Solar permitting is not moving the way the map says

What the numbers are

DC’s solar share of total permits: 4.60% (2020) → 9.14% (2025) — a 1.99× move, on a corpus that grew 3.7% across the same six years. The share move is real, not an artifact of the denominator.

Every count above is a count of residential solar permits, dated by issue date where the jurisdiction publishes one and by application date otherwise, applied identically to every series in this report. All three of these jurisdictions are 100% issue-dated.

Why the windows differ by city. Tempe’s records begin in 2019. DC’s begin earlier, but its 2018–2019 permit counts sit below a step in our own coverage — 29,792 permits in 2018 against 48,317 in 2020 — so a DC series starting in 2018 would measure our ingest rather than DC’s construction. We exclude DC’s first two years for exactly the reason we exclude two entire cities below, and it is the same test in both cases.

Why we think this is real and not our own coverage growing

The obvious way to be wrong about a permit trend is to mistake your own data coverage for the market. A dataset that ingested more of a city over time will show growth that belongs to the ingest, not to the buildings.

So we checked the total permit corpus for every jurisdiction before making any claim about solar.

Two jurisdictions failed that test and were removed. Miami’s total permit corpus goes from 1,390 permits in 2022 to 39,159 in 2023 — a 28× step change that is our coverage, not Miami’s construction. Maricopa County goes from 25 in 2019 to 13,384 in 2025 and never stops ramping. Both classify correctly and both are certified; they simply cannot support a time-series claim, so they are excluded from the trend entirely.

Three jurisdictions passed:

Each corpus is measured over that jurisdiction’s own series window, so the check covers exactly the period the claim covers. In every case the corpus moved in the opposite direction from the effect, or barely at all — so the trend is understated if anything. Scottsdale’s corpus grew 11.0% while its solar permits fell 52%; its records run back further still, and measured from 2018 the corpus grew 13.6% against the same decline.

A note on scale. These three jurisdictions are not the same size. We hold 743,206 permits for DC, 288,142 for Scottsdale and 20,186 for Tempe. Tempe is roughly a thirty-seventh of DC, and a two-market comparison should say so rather than let a table imply parity.

Why we think these are actually solar permits

Permit records do not agree on what to call things. One city writes SOLAR ELECTRIC-RES-GREEN; another writes ELEC - ROOF MOUNTED INSTAL, which names a mounting method and not a technology.

For each jurisdiction we measured classification precision against a written definition and published the result. We report the lower bound of the confidence interval, never the point estimate:

These are measurements against a stated definition over the named populations, not hand-adjudicated audits. Precision measures whether the permits we placed in the cohort belong there. It does not measure whether we found all of them.

What precision is measured over, which matters as much as the number

A precision figure describes the permits our admission rules accept, not every permit the city issues. Those are different sets, and the gap is large:

So “99.71% precision in DC” means: of the DC permits our rules admit and classify, 99.71% are correctly classified — not that we have correctly classified 99.71% of everything DC issues. Roughly half of DC’s corpus is outside the admitted set: permits our rules cannot place, or that carry too little detail to classify. We publish both numbers because the first one alone is misleading, and a provider quoting precision without an admission rate is quoting half a measurement.

Two disclosures belong with those numbers

Scottsdale publishes no free-text description at all — 0 of 9,945 solar records carry one — so cohort membership there rests on the permit type string alone. We corroborated it independently against contractor names: 13 of the top 14 contractors on those permits are named residential solar installers — Tesla, SolarCity, SunRun, SunPower, Titan Solar Power. This is a stronger basis than a description, not a weaker one, because two independent fields could have disagreed and did not.

The corroboration also caught an error, which is the point of running it: 718 permits typed SOLAR HOT WATER-GREEN are solar thermal, not photovoltaic, and their dominant contractor is a plumbing firm rather than a PV installer — the second field disagreeing with the first, exactly where it should. 715 of those 718 predate 2019 and sit outside this report’s window; the three that fall inside it are 0.06% of the series, a correction we have specified and not yet applied — those three permits remain in the counts above. Applying it would move 2025 from 561 to 560 and leave the 2022 peak and the −52% unchanged.

Tempe’s certification was measured over a narrower set than the counts above. It covers permit code 801 — 1,803 of the 1,903 permits in Tempe’s cohort. The remaining 100 are confirmed solar by their descriptions but carry no certification. On the certified subset alone the decline is 62% rather than 60%, so the disclosure does not rest on which side of it you take.

One thing that makes DC’s series unusually clean, and unusually fragile

Every solar permit in DC’s series carries the same source string — CONSTRUCTION - SOLAR SYSTEM — in 100% of records, in every year from 2020 to 2025. That is why the +106% cannot be an artifact of our classification broadening mid-window: there is nothing to broaden. A single stable term maps one-to-one across all six years.

The same fact is a fragility, and we would rather say it than be asked. If DC ever splits that term into subtypes, the series would break silently on our side. We monitor the vocabulary for exactly this reason.

Why we publish it this way

Most permit-data providers publish coverage counts — jurisdictions covered, records held. Very few publish how often their classification is right, because measuring it costs real adjudication time and the answer is rarely 100%.

We publish measured precision with confidence intervals, per jurisdiction and per category, including the ones that fall below our own bar and the ones we have not measured at all. The complete status map is public: every jurisdiction, every category, every value and the date it was measured.

If a number here is wrong, it should be possible for you to find out. That is the point.