Where Bangalore's next government should invest

01Executive summary

We scored each of Bangalore's 369 wards against seven infrastructure dimensions — mobility & connectivity, health access, education, civic safety, water & sanitation, green space, and housing quality — combining peer-relative regression with India's official URDPFI 2014 urban planning norms (and SDG 6 / Jal Jeevan / PMAY-U coverage targets where applicable). Scores run 0–100 where higher means more underserved.

Five findings across all seven dimensions — each independently verified:

• North Corporation inherits the worst infrastructure backlog. Under the new GBA structure (5 corporations replacing BBMP since September 2025), North Corporation averages a gap score of 69 across its 72 wards, with 29 wards in severe-deficit territory (score ≥ 80). East Corporation is a close second at 62 with 10 severe-deficit wards. Central Corporation averages just 40 — the best-served. Of 74 total severe-deficit wards, 39 are in North + East corporations alone.
• Connectivity & mobility gaps dominate the outer growth corridors. 63 wards show mobility & connectivity as their single biggest deficit. Multiple outer wards register fewer than 5 BMTC stops for 20k+ residents — Mangammana Palya (South, 53k, 0 stops), Hegganahalli (West, 52k, 2 stops), Srigandhanagar (West, 31k, 2 stops). The Whitefield IT corridor is split across several new wards in East Corporation; even after the March 2023 Purple Line extension to Kadugodi, most of the wider area is more than 1 km from a metro station.
• Civic safety is a city-wide shortfall — structural, not local. The dataset registers 12 fire stations for 12.5 million residents — URDPFI's 1-per-200,000 norm calls for ~63. The 12-station figure matches the 2013 Karnataka Fire & Emergency Services Annual Report verbatim. 55 wards show civic safety as their biggest deficit; many have zero police AND zero fire stations for 50k+ residents. This is the one finding no corporation can fix locally — it needs coordinated GBA-level investment.
• Health and education capacity is critically short in the largest wards. Dodda Bidarakallu (West Corp) — population 122,996 — has only 3 health facilities where URDPFI would require ~8. 35 wards have health as their top deficit; 41 have education. Notable education gaps include Mangammana Palya and Hongasandra in South Corp.
• Water & sanitation gaps cluster in the North — but Census 2011 is stale. 63 wards flag water & sanitation as their top deficit. North Corp wards Mallasandra, Hebbal and Shettihalli all register piped-water and sewer coverage below 20% in the Census 2011 baseline. These numbers are 15 years stale — Jal Jeevan Mission and BWSSB expansion have added ~50k connections per year since 2018. Use the geographic pattern (North Corp outer growth corridors lag) rather than exact percentages for planning. BWSSB's own city-wide numbers tell a sobering story: 65 L/day per-capita supply (URDPFI norm: 135), sewage treatment covers only 52% of generation (721 MLD treated vs 1,400 MLD generated).

02What the next government should build — Top 20 Priority Wards

These are the twenty wards where ward-level infrastructure investment would have the highest direct impact on resident welfare. Each row shows the score and the single biggest missing thing.

03The Manifesto Kit — one priority per ward

For each of the top-20 deficit wards, the single most-impactful intervention a corporator could promise. Numbers come from the URDPFI 1-per-N norms (verified) applied to the ward's actual population and current facility count.

04Top 10 Opportunities — biggest impact per new facility

If the new corporations had budget for just 10 new facilities across Bangalore, these are the ten placements that would serve the most currently-unserved people. Ranked by population × shortfall.

05The question this report answers

This is a supply-vs-need report focused on physical infrastructure: mobility, health, education, civic safety, water & sanitation, and parks. Retail / commercial supply is intentionally excluded — those are market signals, not government responsibilities.

The report does not measure service quality, staffing, or political priority. It measures provisioning gaps, nothing more. But provisioning is what corporators can actually promise and deliver.

