Southern California Edison · meter deployment

Where the meters are, and what it costs to reach them.

SCE is putting ~5.7 million meters out to bid and published no geospatial data saying where they are. This reconstructs that basemap from fifteen county assessor layers, SCE's own grid geography, census geography and Overture roads — then scores how hard each meter is to reach, which is the question a bid actually turns on.

Interactive · runs entirely in the browser

Deployment cost model

Forty-one controls, each with a range and a provenance label. Move a rate, watch the programme total and the map move with it. Choropleth over six geographies, cost waterfall, schedule, and a sensitivity ranking that says which assumptions are worth arguing about.

Open the cost model →

Method

How it was built

Nine stages from thirty dated sources to one fact table, with the row counts and the decision behind each.

See the flow → For agents

AI access

Catalogue, MCP endpoint and queryable GeoParquet — point Claude or Grok straight at the data.

Connect →

4.81M estimated service points~84% of SCE's ~5.7M

272,354 line sections scoredthe unit SCE dispatches on

4,133,590 parcels in net territoryof 6,478,429 countywide

131,456 KM² net service territoryafter the utility carve-out

15 counties ingestedevery one publishes parcels

30 sources, dated and citedprovenance on every row

In summary

The deliverable is not a parcel count. SCE's solicitation will very likely supply meter counts by district. What it cannot supply — because nobody has built it — is a per-section answer to how hard is each meter to reach. That is what prices a rollout, and it is what this produces.

Four point eight million service points are placed on SCE's own line-section geography, each labelled measured, inferred or modelled. Every one is then scored on meter density per route kilometre, road access, fire-threat tier, multifamily share, structure vintage and disadvantaged-community share, into a composite index and a cost-per-meter multiplier.

At placeholder rates the cost model returns $1.61B, $338 per meter and 236,509 crew-days — 58 months at 250 crews, or 547 crews to finish inside 36. Those are not forecasts. Every rate behind them is a slider with a range and a label saying whether it came from a cost book, a benchmark, or judgement. The point of the tool is to find which of them move the bid by eight figures and which are noise.

The one finding worth leading with: 29.6% of SCE's meters sit in multifamily meter banks — behind a door somebody has to open. That is a scheduling and access problem rather than a technical one, and it is precisely what a parcel-count-only model misses.

Where the build stands

The pipeline runs end to end. All fifteen counties are ingested, clipped, allocated, classed, converted to service points and scored; the cube, the parcel layer, the file exports and the browser tool are built and published. One invariant is asserted on every run — meters summed by block equal meters summed by section equal the grand total — and the build fails if it ever stops holding.

Complete: Harvest (30 dated sources) · territory carve-out · clip · allocate · crosswalk · NNLS calibration · units · service-point fact table, 4,808,866 rows · roads · difficulty index · cost cube · parcel layer · GeoParquet, CSV and file geodatabase · tiles · cost model · MCP endpoint
What the estimate rests on: 39.9% of meters on a published assessor unit count, 13.8% on a structure count, 46.3% on a fitted use-code rate. Reported separately, never averaged.

Open — and the order they matter in: 1. What SCE’s 5.7M actually counts. Meters, service points, customer accounts and billed services are four different numbers. Costs nothing to ask; decides whether the 890,673 gap is even a gap.; 2. Real rates. Every cost control is a placeholder labelled bidder, benchmark or judgement. None is measured.; 3. Address points. The only source that says where a meter sits is blocked, so every location is modelled. The single largest available improvement.; 4. Historical work orders. A few hundred with actual crew hours would turn the whole productivity block from judgement into fitted parameters.

Explore

Interactive

Deployment cost model

The centrepiece. Controls, choropleth, waterfall, schedule, sensitivity. No database behind it — the model is a weighted sum over a pre-aggregated cube.

Open → Diagram

How it was built

The full pipeline, stage by stage, with the row counts and the decisions that shaped each one.

See the flow → For agents

AI access

Machine-readable catalogue, an MCP endpoint, and GeoParquet you can query over HTTPS without downloading it.

Connect → Reference

Field guide to the data

Every table, column and join, and the traps in each source layer.

Read → Assessment

Confidence register

What is measured, what is inferred, what is modelled — and where the estimate is weakest.

Read → JSON

Data catalogue

Every dataset with its URL, column glossary, row count and checksum. Generated by the build, so it cannot drift.

Fetch →

Metadata

Provenance is not a nicety in a bid deliverable. A reviewer who cannot trace a number back to a dated endpoint has no reason to believe any other number on the page, so the register below is generated from what actually ran rather than maintained by hand.

Build: Generated 2026-09-03 · catalogue version 1
Working projection: EPSG:3310, NAD83 / California Albers. Every length, area and distance was computed in it; reprojection happens at display time only.
Grain: One row per metered service connection in core.service_point. Every rollup — block, tract, circuit, substation, county — is a GROUP BY over that one table.
Invariant: Meters summed by block equals by section equals by county equals the grand total. Asserted on every build; the build fails if it does not hold. Holds.

Estimate confidence: Measured 1,918,871 Inferred 662,371 Modelled 2,227,624
Locations: Modelled everywhere. Meters sit at a parcel representative point because the national address-point service is blocked. A measured count with a modelled location is the normal case here — they are separate claims.
Licence and citation: California parcel geometry is a public record under the CPRA and is republished here on the condition that the county is cited as author, with endpoint and retrieval date. Citation file · source register.
Machine-readable reports: catalog.json · manifest.json · source_register.json · units_estimate.json · parcel_allocation.json · territory_carveout.json