Back-testing the model against a published site P&L

Operating data from privately owned Supercharger sites is scarce. Almost every published figure is a projection, which makes any model built from projections hard to check — it can only be compared against other people's assumptions.

So when a site developer publishes first-year figures from a station it actually runs, that is worth taking seriously. SoCal EVSP publishes a year-one summary for a 12-stall site in San Diego on its site: about 1.75 million kWh dispensed, $1,051,200 of gross charging revenue, $105,120 of LCFS credits, $700,800 of electricity cost, a $12,468 lease buy-out line, and $421,320 of net operating cash flow, on a total project cost of $1,216,236.

This post runs the same site through the engine and compares the result line by line. The exercise is a back-test of the model, not a review of the operator — the disclosure is more transparent than most, which is precisely what makes it usable.

Setting up the comparison#

The scenario is configured to match the published site as closely as the public figures allow: 12 stalls, 400 kWh per stall per day (the utilization their own calculator uses as its default), $0.60 per kWh retail, San Diego — so SDG&E's filed EV-HP tariff — and a total project cost calibrated to the $1,216,236 reported. Everything else stays at the model's defaults.

Two differences in convention appear immediately and are not errors on either side: the engine counts 12 months of 30 days rather than a 365-day year, so it dispenses 1,728,000 kWh against their 1,752,000, and prices revenue proportionally lower.

Line by line#

And the lines the published summary does not break out, which the engine prices anyway:

Net operating cash flow, before debt service:

• Engine: $378,121
• Published: $421,320
• Difference: −$43,199, or about 10%

What the comparison shows#

The model lands close, from a harder starting position. The engine deducts $212,300 of costs the published summary does not itemize — the Tesla network fee alone is $172,800 at the modelled $0.10 per kWh — and still arrives within about 10% of the reported figure. Two roughly offsetting differences produce that: the engine's electricity estimate is lower than the operator's actual bill, and its cost coverage is broader.

The operator's real electricity cost exceeded a flat blended estimate. This is the most instructive number in the exercise. The published electricity line implies $0.400 per kWh dispensed. The engine, pricing SDG&E's filed EV-HP tariff with its subscription-block demand charge, models $0.305 per kWh — and a flat blended assumption of $0.30 per kWh, the figure commonly used in quick pro formas, would have been lower still.

In other words, a real operating bill at a real site came in above both the napkin figure and the tariff-derived model. That gap has plausible explanations the public figures cannot settle — a different rate schedule, a different voltage class, demand behaviour worse than the modelled subscription, or costs bundled into the line that the engine books elsewhere. What it demonstrates is the direction of the risk: on the evidence of this disclosure, electricity assumptions err low more readily than high, and the territory's filed structure is the floor rather than the expectation.

Clean-fuel credits arrive net, not gross. The published $105,120 is exactly 1,752,000 kWh at $0.060 — the gross credit. The engine models $80,784, after a 15% aggregator commission and a month-one registration gap. The $24,336 difference is not a disagreement about the program; it is the distinction between credits earned and credits banked, covered in LCFS credits, net of commission.

The published summary does not reconcile exactly to itself. Revenue plus credits, less electricity and the lease line, comes to $443,052 against a reported $421,320 — a $21,732 gap that the itemized lines do not explain. That is unremarkable in a marketing summary, and it is a good illustration of why a full statement is worth having: a P&L that lists four lines cannot show where a fifth went.

What this does and does not establish#

It does not validate the engine. One site, one year, one set of published figures, and several unknowns is not a validation set — and the site in question is a strong performer at high utilization, which is the easiest regime for any model to get roughly right.

What it does establish is narrower and still useful: on the only real operating data available publicly, a model built bottom-up from filed tariffs, statutory tax rules, and contractual fee terms produces a number in the same neighbourhood as reality, while carrying a materially more complete cost stack. The lines it adds are real costs that a shorter summary leaves for the operator to discover later.

The engine is checked continuously against a spreadsheet oracle across 19 named scenarios, which is what actually governs its arithmetic. This comparison is a sanity check against the outside world, and the kind of check worth repeating whenever another operator publishes real numbers.

The methodology page documents every default used above, and the scenario wizard runs the same engine on any address.

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