06The seven infrastructure dimensions

07Methodology — how the gap is computed

Two-axis gap composition

Every dimension's gap is measured two ways, then combined:

Axis A — Peer-relative residual

For each dimension, we fit a log-log regression across all 369 wards:

log(supply_d + ε) ~ log(adjusted_need) + log(area_km²) + log(pop_density)

The residual tells us which wards have less supply than their peers with similar population, area, and density. Adjusted need uses a daytime-load multiplier (IT / business / professional) and transient multiplier (hotels + coaching), ranging 1.00×–1.70×.

Axis B — URDPFI absolute floor

The Ministry of Urban Development's Urban and Regional Development Plans Formulation and Implementation Guidelines (2014) specify provisioning norms per population unit. We compute a shortfall ratio per ward per facility type, with each norm tagged by source:

The peer axis catches contextual underservice. The absolute-floor axis catches city-wide shortfalls. The clearest example: Bangalore has 12 fire stations when URDPFI implies it needs ~63. A purely peer-relative view would call this normal; the absolute floor correctly surfaces it as a city-wide crisis.

08City-wide results

BBMP zone gradient

Average gap score per administrative zone. Bar length scales with score; red = worse off.

Top-deficit dimension distribution

09Per-dimension deep dive

10Ward case studies

11Best-served wards (for contrast)

CBD wards and affluent old residential — exactly where you'd expect. Included as a sanity check.

12Reality-check log & known limitations

Issues found and fixed before publishing

• Bus-stop source was BBMP's fractional reapportioned counts. Switched to Overture's integer counts (KG Halli went from 1 → 8 stops). Mobility URDPFI short-wards dropped from 69% → 49%.
• Daily retail was originally included and flagged 98% of wards short — Overture undercounts kiranas. Dropped entirely in v1.1: retail isn't infrastructure, it's a market signal.
• Bus-stop and daily-retail floors were labeled as URDPFI — they're not. URDPFI has no bus-stop or retail norm. Bus-stop is now flagged as "Heuristic"; daily-retail dimension removed.
• Whitefield metro narrative initially suggested the model contradicted the 2023 metro extension. Investigation showed the administrative Whitefield ward centroid sits ~3 km from the nearest metro station — the March 2023 Kadugodi extension serves a different part of the broader Whitefield area. Verified against the data on every report regeneration.

Standing limitations

• Water/sanitation data is from Census 2011. 15 years stale. Individual ward percentages may have improved significantly. Use the pattern (outer wards worse than inner), not the exact percentages, for investment planning.
• Ward layout: GBA November 2025 369-ward delimitation (current). This report uses the final GBA wards map notified November 19, 2025 — the same boundaries that will be used for the June 2026 elections. Five new corporations (North, South, East, West, Central) replace the old single BBMP. Feature engineering is done via fresh spatial joins to the new boundaries (Overture POIs, buildings, roads, BMTC stops) plus area-weighted reapportionment from the old 243-ward feature matrix for Census 2011 housing, Economic Census, BBMP public goods (facility counts), and property tax.
• Supply matrix uses min-max scaled raw counts, not per-km² densities. This drags R² down for POI-driven dimensions (mobility 0.15, health 0.24, water 0.15, green 0.26). Qualitative rankings are still valid but regressions are noisier than ideal.
• Daytime demand is proxied by establishment counts, which overweights CBD and underweights mega-campus IT corridors. Whitefield's effective demand multiplier (1.50) is probably understated.
• The model does not measure service quality — a ward with one large, well-run hospital scores the same as one with one small, failing clinic. Facility counts only.
• Bus-stop floor (1 per 3k) is a heuristic, not URDPFI. Treat as advisory.

13Ward-level detail reports

Click a ward tab below to see its full infrastructure profile: all six dimension scores, URDPFI facility counts vs provisioning norms, connectivity data, and the specific priority intervention for that ward. Tabs are ranked by overall gap score — #1 is the most underserved.

14Satellite validation — independent ground truth

Every claim in this report has also been cross-checked against ESA WorldCover 2021 v2.0, a 10-metre global land-cover raster derived from Sentinel-1 and Sentinel-2 imagery. This is a completely separate data pipeline from the Overture/BMTC/Census sources used to build the gap model, so agreement between the two is strong independent evidence that the findings are real.

