A real pull of the Miami reroof file

Mix, so you can see this sample is not cherry-picked. The full 139-row pull is 114 residential, 18 commercial, 4 multifamily, 3 unknown. The 137 rows that passed this file's roofing selection are 112 residential, 18 commercial, 4 multifamily, 3 unknown. The 25 shown below are 20 residential, 3 commercial, 1 unknown, 1 multifamily — the 25 slots are allocated in proportion to that mix, highest declared value first within each class, not simply the 25 largest jobs. Until 2026-07-30 they were the 25 largest, which made a file that is mostly houses look mostly like warehouses.

Freshness is a property of the publisher, not of our cron. Each authority posts issued permits on its own delay, so no single sentence is true of both metros at once. Measured from our own daily pulls:

• City of Miami — the newest issue date it had published to us on the 2026-09-12 run was 2026-09-11 (1 day behind that run; 1–2 days across the last 10 runs)
• Miami-Dade County — the newest issue date it had published to us on the 2026-09-12 run was 2026-09-11 (1 day behind that run; 1–4 days across the last 10 runs)
• City of Orlando (other metro, shown for comparison) — the newest issue date it had published to us on the 2026-09-12 run was 2026-09-12 (0 days behind that run; 0–1 days across the last 10 runs)

The file on this page (2026-09-12 run) brought 55 permit numbers that were not in the previous run, out of 139 rows in the file.

Each nightly run re-pulls a rolling seven-day window, so consecutive files overlap heavily: a file is a snapshot, not a day's news. Measured over the 2026-09-05→2026-09-12 runs of this exact file, a run brought 0–55 permit numbers that were not in the previous run (median 13), and 3 of 7 runs brought none at all. That is the real refresh rate of this product, and the median is not a promise about any single day.

Orlando is the fast one; Miami-Dade is not. If you need same-week Miami issue dates, this file cannot give them to you, because the county has not published them at the moment we pull.

A fair question, and the answer is not the one a lead-list vendor usually gives. Miami-Dade's free open-data layer already carries the contractor's phone, the contractor's state licence number and the owner of record. We queried the live layer on 2026-09-12 before writing this: those fields are there, they are free, and anyone can query them. What is not in that free download is in the last column below.
So the honest claim is a narrow one: the merge, the classification, the licence-class decode, the parcel join that supplies roof age and roof area, the occupancy class, and the delta are the work. The phone and the owner on the county half are the county's, and free. Until 2026-07-30 this paragraph left the parcel join and the occupancy class out and the free sample did not ship those columns at all — they were being computed, sold, and never mentioned. That is also why there is no card field anywhere on this site — see the founding list.

Every licence number in the table is checkable in the state's own register: DBPR licensee search. We publish the number and the class exactly as printed on the permit record. We do not assert that any licence is currently valid or in good standing — we have not queried DBPR, and you should before you rely on it.

Masked here, and not sold at all. Phone numbers show the last two digits — 24 of the 25 rows shown carry one on the authority’s record. Owner of record shows the leading words only — 24 of the 25 rows shown carry one. Neither column is in the paid file: files-schema.json lists both as withheld and the publish step strips them from every servable file — we do not sell personal contact details, and the county’s own free layer carries them if you want them. They appear masked here to prove the authority publishes them, not to preview a column you would be buying. Licence numbers are shown in full, because they are the one field you can independently verify. The downloadable sample uses the paid file's own column names — it is a strict column subset of it, asserted on every build — so anything you map against the sample resolves against the real file. Until 2026-07-30 it did not: the sample called the same four fields valuation_declared, applicant_license_no, applicant_phone_redacted and owner_of_record_masked, and an import built on it broke on the file it was previewing.

That assertion used to cover one day. The 14-day spool the same endpoint serves held three incompatible schemas — 13, 17 and 31 columns, with contractor renamed to permit_applicant and contractor_phone to applicant_phone between them — so an importer written to this promise broke on thirteen of the fourteen days. Since 2026-07-31 there is one declared schema, published at /files-schema.json with the date each column first carries data, and verify-spool-schema.py exits non-zero if any file the endpoint can serve does not carry that exact header. A column added after a given day's run is present but empty in that day's file; it is never back-filled with a guess.

What was dropped from this pull, and why. Every reason below describes a record. None describes a company: an earlier version of this page dropped rows by running a regex over the applicant's company name, then published that guess as a statement about a named business's licensed trade. Selection now uses the authority's own discipline coding and the state licence class printed on the record.

The reasons below are the roofing-file selection applied on top of an already-gated product file. The universal gates — government owner of record, MEP sub-permit, corrupt valuation — run in the pipeline that WRITES the CSV, not here. Until 2026-07-30 they ran only here, which meant this page could report rows as filtered while the delivered file still contained them. This page now asserts that the delivered file carries none of them and refuses to publish itself if it does.

• state licence class printed on the permit record is “unrecognised-prefix”, which is not a class that pulls roofing work — 2 rows

Permit status, read back from the source. 24 not mapped from this source, 1 Active across the 25 rows shown, and 1 of them had a status we could read at all — Miami-Dade publishes no status field in its layer, so those rows say so rather than sitting blank. Where finality is published, 1 of the shown rows are flagged not final. Nothing is filtered on status yet: the column ships first so you can see a withdrawn or expired permit, and so the vocabulary each authority actually uses can be watched before anything depends on it. Until it is, read a county row as "a permit with this issue date was in the published feed at our pull", nothing stronger. Declared valuations are the applicant's own figure and routinely understate real project cost. Coverage for this file is Miami-Dade County and the City of Miami only — see coverage for every jurisdiction and date range we hold. This is a lead list built from public records, not a vetted referral, and nothing here is legal or investment advice. Before you call anyone on it, scrub against the federal and Florida do-not-call registries — we have not; see terms.

Built 2026-09-12 from out/permits/miami/solar-2026-09-12.csv · this week's free digest · EarlyDocket