How the opportunity index is built, everything that has changed
since it was published, what each change did to the numbers, and the exact
sources behind each state.
If you saved a screening result, check the changelog.
Scores move when the method changes, not only when the data does. A result you
forwarded last week may not reproduce today, and the entries below say which
changes moved published scores.
What the index does
For every cell in a hexagonal grid (H3 resolution 8 in built-up areas,
resolution 7 elsewhere) it estimates charging demand from public activity
proxies, subtracts distance-decayed existing charger capacity, and rescales the
result across the state to a 0–100 screening score. It identifies
areas worth investigating. It does not identify viable sites, and it has
never been shown to predict actual charging use.
Standing limitations
No state is validated
Only Telangana has ever been backtested against measured charging consumption, and it failed: no statistically significant correlation (rho=0.04, p=0.69) across 85 metered Hyderabad localities. No other state has been tested at all, because no metered series for them has been found.
Urban extent measures mapping, not settlement
Urban classification comes from OpenStreetMap land-use polygons, and mapping convention differs sharply between states. Andhra Pradesh tags whole revenue villages as residential and has 10,229 km2 of mapped built-up land; Kerala tags smaller parcels and has 479 km2 -- yet Kerala is the more urbanised state. No density threshold can fix this, because any threshold ranks states by how they were mapped.
Scores are relative within a state
Scores are rescaled inside each state, so some cell always approaches 100 and a 100 in one state is not equivalent to a 100 in another.
Private and captive chargers are invisible
No public source lists them, so real supply is higher than shown everywhere.
Sources by state
Sources are declared per state. Nothing here is generated by substituting a
state name into a shared sentence — that practice previously put
Telangana’s distribution utility on four other states’ pages.
TGSPDCL open data (Telangana State Southern Power Distribution Company), via the Telangana Open Data Portal
Backtested against metered electricity consumption for 85 Hyderabad localities: no statistically significant correlation with actual charging use (ρ=0.04, p=0.69).
Telangana Electric Vehicle & Energy Storage Policy 2020–2030
Common to every state: OpenStreetMap contributors (ODbL) for geography and
land use; Open Charge Map contributors (CC BY 4.0) for some station records;
VAHAN (Ministry of Road Transport & Highways) for vehicle registrations;
Ministry of Power / BEE charging-infrastructure guidelines, 17 September 2024.
Changelog
Scores changedall
Restricted and captive chargers no longer count as supply
The pipeline read each station's access type but never filtered on it, so gated and staff-only chargers were counted as public supply. A charger the public cannot use does not serve the demand this model estimates, and counting it made underserved areas look served. 116 stations were excluded across the five states (Karnataka 45, Tamil Nadu 27, Kerala 25, Telangana 18, Andhra Pradesh 1). They remain in the registry, graded and sourced; they are excluded only from the supply side of the model.
Effect: Supply fell slightly in affected cells, so their scores rose.
Scores changedall
Station data refreshed from the CPO tracker
501 stations added across the five states. The new records carry location but not charger power, so the imputed share rose in every state: Kerala 27% to 36%, Tamil Nadu 59% to 65%, Karnataka 48% to 54%, Andhra Pradesh 83% to 90%. More records is not the same as more evidence, and the confidence figures on each page moved accordingly.
Effect: Supply rose where stations were added; confidence fell.
Urban extent is derived from land-use polygons fetched over a rectangular bounding box, and those boxes overlap neighbouring states. Only rural cells had been clipped to the state boundary, so each state's grid contained cells that were not in it -- Kerala held 998 (4.4%), Tamil Nadu 10,017. Kerala's top-ranked truck zone was Irugur, Coimbatore, a Tamil Nadu locality ranked as a Kerala opportunity.
Effect: Cell counts fell; out-of-state areas no longer appear.
Scores changedkerala
Fixed a truncated data fetch that understated urban extent
One OpenStreetMap mirror answered valid queries with a well-formed empty response instead of an error, and the pipeline treated that as success. The same query returned 11,078 polygons from one mirror and 1,542 from another within the same minute. Kerala's published grid had been built through the faulty mirror. The mirror was removed, empty results now require confirmation from a second independent mirror, and partial responses (which OpenStreetMap signals with HTTP 200 and a remark field) are rejected.
Effect: Kerala's urban extent had been understated by roughly 10%. Affected cells were re-scored.
Scores changedall
Highway corridors are continuous across state borders
Corridors had been clipped to each state's boundary, so NH-44 was not one route from Telangana into Tamil Nadu but two unrelated stubs in two datasets. Freight routes do not stop at a border, and a corridor just outside a line still serves a cell just inside it.
Effect: Cells near state borders gained corridor adjacency, which raises truck-segment scores there.
Wording onlyall
Confidence figures are per state
Every state page quoted Telangana's 98% charger-power imputation, because the figure was hard-coded rather than read from each state's own data. Kerala's real figure at the time was 27%. The shipped data payload carried the same wrong number.
Effect: No score changed. The stated confidence was wrong before.
Wording onlyall
Evidence grading published
Each state now carries a coarse evidence grade computed the same way for all of them: better-evidenced at 40% imputation or less with 300+ stations; thin at 85% or more, or under 100 stations, or fewer than 5 stations per 1,000 cells. Applied uniformly, this grades Telangana -- the original build -- as thin.
Effect: No score changed. Weak datasets are now labelled as weak.
Full workings
Weights, formulas, the sensitivity analysis and the gap register are
published in the
detailed methodology notes.
Those were written against the Telangana build; the model they describe is the
one every state uses, but the validation results reported there are
Telangana’s alone.