City-wide land-cover composition

Derived from 16.3 million 10m pixels inside the GBA boundary. Built-up dominates at 61% — consistent with a fully urbanised municipal area.

Four validation tests

Top-10 deficit wards — satellite fingerprints

Each tile shows the land-cover composition of one deficit ward, at 10-metre resolution, coloured by ESA WorldCover class: built-up, tree cover, cropland, water. The pattern is consistent — deficit wards are densely built-up with active residential populations, not sparse or rural.

Govindapura

North · pop 36,207

Built 89% · Tree 7% · Crop 3%

100 ≋ Water & Sanitation

Andrahalli

West · pop 148,869

Built 55% · Tree 11% · Crop 20%

100 ■ Civic & Safety

Mallasandra

North · pop 50,220

Built 73% · Tree 11% · Crop 6%

100 ≋ Water & Sanitation

J.P PARK

North · pop 28,304

Built 52% · Tree 37% · Crop 4%

99 ✚ Health Access

Dodda Bidarakallu

West · pop 122,996

Built 64% · Tree 16% · Crop 11%

99 ✚ Health Access

Shettihalli

North · pop 83,569

Built 44% · Tree 37% · Crop 7%

99 ✚ Health Access

Vishwapriya Nagara

South · pop 44,021

Built 61% · Tree 9% · Crop 16%

98 ✚ Health Access

Kempegowda Layout

West · pop 22,053

Built 98% · Tree 1% · Crop 1%

98 ⌂ Housing Quality

Bhoopasandra

North · pop 52,692

Built 54% · Tree 34% · Crop 10%

98 ⌂ Housing Quality

Mangammana Palya

South · pop 53,432

Built 94% · Tree 3% · Crop 0%

98 ◎ Mobility & Connectivity

Data sources for validation

15Accessibility proof — walk-distance coverage per ward

Methodology

• Bus stop access (400m) — % of ward area within 400 metres (≈ 5-minute walk) of a BMTC bus stop. 400 m is the URDPFI walking-access norm.
• Hospital access (1km) — % of ward area within 1 km of an OSM hospital or clinic point. 1 km is a commonly used health-access threshold.
• School access (1km) — % of ward area within 1 km of an OSM school.
• Distance to fire / metro / police — straight-line distance from the ward centroid to the nearest facility, in kilometres.
• Data source: fresh OSM Overpass query for amenity=hospital, clinic, school, fire_station, police, railway=station + subway. Total 2,698 facility points — completely separate pipeline from Overture and BBMP.

City-wide accessibility verdict

The farthest-from-fire ward is Gottigere at 14.3 km from the nearest station. Structurally, Bangalore has 12 fire stations for 12.5M people — accessibility is constrained by count, not by placement.

The iron-clad correlation test

If our gap scores are real, wards with higher gap should have lower measured accessibility. This is a direct, quantitative falsification test.

All three correlations are negative and statistically meaningful across 369 wards. The strongest signal is school access at r = -0.433.

Corporation-level accessibility table

Each cell is colour-graded: teal is best, red is worst.

Key finding: East Corporation has the worst bus stop coverage at just 60% — indicates poor transit network density, not just count. North Corporation has the worst school access at 58% — reflecting outer-corridor growth running ahead of education infra.

Top-15 deficit wards — accessibility fingerprint

Contrast: best-served 5 wards

Best-served wards typically hit 100% coverage on all three walking-access metrics. The contrast with the top-15 deficit list (20-50% coverage, 10+ km to fire station in extreme cases) is the infrastructure gap made concrete.

16Why this approach beats census-only analyses

India's decadal Census is the gold standard for population and household-level statistics — but for a ward-level infrastructure gap report aimed at 2026 elections, census-only methods have three fatal weaknesses. This report uses six independent data sources, cross-validated against satellite pixels and fresh OSM contributions, so findings hold up even if any single source is wrong.

Data freshness — the dealbreaker for census-only work

Six sources, not one — multi-source triangulation

Every major finding in this report is backed by at least two independent data pipelines. Independent means different collection methods, different contributors, different update cadences.

