Does an hour of transit measure Jakarta's inequality?

1 The claim under test

The case computes, for each of Jabodetabek's 1,511 kelurahan and desa, the share of the region's job-dense floorspace reachable door-to-door within 60 minutes on scheduled public transport, and builds an equity argument on the distribution of that number: a Gini of 0.743, a Palma of 109.5×, and a core-to-periphery gap of 88×.

Three questions follow, and they are not the same question. Is the routing right? Is the measure informative? And does the measure support the word "equity"? The first is largely engineering and the case handles it well. The second and third are measurement theory, and that is where this review spends its time.

A cumulative-opportunity measure is a count of what falls inside a boundary. Every property it has — including its inequality — is jointly a property of the city and of the boundary. Publishing the number without the boundary's sensitivity is publishing half the result.

2 Prior art, and what can and cannot be compared

Pereira (2019) is the decisive methodological prior. Running a BRT expansion in Rio de Janeiro through cumulative-opportunity measures at four time thresholds, he found the policy verdict itself changed with the threshold: mean accessibility gains of 13.3% at 30 minutes and 11.3% at 60 fell to 4.2% and 2.4% at 90 and 120, turning a progressive-looking project into a modest, evenly-spread one. He names this the modifiable temporal unit problem. Boisjoly & El-Geneidy (2017), auditing 32 metropolitan transport plans, found practice thresholds cluster between 30 and 60 minutes with no principled basis for the choice.

Wessel & Farber (2019) put a number on the cost of routing a timetable rather than a bus. Rebuilding routable networks from archived vehicle-location data for four North American agencies, they report that schedule-based accessibility measures "may overestimate net accessibility on average by about 5%–15% percent or more", and that the error is spatially patterned rather than random. Those four agencies are among the world's better-adhering; for Jakarta's mixed traffic the range is a floor, not an estimate.

Almost nothing published reports the quantity this case reports. The University of Minnesota's Access Across America series publishes absolute counts, not shares — New York's average worker reaches 1,156,929 jobs within 60 minutes by transit and Atlanta's 59,826, a nineteen-fold spread inside one country, on a quantity that is not a percentage of anything. Brazil's Access to Opportunities Project likewise publishes counts. And its own authors warn against exactly the comparison a percentage invites: IPEA's technical note states the cumulative indicator "is not appropriate for comparing accessibility levels between cities of very different sizes", and Wu et al. (2021), across 117 cities, show job access scales sublinearly with metropolitan population — so enlarging the study area mechanically lowers the share.

One published set does report our exact quantity. Peralta-Quiros, Kerzhner & Avner (2019) measure, for eleven African cities, "the percentage of estimated employment opportunities throughout the city accessible by the average individual within 60 minutes" by public transport, population-weighted, over a 07:00–09:00 window — and publish a Gini of the same measure. Six of the eleven use an "Employment Opportunity Area" grid scored from mapped establishments rather than payroll counts, which is conceptually close to a floorspace proxy. §5 uses it.

On the equity side, other published accessibility Ginis measure different constructs. Guzman, Oviedo & Rivera (2017) report 0.348 for Bogotá, but on a gravity measure with continuous decay; Delbosc & Currie (2011) report 0.68 for Melbourne, but of transit supply. A binary threshold manufactures exact zeros a gravity measure never produces, so a cumulative Gini will normally exceed a gravity Gini on the same city. That is testable here, because the pipeline also publishes an exponential-decay measure with half-weight at 45 minutes: on that, Jabodetabek gives 0.723.

What we could not obtain. Andani et al. (2025) is the closest published comparator — a district-level cumulative-opportunity job-accessibility study of Jakarta across car, motorcycle and public transport, reporting that motorcycles deliver the broadest and most equitably distributed access and public transport the most severe inequality. It sits behind a paywall we could not pass, so we cite its qualitative conclusions and quote no numbers from it.

3 What was actually computed

Routing. r5py/R5 over the OpenStreetMap street and footpath graph plus three GTFS feeds: the official TransJakarta feed (240 routes including 98 Mikrotrans, 8,091 stops), a hand-encoded frequency feed for 9 rail lines, and a community angkot feed for Bogor (24 routes). Departures every minute across 07:00–09:00 on 2026-09-02, p50 over the window. Travel time is door-to-door: access walk, wait, in-vehicle, transfer and egress, with a 30-minute walking budget in every scenario and a 90-minute cap. Waiting and transfers are therefore in; ojek, becak and unscheduled paratransit are out. That is the right architecture, and the first thing a reviewer should check.

