We back-tested the model against a real Supercharger P&L. All-in power cost landed 1% apart.

Every Supercharger ROI model is a claim about a site nobody has operated yet. That makes them almost impossible to check — you can compare one set of assumptions against another, and learn nothing.

So when a developer publishes year-one figures for a station it actually runs, that is worth taking seriously. SoCal EVSP publishes a 12-stall San Diego P&L on its site: 1,752,000 kWh dispensed, $1,051,200 of charging revenue, $421,320 of net operating cash flow.

We ran the same site through our engine — built bottom-up from SDG&E's filed tariff, with no reference to their numbers — and compared line by line.

The result: the cost of getting a kilowatt-hour into a car — the utility bill plus Tesla's network fee — came out 1.05% apart: $708,162 against their $700,800. One figure is derived from SDG&E's filed tariff; the other is a working developer's own number for a station they run.

---

The scoreboard#

Normalized so both sides dispense exactly 1,752,000 kWh:

Rows one through three agree by construction — they are products of inputs taken from the source page, so matching there proves arithmetic, not accuracy.

The power-cost row is the real test, and it is deliberately stated as a combined figure. The source publishes one line, "Cost of Electricity: $700,800," and does not say whether Tesla's per-kWh network fee sits inside it. Comparing combined totals sidesteps that question entirely: on both sides the figure is everything paid, per kWh delivered, to the utility and to Tesla.

Our side of that row is built up rather than assumed:

$0.4042 against $0.4000. A model that reads utility filings and a developer who operates the asset land four tenths of a cent apart per kWh — and ours additionally shows that $109,633 of it is demand charge, which a single blended rate cannot reveal.

What this comparison rests on#

Being precise about the load-bearing assumption, because it is the one a reader should check.

The comparison is valid if their $700,800 covers both the utility bill and the network fee. Two pieces of evidence point that way: the line works out to exactly $0.400/kWh, and the same operator's public calculator uses $0.30/kWh for electricity — leaving exactly $0.10 for the fee, which is the standard per-kWh figure. Their published P&L also shows no separate fee line anywhere, and the only unexplained residual in it is $21,732, far too small to be a $175,200 fee.

If instead that line is electricity alone and the network fee is simply absent from their summary, the comparison changes materially: our utility bill of $532,962 would sit 24% below their $700,800, and their reported net operating cash flow would be overstated by the missing fee. We cannot settle this from public information, so both readings are stated rather than one being asserted.

One further caveat on how much the agreement proves. Our $0.4042 is derived from SDG&E's filings; their $0.30 electricity figure is a general-purpose default they apply broadly, not a San Diego derivation. So this is a derivation agreeing with an experienced practitioner's rule of thumb — real corroboration that the filings support the number, but not two independent derivations converging. The territory table further down shows how quickly that rule of thumb stops working elsewhere.

---

Three corrections that made it a fair fight#

The first pass showed a 10% gap. Almost all of it was measurement error on our side, not modelling error. Three fixes:

1. Scope. We price the utility bill and Tesla's network fee as two separate lines; the source publishes one. The original comparison set our electricity-only figure against their combined line, and then counted the fee again on our side — double-counting it. Comparing combined totals, as the scoreboard now does, removes the error without requiring anyone to guess how their line is composed. For reference, the decomposition that motivated the question:

That reading is plausible but unconfirmed, which is why the headline figure above compares totals instead of relying on it.

2. Day count. Our year is twelve 30-day months (360 days); theirs is 365. Utilization was scaled to 405.56 kWh/stall/day so both sides dispense identical energy.

3. Gross vs net credits. Their $105,120 is the gross LCFS credit; we book net of commission and the registration gap. Comparing those measures an accounting convention, not a disagreement.

---

Where a flat model runs out of road#

Here is the part that matters more than any single-line agreement.

Their model runs 400 kWh per stall per day, flat, from month one. Real sites ramp. We ship three calibrated ramp curves and price whichever applies, month by month.

Same site. Same tariff. Same $1,216,236 capex. Only the ramp changes:

Payback moves from 21 months to 158 months — under two years to over thirteen — on the ramp assumption alone. Nothing else in the scenario changed.

That is the difference between a site that self-finances in year two and one that absorbs cash for a decade. A flat assumption cannot represent either end of it.

Two things fall out of that table:

Effective electricity cost is not a property of the tariff. It climbs from $0.3042 to $0.4217 per kWh — 39% higher, with no rate changing. Demand charges are set by peak kW and energy charges by kWh, so a site running below its eventual volume spreads the same demand bill over fewer kilowatt-hours. A single blended rate is blind to this, in precisely the early years when a site can least absorb it.

The flat case is the optimistic bound. Even our Premium curve — the strongest ramp we ship — dispenses 27.8% less energy in year one than a flat 400, turning a $163,871 first year into $17,104. Modelling the steady state as year one front-loads several hundred thousand dollars.

---

The same site, priced in six territories#

The 1.05% agreement is a San Diego result. Their electricity figure is a generic default, not an SDG&E-derived one. Hold the site identical, change only the filed tariff:

A flat $0.30 describes this San Diego site closely and misprices the identical site by −$285,000 to +$868,000 a year elsewhere. On Maui the error alone exceeds the site's entire published net operating cash flow.