Pearson correlations between independent sources: Overture buildings ↔ satellite built-up pixels r = 0.83. Different pipelines, same result — that's strong evidence.

Counts vs accessibility — a conceptual upgrade

Census tells you how many households in a ward have a piped-water connection. It does not tell you how far you have to walk to a hospital, or whether a bus stop is within 400 metres of every resident. This report measures walk-distance coverage for every facility type across every ward, using distance-transforms over actual facility locations. That is the difference between what exists and what is reachable.

Continuously rebuildable

Census is released once a decade. This report's pipeline — feature matrix, gap model, satellite validation, accessibility, graph analysis — can be re-run end-to-end in under 20 seconds whenever new OSM or Overture data arrives. Every claim in this report carries a traceable data lineage and can be audited against the live parquet files on the compute server.

17Ward embeddings & influencer graph

Beyond per-ward gap scores, two graph analyses let the report do things a single-number score can't: find peer wards (so policy lessons from well- served wards can be transferred to similar-structure deficit wards), and identify influencer wards (where infrastructure investment has multiplicative rather than local effects).

Ward embeddings — a vector for every ward

Each of the 369 wards is represented as an 8-dimensional vector computed by running PCA on 29 infrastructure features simultaneously (population, density, buildings, roads, POIs, water %, housing quality, etc.). In this embedding space, wards with similar infrastructure profiles end up near each other. Distance in the embedding = “how similar are these two wards across everything the model measures”.

Peer wards for the top-10 deficit

For each deficit ward, we list its 5 nearest neighbours in embedding space — these are the wards whose full infra profile matches most closely. This is practically useful: if ward A solved a connectivity problem, its peers B/C/D are likely to benefit from the same intervention.

The peer sets for extreme-deficit wards cluster strongly by geography and character — e.g., Govindapura's peers are all other North/West outer-corridor water-crisis wards. This is not a coincidence; it's the embedding rediscovering the underlying urban structure without being told.

BMTC transit graph & influencer PageRank

Using GTFS data (4,284 routes, 9,498 stops, 1.49 million stop-times), we built a ward-ward transit graph where edge weight = number of BMTC routes that visit both wards. The result: 369 nodes, 21,438 edges — a dense mesh reflecting Bangalore's transit reach.

Running PageRank on this graph identifies wards that sit at the heart of the transit network. A high-PageRank ward is one that shares routes with many other wards, often transitively — it's a hub. Improvements to service at a hub ripple to every ward connected via shared routes.

Top 15 transit-influencer wards

What this enables for policy

• Peer-based policy transfer: if Nagavara (deficit, score 95) has peer Mallasandra (deficit, score 99) — a successful intervention at Nagavara will likely translate to Mallasandra given their near-identical infra profiles.
• Transit-ROI prioritisation: investment in the 15 transit-influencer wards improves connectivity for the majority of the BMTC-served city; investment in low-PageRank outer wards needs to be paired with network-extension projects to achieve similar scale.
• Cluster identification: the embedding space naturally clusters wards by infra character. Every outer-corridor water-crisis ward has the same half-dozen peers — an analyst can look at one, know the dozen-ward peer set is similar, and size policy at the cluster level.

Reproducibility

Full pipeline at gap_model/scripts/13_embeddings_graph.py. Outputs: ward_embeddings.parquet, ward_similarity_graph.parquet (1,845 edges), ward_transit_graph.parquet (21,438 edges), and ward_graph_metrics.parquet (per-ward PageRank + degree). All reproducible from the feature matrix + GTFS — no external ML models, just PCA + NearestNeighbors + NetworkX PageRank.

18Honest scorecard — what this validation really proves

What satellite imagery genuinely proves

• Where people live: ESA WorldCover confirms that deficit wards are built-up (68% avg vs 60% city). Overture buildings at r=0.83 with satellite built-up pixels. Two-pipeline agreement on urban extent is real validation.
• Urban growth direction: cropland-to-built-up transition maps cleanly to outer corps. Population has moved faster than facilities — that spatial pattern is visible in pixels.
• Presence of large green / water features: parks and lakes >10m are visible. BBMP lake counts correlate with pixel water (4× more in lake-recorded wards).