Opportunity. A 1000 m lattice of 5,766 cells carrying GHS-BUILT-S NRES non-residential built surface as the jobs proxy, capturing 91.8% of the region's proxy and 92.7% of its population. Destinations are routed at each cell's population-weighted centre of mass rather than its geometric centre — a good decision that removed a large earlier artefact.

Denominator. The denominator is the whole region's floorspace, and the region is not the city.

Checks, as published. 1 of 3 hard checks passed. The timetable check failed 1 of 4 legs; network integrity failed on 10 unroutable mainland origins; the external routing comparison was never evaluated ("requires live Google Routes API calls; the user has not authorised them, so no external routing comparison was made."). The case publishes all of this unretouched, which is the correct behaviour and the reason the rest of this review can be specific.

4 Finding one — the hour is the finding

The pipeline computes access at 30, 45, 60 minutes. The equity chapter used only the last. Recomputing the entire distributional package at each threshold shows how much of the headline belongs to the clock.

The Gini moves from 0.655 to 0.743. The Palma multiplies by 5.0. And the statistic the case leads with — the ratio of the DKI median to the Bodetabek median — moves from 14× to 88×, a factor of 6.5. The headline "a gap of 88×" is, on the case's own data, equally a headline of "14×".

The direction is worth dwelling on, because it is counter-intuitive and it is not what the Rio study found. Lengthening the budget makes the distribution more unequal here. The mechanism is compounding: fifteen extra minutes given to someone already standing on a corridor buys another interchange and another wedge of the city, while the same fifteen minutes given to someone with a two-kilometre walk to a thirty-minute-headway feeder buys nothing. Access to a network is convex in time for those on it and flat for those off it.

That is a real result about Jabodetabek, and it is more interesting than the headline it undermines: the region's transport disadvantage is not a gradient, it is a threshold. You are on the network or you are not, and time only helps the former.

The gravity measure the pipeline already computes — no cut-off at all — gives a Gini of 0.723, between the 45- and 60-minute cumulative values. Its rank correlation with the 60-minute measure is 0.997, and the 30- and 60-minute rankings correlate at 0.970. The levels move enormously; the ordering barely moves at all. That distinction runs through every finding below.

5 Finding two — and so is the frame

The second half of the problem is spatial rather than temporal. The measure asks what share of a 6,858 km² region's floorspace a resident can reach. Every published benchmark asks what share of a city's. Jabodetabek as drawn here is 2.8 times the area of Cape Town's study region and 8.3 times Nairobi's, and 62% of its floorspace is outside DKI.

So we re-based the measure without re-routing anything: DKI residents, DKI destinations, DKI's own floorspace as the denominator, on the identical published travel-time matrix.

On the city frame DKI's population-weighted mean access is 11.5%, not 1.5%, and its Gini is 0.350, not 0.743. That places Jakarta between Cape Town (6.5%) and Dar es Salaam (12.2%) on level, and more equal than 8 of the 11 on distribution. Measured as a city, Jakarta is unremarkable. Measured as a 6,858 km² region it falls below all 11 of them, Cape Town included — the most spatially mismatched large city in the set, at 6.5% and a Gini of 0.63.

Both numbers are correct. They answer different questions, and the case answers only the second while writing headlines that read as the first. The regional framing is the right one for a Jabodetabek planning authority — the periphery's disconnection from the region's own industrial floorspace is exactly the policy problem — but it must be labelled as a regional statistic, with the study area stated, every time it is quoted.

Where this comparison is weak. The African cities' opportunity variable is an establishment-scored grid or, for two cities, a census of employees; ours is built surface. Their study areas are drawn on access patterns; ours on administrative boundaries. Their transport data are largely mapped paratransit — which is precisely what Jabodetabek's periphery lacks in this build. Every one of those differences would tend to make Jakarta look worse than a fair comparison would, so read Figure 4 as an order of magnitude and a warning about framing, not as a league table.

6 Finding three — a compass reproduces most of the map

Before crediting a routing engine, ask what a ruler would have said. We built the crudest possible alternative: for each kelurahan, the share of the region's job-dense floorspace lying within 10 km as the crow flies — no streets, no timetables, no transfers — plus straight-line distance to the region's floorspace-weighted centre of mass.

Circle and distance together explain R² 0.604 of the log of routed access across the 1,081 kelurahan that reach anything, and the rank correlation between circle and routed measure across all 1,511 units is 0.819. Distance to the centre alone gives 0.519; the 5 km circle alone, 0.534.