That is the case for reading filings. Not that it beats a blended rate everywhere — here it essentially ties — but that it is right where a blended rate fails badly, and a blended rate gives no signal about which case it is in.

---

What we carry that the summary doesn't#

Plus clean-fuel credits at $81,906 net rather than $105,120 gross — the aggregator commission and month-one registration gap — with a quarter of it landing as cash the following year.

And past year one: annual escalation on energy, capacity and rent; the demand charge re-derived each year as utilization moves; depreciation, the tax stack, debt service; and a utilization cap so growth cannot compound past what the hardware can physically deliver.

---

One thing the source does not state: the hardware generation#

Our engine models Tesla V4 Superchargers. The published page does not say which generation the San Diego station runs — it is described only as a "12-stall flagship," with no stall power rating, cabinet specification, or opening date given. The operator's own FAQ states company-wide that they "design, fund, permit, build, and operate Tesla Supercharger V4 stations and NACS-compatible DC fast charging infrastructure," and their installation guide works in V4 terms (350 kW per stall, 1.5–2.0 MW for an eight-stall site). Neither statement is specific to this station.

The deployment timeline narrows it somewhat. V4 dispensers — V4 posts running on V3 cabinets — have appeared in the US since October 2023, so any station built from 2024 onward could carry them. Full V4 sites with V4 power cabinets are far rarer: the first opened in Redwood City, California in September 2025, followed by Utah in January 2026 and Kissimmee, Florida in March 2026. A station that has already completed a full operating year at 1.75 million kWh therefore almost certainly is not a 500 kW V4 site, whatever its dispensers look like. No independent record of the station's opening date could be found; the "California's first private 24/7" description appears only on the operator's own pages.

For this particular comparison the ambiguity matters less than it first appears, because the three biggest levers are pinned to their figures rather than to hardware:

• Revenue and energy are matched exactly to their stated 1,752,000 kWh, so generation cannot move them.
• Capex is calibrated to their stated $1,216,236 total, so V3-versus-V4 hardware pricing is already absorbed.
• The energy charge follows kWh purchased, which is throughput-driven.

Where it could matter is the demand line. Our $109,633 comes from a subscription of 150 kW per stall — an operating choice about contracted capacity rather than a hardware nameplate, and one already listed among our supplied assumptions. A station that contracts materially more or less capacity per stall would move that line, and with it the 1.05%.

So the honest position: the comparison is generation-agnostic on revenue and capex, and generation-adjacent on demand. If the operator confirms the station's generation and its contracted capacity, the demand row can be re-derived exactly.

---

Everything we supplied#

From the source: stall count (12), retail price ($0.60/kWh), LCFS rate ($0.060/kWh), project cost ($1,216,236), lease buy-out ($12,468), throughput (1,752,000 kWh).

From us:

Comparison rows are confined to operating lines, since the published figure is described as operating cash flow. The ramp table's payback and 15-year columns are full-model figures on our own definitions, internally consistent with each other rather than with the source.

---

The verdict#

The model checks out on the line that can actually be checked. Against a P&L from people who build and operate these sites, it prices the all-in cost of delivered power within 1.05% — $708,162 against $700,800 — with no knowledge of their figures, and while carrying a materially more complete cost stack.

For a 15-year projection of a capital asset, agreement inside about one percent on the largest operating line is a strong result. It also arrives from the harder direction: not by tuning to match, but by reading filed tariffs and letting the number fall where it falls.

What it is not is a full validation. One line is genuinely testable here; the revenue and credit rows agree only because they were set from the source's own inputs, and the bottom-line comparison inherits a $21,732 residual the source does not explain. The honest scope of the claim is the power-cost row, under the scope assumption set out above.

Where the two models diverge, the divergence is the product working. A flat pro forma returns one answer. Ours returns the answer plus its sensitivity — that this site's payback is 21 months if it opens at full volume and 158 months if it ramps slowly, that its effective power cost swings 39% on utilization alone, and that the identical asset is $868,000 a year worse on Maui than in San Diego.

One honest note on the source: those published figures are modelled, not metered. Each resolves exactly to throughput times a round rate — 1,752,000 is precisely 12 × 400 × 365, revenue precisely × $0.60, electricity × $0.400 — and the stated lines miss the stated total by $21,732. So this is a comparison against another practitioner's model of a real site, not an audited ledger. That is still a meaningful test, because the two share no method — and if the operator's metered bill, hardware generation, or contracted capacity differ from what is assumed above, those corrections would be more valuable to us than the comparison itself.

Our arithmetic is governed by a spreadsheet oracle across 19 named scenarios; that is what holds it to account day to day. This is the outside check — and it held.

The methodology page documents every default used here, the utility rankings publish the full tariff ordering, and what a flat blended electricity rate hides explains why the territory spread is so wide. To run this on a real address, the scenario wizard uses the same engine.

Published figures are quoted as they appear on the source page, read 2026-09-09, are the operator's own, and have not been independently verified; the reconstructions above are arithmetic on those published numbers. Engine outputs are model projections under the stated assumptions, not advice or guarantees.