What satellite imagery does not prove

• Water-supply quality. A 10-metre pixel can't see inside a pipe. Census 2011's "6.6% piped water in Mallasandra" is still the only ward-level piped-water signal we have, and it's 14 years old. Satellite confirms the ward is populated, not that its water situation is still bad.
• Health facility functionality. An OSM point tagged amenity=hospital exists — but we don't know its bed count, staffing, opening hours, or if it's operational. Deficit wards might have hospitals on paper that aren't running.
• Transit service reliability. GTFS tells us the schedule. It doesn't tell us bus-on-road frequency, overcrowding, or whether a ward's assigned routes actually arrive.
• Fire / police response time. We measure straight-line distance. Actual response depends on road network state, traffic, and staffing — none of which we observe.
• Sewerage connection quality. Even if Jal Jeevan/BWSSB added connections since 2011, we can't verify operational status from pixels.
• Education quality. A school POI's existence ≠ enrolment ≠ pass rates ≠ teacher-student ratio.

What the three-pipeline agreement does prove

Where Overture (Meta/Microsoft/Esri community), ESA WorldCover (Sentinel satellites), and OSM (independent community) all agree, the underlying physical fact is very likely real. They agree on:

• Population distribution — Kontur matches Overture building density matches satellite built-up.
• Facility locations — OSM hospital points land in wards that Overture flags as health-POI dense.
• Urban extent — no deficit-flagged ward turns out to be a farm or lake.

Where we are intentionally relying on single-source signals

• Water/sanitation coverage: Census 2011 only — flagged. The pattern (outer wards lag) is likely still true; the exact percentages are 14 years stale.
• Fire/police counts: BBMP administrative records. The 12-fire-station figure matches the 2013 Karnataka Fire & Emergency Services Annual Report, but no live audit this decade.
• Housing quality: Census 2011 dwelling-condition self-reports. No satellite substitute exists at ward scale.

Summary verdict

19Auxiliary truth sets — what else we could validate against

Dozens of freely-accessible datasets exist that would strengthen this report's functional-quality claims. This section maps each infra dimension to specific validation datasets we could integrate in a v2.3 update.

Datasets we could add right now (free / open)

Datasets that would require authentication or partnership

• Planet Labs PlanetScope — 3m daily imagery. Paid but some research programmes.
• Google Earth Engine — free for research/non-profit; gives access to VIIRS, MODIS, Sentinel archive.
• BESCOM feeder-level consumption — proves electrification + activity at street scale.
• BWSSB SCADA water-supply data — actual flow rates per DMA (district metering area).
• KSRTC / BMRCL ridership — actual usage, not just schedule.
• Mobile-phone-based movement data (anonymised) — commuter flows between wards.

Why we didn't integrate all of this yet

This report was built in a single session on a compute server. The sources in the first table are all freely accessible and could be wired in within 1–2 days each. The prioritisation should be driven by which claim is most politically contested or operationally important:

• If the "water crisis" claim is contested → pull Jal Jeevan Mission current coverage.
• If the "fire station shortage" claim is questioned → obtain current BBMP fire station list via RTI.
• If the "North Corp worst" finding is contested → cross-check against NITI Aayog municipal performance ranks.
• If the "outer-ring growth" narrative needs strengthening → GHSL 1975-2030 time-series visualisation.

What's robust today without more data

Even without any additional validation, the spatial structure of the findings is well-supported: three independent data pipelines agree that the deficit wards are populated, outer-ring, and under-provisioned on measured accessibility metrics. Adding auxiliary truth sets would harden the service-quality claims (which we've already flagged as directionally rather than precisely proven) — they wouldn't overturn the spatial findings.

20Independent population cross-check — WorldPop 2020

Headline numbers

Honest interpretation — this is more nuanced than a simple pass/fail

Per-ward correlation is moderate (r=0.46) and only 174/369 (47%) of wards fall within a 0.5×–2× agreement window. At first glance that looks like weak validation — but the disagreement pattern is informative, not random.

