Your Walkability Score Is Wrong: New Tool Lets Residents Redefine Their City

Every walkability score you have ever seen about your neighbourhood may be quietly lying to you. That is the provocative implication of new research from Bengaluru, where computer scientists have built and deployed an interactive framework that lets ordinary residents — not analysts, not vendors, not planning departments — define what accessibility means to them and watch a city-wide map redraw itself in real time. The study, published in Discover Cities, was tested across all 198 wards of Bengaluru, India’s third-largest metropolitan area, and its central finding is startling: who is asking the question changes the answer far more than the mathematics used to compute it.

The problem with existing tools is deceptively simple. Walk Score, the most widely used public walkability index, assigns a single number to every location, computed from fixed amenity categories and secret, patented weights calibrated on North American service landscapes. Everyone sharing a location receives the same value, regardless of whether they are a parent walking children to school, a night-shift worker, or a retiree managing a tight budget. The researchers point out that this one-size-fits-all assumption is least defensible precisely where populations are most mixed — and Indian cities are defined by extreme density, socioeconomic heterogeneity and informal urban growth, with households of vastly different incomes routinely co-located in the same neighbourhood.

Consider what a fixed category definition actually measures in Bengaluru. An index whose grocery category is defined by supermarkets captures something very different in a city where daily provisioning runs through neighbourhood kirana stores. The researchers cite evidence from Chile, where Walk Score omitted locally important destinations such as street markets and overweighted eating and drinking out relative to local travel patterns. In Bengaluru the stakes are higher still: roughly thirty per cent of roads carry pedestrian footpaths, and pedestrians accounted for forty-four per cent of road accidents in the city. A tool calibrated where pedestrian provision is continuous does not transfer unexamined to a setting where the walk itself is contested ground.

Academic accessibility research has known for decades that accessibility varies between individuals at the same location, and has developed sophisticated preference- and person-sensitive measures to capture this. But those measures run as analyst-operated GIS workflows: assembling network data, routing, intersecting, aggregating — and changing a single weight means running everything again. Public tools offer the opposite trade: immediacy with almost no preference expression. The new framework supplies the missing delivery, allowing preference specification at urban scale and at interactive speed. Users state which amenity categories matter and how much, which are non-negotiable, which to avoid, which substitute for one another, and how far they will walk for each. The map updates accordingly.

The technical trick that makes this possible is elegant. The framework replaces the continuous distance-decay functions used in established measures such as the WalkTHERE index with a rectangular impedance — an amenity is either reachable within a user-set walking-time threshold or it is not. Under that simplification, the entire accessibility summation collapses to a count of amenities within the threshold, and counts can be precomputed once by shortest-path routing. The team built a pedestrian network of 619,696 nodes and 700,377 edges covering 23,367 kilometres, routing independently at every threshold from three to fifteen minutes. At query time, only arithmetic on stored integers remains, so a reconfiguration costs a single lookup and a few operations per category. The score stays bounded between zero and one, and each category’s contribution remains individually readable.

Each control does precise mathematical work. Category selection determines which terms enter the sums. A saturating exponential decay, parameterised by an intuitive half-life, encodes diminishing returns so the tenth nearby café adds less than the first. Grouping asserts functional substitutability: combining restaurants, fast food and cafés into one dining group means any one of them satisfies the need, with counts summed before decay. The framework’s sharpest innovation is its pair of non-compensatory constraints. Marking a category Required sets the score to zero wherever it is unreachable, however excellent everything else is; marking one Never does the same in reverse wherever it is present. No weight in a weighted linear combination can reproduce either, because raising a weight cannot zero out a location on the strength of one absent amenity.

The empirical results quantify why this matters. The researchers built four resident profiles — a worker, a non-worker, a school-run household and a social-recreational profile — with thresholds derived from a 2,400-household Bengaluru travel survey reporting acceptable walking distances as short as 263 metres for personal errands, roughly three minutes rather than the fifteen the eponymous city concept assumes. Measured against a fixed-preference reference, the profiles agreed at correlations ranging from 0.93 down to 0.62, while the framework’s computational simplification agreed with the measure it extends at 0.973, and four standard measures given uniform weights agreed at 0.956 or better. In other words, configuration moves ward rankings by an order more than any choice of aggregation. The worker and non-worker surfaces agreed with each other at only 0.585.

The hard constraints produced the most dramatic divergences. A ward with excellent dining, banking and healthcare but no reachable school scores zero for the school-run household — behaviour no amount of other provision can compensate, and exactly what a user declaring a non-negotiable need should expect. The avoidance mechanism revealed something subtler: the same bars and pubs that topped the social-recreational profile’s map were the ones penalised in the non-worker profile. The ward of Domlur was simultaneously the best-scoring location for one profile and the most heavily penalised for another. A measure that treats amenity presence as unambiguously good cannot express that tension, and any single averaged score for such wards answers one question while appearing to answer both.

The study is candid about its limits. OpenStreetMap amenity data, comprising 20,790 features across 50 categories, is markedly more complete for formal and institutional amenities than for the informal retail on which lower-income households rely. General and convenience stores, greengrocers and kiosks together number just 806 records across the metropolis — implausibly low for a city whose daily provisioning is dominated by small independent shops. A profile requiring daily essentials returned zero for 96.7 per cent of the city, so income-differentiated configurations remain unevaluable. The framework also treats every instance of a category as equivalent — a single-room clinic counts the same as a tertiary hospital — and measures proximity only, not footpath quality, shade or safety.

Those caveats frame who the tool is for and why its governance matters. The primary user is a resident making a location decision, where the preferences encoded are the user’s own and the consequences fall on them. Community organisations can characterise service gaps without commissioning technical analysis. But the consequential case is institutional: when planners configure the tool, the person setting preferences is not the person living with the result. A tool that asks where is good for people like me also makes it easy to ask which people an area is good for — a question with very different stakes when the answerer allocates resources. The researchers propose two safeguards: recording configurations alongside results so scores can be interrogated rather than merely cited, and never treating a personalised score as sole justification for an allocation decision. It is a rare piece of software research that ends by warning about its own success, and a fitting note for a tool whose deepest contribution is making visible that every accessibility score — averaged or personalised — encodes someone’s preferences. The difference is that now, at last, the someone can be you.

Subject of Research: Personalised interactive urban accessibility measurement in Bengaluru

Article Title: An interactive framework for personalised urban accessibility assessment

Article References: Ghuriki, P., & Chanti, S. (2026). An interactive framework for personalised urban accessibility assessment. Discover Cities, 3(1), Article 204. https://doi.org/10.1007/s44327-026-00384-z

Image Credits: AI Generated

DOI: 10.1007/s44327-026-00384-z

Keywords: urban accessibility, walkability, 15-minute city, Bengaluru, GIS, OpenStreetMap, personalised scoring, cumulative opportunity measures, multi-criteria decision analysis, pedestrian networks, urban planning, spatial analysis