential`	2026-10-06 16:39	241 station areas	STATION-POTENTIAL
`station-proximity-rule`	2026-10-06 17:04	606 sectors, 1,626 units	STATION-PROXIMITY-RULE
`station-monographs`	2026-10-06 17:02	241 monographs	STATION-MONOGRAPH
`opportunity-ledger`	2026-10-06 16:57	2026 snapshot	OPPORTUNITY-LEDGER
`pdu-area-sheets`	2026-10-06 17:40	86 sheets	PDU-AREA-SHEETS
`real-projects`	2026-10-06 17:32	35 projects	REAL-PROJECTS
`capacity-funnel`	2026-10-06 17:09	606 sectors	CAPACITY-FUNNEL
`job-accessibility`	2026-10-06 17:15	2,151 sections	JOB-ACCESSIBILITY
`industrial-land-demand`	2026-10-06 17:11	36 municipalities	INDUSTRIAL-LAND-DEMAND
`productive-land-balance`	2026-10-06 16:40	activity-land pieces	(INDUSTRIAL-LAND-DEMAND)
`impact-lens`	2026-10-06 17:06	2,151 sections, 1,932 units	IMPACT-LENS
`new-homes-services`	2026-10-06 16:40	606 sectors	(TRANSFORMATION-AREAS, PROPOSAL-APPRAISAL)
`transformation-climate`	2026-10-06 16:37	1,626 units	(OPPORTUNITY-DETECTOR)
`metro-scenarios`	2026-10-06 16:56	36 municipalities	METRO-SCENARIOS
`proposal-appraisal`	2026-10-06 18:54	582 units	PROPOSAL-APPRAISAL
`planning-constraints`	2026-10-06 18:51	1,626 units	PLANNING-CONSTRAINTS
`public-land`	2026-10-06 17:52	59 located assets	PUBLIC-LAND
`value-capture`	2026-10-06 18:44	582 units	VALUE-CAPTURE
`ex-post-transformations`	2026-10-06 18:11	15 cases	EX-POST-TRANSFORMATIONS
`protected-housing-balance`	2026-10-06 18:57	36 municipalities	PROTECTED-HOUSING-BALANCE
`older-alone-heat`, `infill-opportunity`, `housing-mobilisation`	2026-10-06 17:16–17:18	2,151 sections	CRITERIA-CROSSING-VIEWS
`tourism-pressure`	2026-10-06 18:07	2,151 sections	TOURISM-PRESSURE
`nursery-pressure`	2026-10-06 18:44	2,151 sections, 568 centres	NURSERY-PRESSURE
`primary-care-pressure`	2026-10-06 18:49	2,151 sections, 147 health areas	PRIMARY-CARE-PRESSURE
`energy-retrofit`	2026-10-06 17:24	2,151 sections	ENERGY-RETROFIT
`play-spaces`	2026-10-06 17:39	2,151 sections, 1,784 play spaces	PLAY-SPACES
`economic-dynamism`	2026-10-06 19:20	36 municipalities, 65 poles, 154 parks	ECONOMIC-DYNAMISM
`economic-complexity`	2026-10-06 19:41	36 municipalities (295 for estimation)	(manifest only)

Figures. Generated by scripts/build_paper_figures.py from the files above; provenance (paths, dates and SHA-256 of each input) is in docs/paper/figures/figures.json. Palette validated with the data-visualisation palette validator (three-colour categorical, all pairs; single-hue ordinal ramps).

Reproduce.

.venv-data/bin/python scripts/build_paper_figures.py           # CA/EN SVG + figures.json
.venv-data/bin/python scripts/build_paper_html.py              # docs/paper/xamfra-methods-paper.html
.venv-data/bin/python scripts/build_paper_html.py --lang en    # docs/paper/xamfra-methods-paper.en.html