The disagreement is concentrated — and it tells a story

Top 10 wards where Kontur >> WorldPop (rapid growth signature)

These wards — Anjanapura, Horamavu, Dodda Bidarakallu, Andrahalli, Jakkur — are all in our top-deficit lists and they're exactly the outer-corridor growth wards the gap model flags as infrastructure-starved. The fact that WorldPop misses their 2022 population is itself evidence that this is where recent, rapid, unserviced population growth occurred.

What this cross-check actually proves

• City-wide totals are consistent. Kontur 12.5M vs WorldPop 9.7M (ratio 1.29) aligns with 2-year growth plus WorldPop's UN-adjustment conservatism.
• North and West corps are cleanly validated — ratios 0.95 and 1.01 imply near-perfect per-corp agreement on total population.
• East and South disagreement is explained — WorldPop's 2020 baseline predates the IT-corridor growth surge. Kontur (2022 H3) is the more current signal; WorldPop is the older lower bound.
• Where both agree per-ward (47% of wards, ratio within 2×), the correlation is high — meaning the gap model's population foundation is robust for the non-growth-corridor wards.

What this cross-check does NOT prove

• It does not resolve whether Kontur is overcounting East/South or WorldPop is undercounting. Ground truth is BBMP voter roll + Aadhaar records (not publicly available at 369-ward granularity).
• The 2-year gap (2020→2022) inherently limits how closely any two pop models can match in a growing city.

21Growth patterns 2000→2020 (GHSL time-series)

Headline: Bangalore's built-up area grew 59% in 20 years

The corporation-level gradient — growth where gap is worst

Growth was not uniform. The outer corps — especially East Corp — grew dramatically. Central Corp barely grew at all because it was already saturated. The exact pattern of our gap findings.

Top 12 fastest-growing wards (built-up ratio 2020/2000)

Wards with at least 0.1 km² baseline built-up in 2000, ranked by growth ratio. Several of these are in our top-deficit list — growth and under- provisioning go together.

Statistical cross-validation

Note on VIIRS nightlights

What this section proves

• Outer-ring deficit is a growth story. East grew 2.44× — the corps where infra is worst are exactly the corps where population/built-up grew fastest.
• Top-individual-ward growth (Andrahalli +528%, Horamavu +461%) matches our top-deficit list. Three independent pipelines (GHSL pixels, WorldPop underestimation in §20, and our gap model) now agree on which outer-corridor wards grew fastest and have the worst infrastructure.
• Central Corp's low gap score is earned structurally — it grew only 8% because it was already saturated by 2000, and inherited a century of infrastructure investment.

22Building triangulation — Overture vs OSM (independent pipelines)

Headline numbers

Per-ward correlation — the validation

Across all 369 wards, the two independent building counts correlate strongly: Pearson r = 0.801 for counts, r = 0.948 for total footprint area in km². The footprint-area correlation at r = 0.95 is exceptionally strong — it means the two pipelines agree almost perfectly on how much land is built on, per ward, even though they disagree on the exact count (because OSM traces large compound buildings as single polygons while Overture ML often splits them into many).

City-wide totals

Overture catches 1,100,344 buildings; OSM catches 701,722 — about 64% of Overture's count. The gap is consistent with what we know about OSM: its community traces branded and obvious buildings well, but informal settlements and recently-constructed low-rise buildings often get picked up first by ML (Overture) before volunteer mappers catch up. Overture's extra 398,622 buildings represent where the ML-derived layer sees structures that OSM volunteers haven't yet traced.

Corporation-level OSM / Overture coverage

OSM coverage ratio is fairly uniform across corporations (0.60–0.74) — no systematic bias. Central Corp has the best OSM coverage (0.74), probably because the central commercial/heritage buildings attract more volunteer mapping attention.

Where Overture and OSM most disagree

These are wards where Overture's ML picked up many more buildings than OSM has traced. They're all in rapidly-growing outer corridor wards — consistent with the pattern that ML catches new construction before OSM volunteers.