This is not a demolition — two-fifths of the variance is genuinely network. But it does mean the choropleth is not the product. A map shaded by routed access looks, to the eye, almost identical to a map shaded by distance from the middle of Jakarta, and a reader who takes the headline from the picture has learned something they already knew.

7 Finding four — the residual is the product

If the circle predicts most of the map, the places where it fails are where the analysis earns its keep. We asked a deliberately blunt question: which kelurahan have more than the median amount of the region's floorspace within 10 km of them and still reach none of it inside an hour?

There are 34 such kelurahan, holding 739,629 people. Bitung Jaya in Cikupa has 13.4% of the whole region's job-dense floorspace within 10 km of it and reaches none of it. These are not remote villages; they are the dormitory desa of Jabodetabek's manufacturing belt, and the finding is that the region's industrial workforce lives inside a labour market it cannot reach on scheduled transport.

The mirror image is equally diagnostic, once a handful of remote desa are set aside — Cikasungka, Cisarua and Batok top the residual list only because their local circle is so close to nothing that reaching anything at all beats the prediction. The substantive over-performers are the next four: Pabaton, Paledang, Empang and Gudang, all in the centre of Kota Bogor, which reach roughly 39 times what their own neighbourhood would predict. That is the KRL Bogor line, and it is what a corridor is for. A ranked list of both residuals is a project pipeline; the shaded map is a poster.

One of these has a mundane cause. Bitung Jaya is also on the case's own list of 10 mainland origins whose population-weighted centre snaps to an isolated street fragment, so its zero is a graph defect rather than a service fact. The other 33 route normally. The case is right to publish the unroutable list; it should also cross it against this one, so a reader can tell a broken graph from a broken bus network.

8 Finding five — the system is relatively regressive

The case prices each layer by what it adds to the average resident: rail 0.26%, the whole public-transport system 1.32% over walking. A mean says how much opportunity a network creates. It says nothing about who receives it.

94% of everything public transport adds over walking accrues to the 30% of the region's people who live inside DKI Jakarta, and 54% to the best-served tenth. For the rail layer the figures are 89% and 80%. The mean DKI resident gains 4.14% from the system; the mean Bodetabek resident gains 0.106% — a factor of 39.

The Gini follows: 0.629 on foot, 0.737 with buses, 0.743 with everything. Every layer of Jabodetabek's public transport network raises access and raises inequality at the same time.

Three things this does not mean. First, nobody loses: the case's own monotonicity check confirms access never falls when a layer is added, so this is relative regressivity in a rising tide, not harm. Second, a rising Gini alongside a rising mean is a property of any concentrated improvement and is not by itself an indictment. Third, and most important, the periphery's paratransit is almost entirely missing from the data — one community angkot feed exists — so the Bodetabek baseline is a floor and the measured concentration is an upper bound. State all three or state none.

9 Finding six — the equity axis, computed

The case is titled "Transit Access & Urban Equity" and reported its poverty cross-link as pending, because the poverty case it depends on had not published when it ran. It has since. We ran the link.

Two things were wrong beyond the ordering. The cross-link code searched for a column whose name contained both "adm3" and "code"; the estimates publish that key as pcode, so the join would have reported an unrecognised schema even after the dependency was met. And the column search would have selected the official regency headcount broadcast downward rather than the modelled small-area estimate. Both are fixed in the case's pipeline, and the axis is now computed on the 2025 vintage across 187 kecamatan.

The rank correlation between access and poverty is -0.509 and the population-weighted concentration index of access over the poverty ranking is +0.184 — access accrues to the less poor. The blue line is the cleaner statement: the share of kelurahan reaching nothing measurable rises monotonically from 1.6% in the least-poor fifth to 56.7% in the poorest, a factor of 35. 170 kelurahan — 1,057,344 people, 2.9% of the region — fall in both the worst access quintile and the worst poverty quintile.

Read the bars carefully; they are not monotone. Median access falls from 0.98% in Q1 to 0.071% in Q3 and then rises again, and the population-weighted mean is non-monotone too. The reason is real rather than statistical: Jakarta's inner-city kampung are both poor and extremely well connected, so the poorest quintile holds two populations — the connected urban poor and the disconnected peripheral poor — and averaging across them hides the thing that matters. The zero-access share separates them cleanly, which is why it, and not the mean, is the equity statistic to publish. Note also that poverty is constant within a kecamatan (187 distinct values across 1,511 units), so within-kecamatan variation in access is orthogonal to it by construction.

