Load Ledger

Two questions, answered from the paper: is this project real — and how sure am I. Everything below is how I decide.

SB 6 made the ERCOT large-load queue confidential, so the signal moved to the public record: air permits, water contracts, tax-abatement filings, PUCT dockets, transmission applications, county agendas. I read that paper every week and score each project. No project data is confidential, and every score traces to a document you could pull yourself.

Score by domain, not by document

This is the core rule, and the reason a well-publicized phantom can't beat a quiet real project. A real large-load project leaves paper across independent domains — power, air, gas, water, land, and fiscal records — correlated in one place over a few quarters. A phantom appears in a press release and nowhere else. So each domain contributes to the score once, at the strength of its best evidence. Ten documents in one domain never out-score four domains lit up together — and I collapse an abatement filing, its press release, and the agenda item approving it into one commitment, not three signals.

A project lit across four or more independent domains is real. Weights are recalibrated quarterly against what actually happened.

The four tiers

What you act on is the change: a project moving Filed → Committed in a quarter is the signal your window is opening now.

Real vs. phantom

Scores decay when no new paper appears, and drop on active negative events — a withdrawn permit, a terminated abatement, a resold parcel. A project that's loud in one cheap domain (announcements) while the expensive domains (gas, water, transmission) stay dark earns suspicion, not partial credit. That's the Phantom File: each week I name one announced project with no permit, water, or abatement paper found as of that date — stated as fact, with the methodology linked.

The rules I don't break

• Documents only. I assert that documents exist and scores derived from them — never that a company is lying.
• Null over guess. A missing value is absence of signal, never a fabricated one.
• Provenance on every claim. Source URL, retrieval date, archived document.
• Independence. I hold no positions in any project I cover.

I publish how often I'm right

This is the part no aggregator will match. I grade past scores against independent, physical outcomes — appraisal-district value jumps, metered load at the point of interconnection, construction imagery — signals the score never reads, so the test is honest. Two numbers, published and dated: the hit rate (of what I called real, how much got confirmed) and the catch rate (of the projects that got built, how many I flagged, and how early). Score conservatively, date every call, let the track record earn the trust.