Why this matters for the gap model

• r = 0.80 on counts across 369 wards is very strong — two entirely different data-creation processes converge on the same per-ward building reality.
• r = 0.95 on footprint area — the total built-up area per ward is essentially triangulated. Any per-ward population estimate that relies on building data is now standing on a three-source foundation (Overture + OSM + ESA WorldCover pixels from §14).
• The systematic Overture > OSM gap (ratio 0.64) tells us OSM isn't yet caught up on informal / new construction in outer corps. This supports rather than contradicts the gap model's outer-ring-growth narrative.

Note on Microsoft Global Building Footprints

25Google Open Buildings — four-source building triangulation

Four-source headline

Triangulation matrix — remarkably strong agreement

r = 0.979 between Google and Overture is the strongest agreement between any two independent building-footprint sources we've tested. These two pipelines use completely different ML models trained on different satellite imagery (Maxar for Google, community contributions compiled into Meta Facebook footprints for Overture). Near-perfect agreement at the ward level validates the spatial foundation of every per-ward claim in this report.

City-wide totals — Google catches the most buildings

Google Open Buildings v3 detects 1,592,124 buildings in BLR — 1.45× Overture's count and 2.27× OSM's. This is because Google's ML was trained on high-resolution Maxar imagery from 2020-2022 and picks up small informal structures, rural buildings, and partial outlines that the other two sources miss.

Corporation-level four-way view

Google/Overture ratio is remarkably uniform at 1.40–1.50× across every corporation — indicating no systematic geographic bias. Google sees ~45% more buildings than Overture everywhere, not just in specific wards.

Where the gap is biggest — the growth-corridor signature again

Wards where Google detects the most extra buildings vs Overture — Thanisandra, Kogilu, Talagattapura, Anjanapura, Sampigehalli, Jakkur — are the same North/South outer-corridor wards that §20 showed as WorldPop-undercounted and §21 showed as fastest-growing 2000-2020. Four independent sources continue to converge on the same growth-and-deficit pattern.

What this section adds to report credibility

• Google OB is Tier A (rock-solid) data. Published globally by Google Research, peer-reviewed methodology, and the r=0.98 with Overture means the count is not an artefact of either pipeline.
• Four-source convergence. Overture + Google + OSM + ESA WorldCover pixels all now agree on per-ward building presence. A ward the gap model flags as "densely populated" is densely populated by four independent yardsticks.
• The footprint correlation (r=0.975) means aggregate built-up area is essentially a triangulated quantity. Google says 175 km² of buildings; Overture says 35 km²; OSM says 120 km² — the absolute totals differ because of different polygon-traceing granularity, but per-ward ratios match across all three.

26Fresh auxiliary data — BWSSB infrastructure + Foursquare POIs

A. BWSSB water infrastructure (November 2025 data from OpenCity)

Borewells mapped across BLR — 7,433 BWSSB municipal borewells

14,876 municipal borewells (dug by BBMP, handed over to BWSSB) pulled as KML from OpenCity, spatial-joined to 369 wards. 7,433 fell inside ward boundaries; the rest were outside municipal limits or in boundary-edge cases.

Important interpretation: these are BWSSB-managed supply borewells, not private wells. More borewells → more public water infrastructure, not more water stress. The Pearson correlation with Census 2011 piped-water coverage is r = +0.179 (weak positive) — confirming that BWSSB borewells supplement, not substitute, Cauvery supply.

Cauvery water supply stage coverage

BWSSB's Cauvery pipeline has been built in 5 phases. Each phase extended supply to new localities. We fuzzy-matched the 1,220 stage-assigned locality names to ward names and identified 146/369 wards with at least one documented Cauvery stage connection (remaining wards either weren't matched by locality name or aren't yet on the Cauvery network).

Corporation-level water infrastructure table

West Corp has the most borewells (3,098) because it has the largest number of outer-fringe wards needing groundwater supplementation. Central Corp has the fewest (11.8 avg per ward) because it's fully on the Cauvery pipe network.

