Where the numbers come from.

The four sources we'd use. The two we actually do.

ESRI World Imagery and Copernicus Sentinel-2 are both free, both with predictable refresh, and together cover every site on the list. Planet and Maxar are higher cadence + resolution but cost real money — both are on the build list, neither is yet wired.

Built area ÷ planned area. Polygons drawn by hand.

• Built footprint. A polygon drawn on the latest cloud-free satellite capture, around finished structures only. Active construction zones are excluded so a half-built building doesn't count as built.
• Planned footprint. Official site plans, building permits where public, FAA filings for SpaceX, and zoning records. When a company has published a master plan, we use that.
• Cross-checks. County tax records, GIS portals, drone footage from local reporters (cited inline on the site page), and satellite timelapses.

These are estimates, not audited numbers. A human (the project operator) re-traces the polygons every refresh cycle.

Every percentage on the site has a measurable error bar. Here's how big.

The typical band is ±3 percentage points around the reported figure, based on the operator's own re-trace experiments across 10 sites in May 2026. Below: one site's progress curve drawn with that band shaded in. The dark line is what we publish; the grey envelope is the range the same operator produced on repeat traces of the same imagery.

Where you see it: the aggregate /pulse "Empire build-out, 2020 → today" chart now renders this same ±3pp band as a translucent strip behind the build-out curve — apply the same mental band to per-site sparklines (those are too small to show it without crowding).

Every milestone wears one of four labels.

The same tier system is tagged into the JSON dataset (one of high, medium, low, speculative ), so anything you read on the site can be traced back to a specific evidence standard.

Every site, every timelapse, every last-capture date.

Live numbers from the build pipeline. The timelapse module pulls one frame per capture date; 493 frames across 16 sites today.

What satellites can't tell you.

• ·See section 3 (Uncertainty) for the actual error-bar width on every published percentage (±3 pp typical).
• ·Indoor work is invisible. A 100%-built site can be at 0% production capacity until racks land.
• ·"Operational" means finished structures plus visible activity, not nameplate output.
• ·Cloud cover, sun angle, and seasonal vegetation move the polygon by ±2-3% between captures. We do not report sub-5% deltas as news.
• ·Investment dollars are the company's number, not ours — drawn from disclosures, IR decks, and government incentive filings.
• ·Announced projects with no visible groundbreaking get the speculative tag and no progress %.
• ·This is a one-person tracker, not a satellite intelligence firm. Audited capacity numbers belong to your equity analyst.

Primary first. Secondary as fallback. Reddit and AI never.

Every verified milestone links to a primary source — FAA notice, SEC EDGAR filing, county permit, IR press release, NASA mission page, ClinicalTrials.gov entry, NRC docket, or government data table (USGS, EIA, KOSIS, DART). The full bibliography lives at /sources — every URL we cite, grouped by site.

When sources conflict, both are noted on the site page (for example, company guidance $20-25B, industry estimate $35-45B). When something is an estimate, we tag it "est." Milestones without a verified primary-source URL are left uncited rather than padded with weak references — under-citing beats inventing.

Pre-pivot site records may still carry a Wikipedia revision pointer as the entry reference while migration to primary sources finishes through Q2-Q3 2026. Reddit and AI-generated content are never used as a sole source.