10 Finding seven — a footprint is not a floor area

The jobs proxy is GHS-BUILT-S NRES: non-residential built surface, in square metres of ground covered. A single-storey warehouse in Cikarang and a forty-storey tower on Sudirman contribute in proportion to their footprints. Every tall building in Jakarta is counted at a fraction of the employment it holds, and every shed on the periphery at a multiple.

This is cheap to test without re-routing. The same GHSL family publishes GHS-BUILT-V NRES — building volume, same 100 m grid, same epoch, same licence. We downloaded the matching tile, re-weighted the identical published travel-time matrix, and recomputed the measure.

The DKI median rises from 4.29% to 5.28% — 23% higher — the Bodetabek median falls from 0.049% to 0.036%, and the core-to-periphery gap widens from 88× to 147×. The Gini rises from 0.743 to 0.781 and the Palma from 110× to 223×.

Two conclusions, pointing opposite ways. The published levels understate concentration — the footprint proxy flatters the periphery by counting its sheds at full weight. And the rank correlation between the two measures is 0.9988: not a single kelurahan changes place meaningfully. Volume is the better default and the case should switch to it, but nothing already published about the ordering is wrong.

11 Finding eight — the timetable runs fast, and reality runs slower

No operator publishes a GTFS feed for KRL, MRT or either LRT, so the case hand-encodes them from published headways and end-to-end journey times, distributing run time along the inter-station geometry at each line's average commercial speed. It flags a ±15 % caveat and tests three lines against their operators' own figures. All three pass. But all three deviate in the same direction.

The mean optimism is 6.8% and the mean chord shortfall is 5.7%. For the MRT the two are 8.0% against 8.9%, and for the LRT 5.8% against 5.6% — close enough that the geometry is plainly the mechanism. For the KRL Bogor line the chord explains only 2.5% of a 6.5% gap, so something else is fast there too. A single scale factor per line, calibrated to close the published gap, would remove this in an afternoon; a ±15 % caveat does not, because the error is a bias, not a spread.

The bus side of the same check fails, and its framing deserves a second look. The official TransJakarta feed contains no end-to-end Corridor 1 trip, so the check compares the feed's scheduled commercial speed — 11.2 km/h over 13.83 km of straight-line inter-stop distance — against brtdata.org's published 19 km/h corridor average, and concludes the routed bus times are "conservative, not optimistic". Two caveats belong with that. Straight-line distance understates the real corridor by roughly a tenth, so the true scheduled speed is nearer 12 than 11.2. And the 19 km/h benchmark carries a 2013 date on brtdata.org — a thirteen-year-old grey-literature figure being used to fail a 2026 feed.

More fundamentally, Wessel & Farber (2019) show that any schedule-based accessibility measure overstates by 5%–15% even when the schedule is honest, because real headways vary and variance lengthens average waiting time. That correction applies on top of everything above; it bites hardest on the high-frequency corridors where this case's access is concentrated; and their four North American agencies are a best case. The case states "scheduled, not congested" clearly and repeatedly. What it does not state is that the error has a known sign and a known rough magnitude.

12 Finding nine — the check that passed does not test what it claims

Of 3 hard checks, one passed, and its most quotable component is a replication of ITDP's People Near Transit indicator. The build reports 98% of Jakarta's population and 42% of Greater Jakarta's within 1 km of service at a headway of 15 minutes or better, against ITDP's 2016 anchors of 44% and 16%, and notes that it "should read high, and does".

We read ITDP's report. Its Table 2 gives Jakarta City 44% (4,410,442 of 9,991,788) and Jakarta Metro 16% (4,410,442 of 28,019,545) — the same numerator, because, as ITDP states, the rapid-transit network did not extend past the city borders. Its definition of rapid transit is BRT, LRT or metro meeting the BRT Standard's Basics, at headways under 20 minutes between 06:00 and 22:00. In Jakarta in 2016 that meant TransJakarta's BRT corridors and nothing else: KRL Commuter Line was not counted.

The build's replication counts every stop in the feed at a 15-minute headway or better — 7,517 of them, Mikrotrans and commuter rail included. That is a defensible measure of service coverage. It is not ITDP's measure, and the difference between 16% and 42% for Greater Jakarta is mostly the decision to count KRL, not eight years of construction. Two further problems compound it: at 98% of Jakarta the statistic has no room left to discriminate — almost any network would clear it — and the anchors are a decade old, which makes "above the anchor" a statement about 2016 rather than about this model.