Top-10 wards by BWSSB borewell count (infrastructure concentration)

IISc/BWSSB Groundwater Outlook April 2025

Additionally pulled: iisc_groundwater_apr2025.pdf (697 KB) — Interim Report II (April 2025 projections) by Prof. Lakshminarayana and Prof. Sekhar Muddu (IISc) for BWSSB. This is the most recent authoritative water-supply projection for Bangalore, covering groundwater dynamics and future shortfall scenarios. Cited in the water-gap narrative but not numerically integrated yet (PDF extraction would take a separate pass).

B. Foursquare Open Source Places (Nov 2024 global release)

Headline

r = 0.916 per-ward correlation with Overture validates that both crowdsourced/hybrid POI pipelines are converging on the same ward-level point density. 40,304 places (36% of FSQ BLR) were refreshed in 2024 — that's the freshness advantage Foursquare brings.

FSQ coverage by corporation

Central Corp has the highest FSQ/Overture ratio (0.84) — FSQ's user-checkin-driven data captures CBD commercial activity densely. Outer corps (East, North) have ~0.52 ratio, meaning Overture catches ~2× more places there — FSQ's coverage bias is toward already-commercially-dense areas.

Top FSQ categories in BLR

FSQ's hierarchical categories give us rich semantic labels vs Overture's flatter taxonomy:

27VIIRS nightlights — the real-time activity pulse

Headline pattern — nightlights don't map to infrastructure gap

Corporation-level brightness + growth

South Corp is brightening fastest at +8.2%/year — outer-southern growth corridor lighting up. West is the brightest already (80.9 nW) but slowest-growing (+2.2%) — it's saturated. Central Corp's modest growth (+5.7%) reflects already-mature commercial density.

Fastest-brightening wards — the IT corridor boom in real time

Seven of the top-10 fastest-brightening wards are in East Corporation — Kadugodi, Varthur, Panathur, Belathur, Channasandra, S.M Krishna Ward, Manjunatha Nagar (North). These are all IT-corridor wards near the Whitefield/Kadugodi tech cluster. Brightening +25-31%/year is explosive commercial growth. This is the same growth corridor §20 showed as WorldPop-undercounted and §21 showed as 2.44× GHSL built-up growth 2000-2020 — VIIRS confirms it's still accelerating in 2022-2023.

Top 5 brightest vs darkest (2023)

What this adds to the evidence stack

• Real-time activity confirmation: VIIRS 2022-2023 shows East Corp's IT corridor is actively brightening at 25-31%/year. This is the third temporal signal (after WorldPop pop-undercount and GHSL built-up growth) confirming the outer-corridor growth thesis.
• Infrastructure gap ≠ activity level: the weak r=+0.19 with gap score validates that our model is measuring infrastructure provisioning, not just commercial presence. Brightness and infra adequacy are orthogonal — bright wards can still lack schools/hospitals/water.
• Tier A data now (directly from NASA, atmospherically corrected): this upgrades the "commercial activity" evidence from our prior POI-based proxy (flagged as heuristic in §19) to direct satellite-measured radiance.

23Reliability scorecard — what to trust, with what confidence

The four reliability tiers

Specific claim ratings

Anti-bias disclosure

The model and validation pipeline were built without fitting to any specific political narrative. The independent satellite/OSM/WorldPop sources used for validation were not chosen post-hoc to confirm the gap model — they're standard global datasets that any analyst would reach for. The convergence of five sources on the same spatial story is therefore meaningful evidence, not selective citation.

What would strengthen the report from B/C → A

• Water/sanitation: integrate Jal Jeevan Mission live ward-coverage data → Census 2011 → Tier A (single-issue)
• Fire/police: RTI for current BBMP fire station + Karnataka police station list → Tier A
• Hospital quality: link to National Health Facility Registry (HFR) → adds operational status
• School quality: UDISE+ 2023-24 enrolment + teacher data → Tier B for service quality
• Transit reliability: BMTC AFCS ridership + on-time performance via RTI → Tier B for service

24Presenting this to government — playbook by audience

The five government audiences in the GBA structure