A check that cannot fail is not a check. This one should be relabelled as what it is — a coverage statistic — and either dropped from the check list or rebuilt to ITDP's actual definition, where it would be a genuine and interesting test.

13 Finding ten — modifiable units, and an edge that turns out not to matter

Kwan & Weber (2008) state the general problem plainly: "MAUP will be a problem for any zone-based accessibility measure" — though their own empirical test, on individual space-time measures, found little variation across zone schemes. Here the effect is large, and easy to measure because the P-codes nest.

Aggregating from 1,511 kelurahan to 14 kabupaten and kota moves the Gini from 0.743 to 0.660 and the Palma from 109.5× to 12.6× — a factor of 8.7 on the Palma, from redrawing lines on a map. The units are also wildly unequal in size, from 0.148 km² to 61.5 km², a 415-fold range, and access correlates -0.48 with area. Kelurahan is the right unit for this question — it is where people and services actually are — but the Palma in particular should never be quoted without it.

The edge effect, by contrast, is a null. The study clips at a bounding box, so opportunities just outside it are invisible: Karawang's industrial estates east of Cikarang, Serang to the west. We scanned access against distance from the boundary. Only 45 kelurahan — 0.8% of the region's population — lie within 10 km of the edge. Truncation is not a material driver of the zero mass, and we say so rather than leaving it as an unquantified worry.

One publication defect, now fixed. The per-kelurahan file the map reads rounded access to four decimal places. The population-weighted median of that measure is 0.0021, so the rounding step was 5% of the median value, and 217 kelurahan holding 3,289,289 people (8.9% of the region) were published as exactly zero when they were not. The map painted them the same colour as the 430 genuine zeros. The export now writes six decimals.

14 What follows for decisions

Evidence is worth gathering only if it changes an action. Read strictly, this instrument supports four uses and forbids a fifth.

1. A corridor pipeline, for Dishub DKI and the Jabodetabek transport authority. The 34 kelurahan with opportunity next door and no way to reach it — 739,629 people, concentrated in the Bekasi and Tangerang industrial fringe — are a ranked, named, defensible list of feeder-route candidates. This is the highest-value output in the whole case and it is currently invisible on the page.
2. A targeting layer for TransJakarta's Mikrotrans expansion. The residuals in §7 identify precisely where a low-cost feeder converts nearby floorspace into reachable floorspace. Kota Bogor's centre shows the upside when it works.
3. A before-and-after frame for KAI Commuter and MRT Jakarta. The scenario machinery already prices the rail layer; §8 shows its incidence. Any new line can be scored the same way — and scored on who gains rather than on the mean.
4. An equity screen for Bappenas. §9's double-disadvantage list — 170 kelurahan, 1,057,344 people, poor and disconnected together — is exactly the intersection a national planning agency needs and rarely has, and it exists only because two Demo Lab cases were joined.
5. Not: a cardinal statement of the size of the gap. "88×" is 14× at a shorter budget, 147× on a better proxy, and unquotable without the reporting unit and the study area. Publish the ordering and the named lists; do not publish the multiplier as though it were measured.

The same discipline applies to the hospital figure. 38% of the region cannot reach a hospital in an hour on scheduled public transport — but Jabodetabek moves on motorcycles, and Andani et al. (2025) find motorcycle access in Jakarta both broader and more equally distributed than transit access. The statistic is a real indictment of the transport network. It is not a statement about healthcare access, and the page has been corrected to say which one it is.

15 What remains open

1. Route the motorcycle. Every equity conclusion here is conditional on a mode that carries a minority of the region's trips. A second matrix on the same graph, with ojek speeds and no timetable, would turn "who can reach work by bus" into "who can reach work", and would test the case's central claim directly.
2. Validate against realised times. The external routing comparison was never run (requires live Google Routes API calls; the user has not authorised them, so no external routing comparison was made.). Any archived vehicle-location or probe feed would let the 5%–15% schedule optimism be measured for Jakarta rather than imported from Toronto and Boston.
3. Calibrate the hand-encoded rail per line to close the published gap. §11 shows the error is a bias with a geometric cause, not a spread. One scale factor per line removes most of it, and the ±15 % caveat can then mean what it says.
4. Obtain Andani et al. (2025) and compare like for like. It is the only published cumulative-opportunity accessibility study of Jakarta we could find, it appears to use a different threshold again, and it is the one document that could turn this case's headline from unbenchmarkable into benchmarked.
5. Publish the measure at every threshold and both frames by default. Not as a robustness appendix — as the result. §4 and §5 are not caveats about the finding; they are the finding.

16 References and reproducibility

1. Pereira, R.H.M. (2019). Future accessibility impacts of transport policy scenarios: Equity and sensitivity to travel time thresholds for Bus Rapid Transit expansion in Rio de Janeiro. Journal of Transport Geography 74, 321–332. doi:10.1016/j.jtrangeo.2018.12.005
2. Peralta-Quiros, T., Kerzhner, T. & Avner, P. (2019). Exploring Accessibility to Employment Opportunities in African Cities: A First Benchmark. World Bank Policy Research Working Paper 8971. doi:10.1596/1813-9450-8971 — Tables 1 and 3, pp. 9–11.
3. Wessel, N. & Farber, S. (2019). On the accuracy of schedule-based GTFS for measuring accessibility. Journal of Transport and Land Use 12(1), 475–500. doi:10.5198/jtlu.2019.1502
4. Wu, H., Avner, P., Boisjoly, G., Braga, C.K.V., El-Geneidy, A., Huang, J., Kerzhner, T., Murphy, B., Niedzielski, M.A., Pereira, R.H.M., Pritchard, J.P., Stewart, A., Wang, J. & Levinson, D. (2021). Urban access across the globe: an international comparison of different transport modes. npj Urban Sustainability 1, 16. doi:10.1038/s42949-021-00020-2
5. Owen, A., Liu, S.S. & Lind, E.M. (2025). Access Across America: Transit 2024. Report CTS 25-18, Accessibility Observatory, University of Minnesota. Dataset doi:10.13020/gd60-km49. Table 2, p. 3.
6. Guzman, L.A., Oviedo, D. & Rivera, C. (2017). Assessing equity in transport accessibility to work and study: The Bogotá region. Journal of Transport Geography 58, 236–246. doi:10.1016/j.jtrangeo.2016.12.016
7. Delbosc, A. & Currie, G. (2011). Using Lorenz curves to assess public transport equity. Journal of Transport Geography 19(6), 1252–1259. doi:10.1016/j.jtrangeo.2011.02.008
8. Boisjoly, G. & El-Geneidy, A. (2017). How to get there? A critical assessment of accessibility objectives and indicators in metropolitan transportation plans. Transport Policy 55, 38–50. doi:10.1016/j.tranpol.2016.12.011
9. Kwan, M-P. & Weber, J. (2008). Scale and accessibility: Implications for the analysis of land use–travel interaction. Applied Geography 28(2), 110–123. doi:10.1016/j.apgeog.2007.07.002
10. Andani, I.G.A., Qamilla, N., Izdihar, R.P., Safira, M., Sakti, A. & Syabri, I. (2025). Spatial, mobility, or socio-economic inequity? A district level job accessibility evaluation in Jakarta, Indonesia. International Journal of Urban Sciences. doi:10.1080/12265934.2025.2553714 — cited qualitatively; full text not obtained.
11. Institute for Transportation and Development Policy (2016). People Near Transit: Improving Accessibility and Rapid Transit Coverage in Large Cities. New York: ITDP. Table 2, p. 11. Grey literature, no DOI.
12. Pereira, R.H.M. et al. (2020). Desigualdades socioespaciais de acesso a oportunidades nas cidades brasileiras. Texto para Discussão 2535, IPEA, Brasília, p. 29. Grey literature; no resolvable DOI.
13. 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(1), 45–53. doi:10.3141/2653-06 — the R5 engine used here through r5py.
14. European Commission JRC (2023). GHSL Global Human Settlement Layer R2023A — GHS-BUILT-S NRES and GHS-BUILT-V NRES, epoch 2020, 100 m Mollweide; GHS-POP epoch 2025. CC BY 4.0.
15. BRT+ Centre of Excellence & ITDP. Global BRTData, Jakarta entry (commercial speed 19 km/h, labelled 2013). Grey literature.

The only hand-entered values here are the literature constants in §2 and Figures 4 and 11, each transcribed from the published table named beside it. All eleven review tests were specified before they were run and all eleven are reported, including the two that returned nulls — edge effects, and the invariance of the ranking under the surface-to-volume swap. Data vintage 2026-08-30; routed 2026-09-02 07:00–09:00, p50 over the window.