865,275 Sealed Predictions: USTA Permit Ledger

This is a transparency post, not an accuracy claim: US Tech Automations Research seals every permit prediction before outcomes exist, and the ledger now holds 865,275 sealed predictions across 12 jurisdictions. Every record is hashed, timestamped, and locked in a content-addressed store before any outcome is observable, then scored against public records later, across 711 seal runs so far in the window June 7 – July 25, 2026.

Property transfers observed against those predictions stand at exactly 0, so no claim of predictive performance is possible yet, and none is made here. The 98.5% figure reported below is a record-join rate only: whether a sealed record could be re-identified in public permit data on the resolution date. It is not precision, recall, accuracy, or a success rate of any kind, wherever it appears in this post.

A sealed prediction, in plain terms, is a claim about a future public record that is locked by cryptographic hash before that record exists, so it cannot be edited once outcomes start to arrive. That definition matters more than it sounds: without it, there would be no way to distinguish a genuine forecast from a story told after the fact once the outcome was already known.

The ledger holds 865,275 sealed predictions across 12 jurisdictions.

Key Findings

• 865,275 predictions sealed across 711 seal runs in 12 jurisdictions, per the sealed ledger tables.

• Ledger window: June 7 – July 25, 2026 — every record hashed and timestamped before any outcome could be known.

• The first resolution pass checked 13,743 records and matched 13,543, a 98.5% record-join rate, not a measure of predictive accuracy.

• Property transfers observed so far: 0 — no predictive-performance claim is made.

• Resolution ran on July 23, 2026, against the same fixed set of sealed predictions.

The Ledger at a Glance

Each row above is a literal count from the sealed ledger tables. The match rate deserves the same plain-English caveat every time it appears: it measures whether a sealed record could be re-identified in public permit data on the resolution date. It says nothing about whether any prediction came true. The row that actually answers that question, property transfers observed, currently reads 0.

None of the rows in that table are estimates or projections; each is a direct count read from the sealed ledger tables on the dates given, the same discipline applied to every permit figure elsewhere on this site.

The ledger has grown since its first, June 2026 edition, both in prediction volume and in jurisdiction coverage, and the June edition of this ledger documents where it started. We do not restate that edition's figures here; each edition's numbers stand on their own sealed set.

865,275 predictions sealed, 12 jurisdictions covered, 0 property transfers observed to date. Sealed does not mean scored, not yet.

The predictions themselves draw on the same sealed permit snapshots behind our metro reports, including the Austin building permit report and the Cincinnati building permit report. The underlying edition snapshot looks like this:

Scope matters here too: these figures cover residential building permits (single-family and small multi-family); commercial and sub-trade permits are excluded at ingest. This is not a count of all construction permits issued in each city. All figures are computed directly from US Tech Automations' sealed daily permit snapshots; nothing is estimated, modeled, or extrapolated. That same restriction applies to the ledger figures above: a seal run, a match, or a transfer is either recorded in the sealed tables or it is not; there is no intermediate category of a likely or probable one.

How Sealing Works

The point of sealing is simple: make it impossible to quietly rewrite a prediction after the outcome arrives. The pipeline runs on a fixed, deterministic cadence.

1. Snapshot. Raw permit records are pulled from official municipal open-data portals for each covered jurisdiction and staged without modification, alongside the prediction records generated from them.

2. Hash. Every staged record is content-addressed: a cryptographic digest is computed over the exact bytes of the record, so any later edit, even a single character, produces a different digest.

3. Seal. The digests and records are committed to an append-only ledger with a timestamp. A seal run is one such commit; the ledger reflects 711 of them so far. Sealed entries are never updated or deleted.

4. Resolve against assessor data. On a resolution date, sealed records are joined against fresh public data, permit records now, county assessor and transfer records as the observation horizon lengthens, and the join result is recorded next to the original seal.

5. Publish the join. The joined result is published as-is, including the uncomfortable rows. A ledger that only surfaced flattering numbers would be marketing, not a ledger.

Because the hash is computed before outcomes exist, a third party can verify that a published prediction is byte-identical to what was sealed. That verifiability, not any performance number, is the product of this exercise. Permit feeds themselves are also revised after the fact by the jurisdictions that publish them, one more reason predictions are sealed against a point-in-time snapshot rather than re-derived from whatever the live feed says today.

That is also why this ledger reports seal runs and predictions sealed as two separate figures rather than one blended count: a seal run is an operational event, while a sealed prediction is the object being scored, and conflating the two would obscure how the pipeline actually operates.

711 seal runs have locked in 865,275 predictions so far.

First Resolution Pass

The most recent resolution pass ran on July 23, 2026. It took a fixed subset of sealed records, 13,743 of them, and attempted to re-identify each one in current public permit data. Of those, 13,543 matched, a 98.5% match rate.

13,543 of 13,743 checked records matched, a 98.5% record-join rate.

On July 23, 2026, the resolution pass checked 13,743 sealed records and matched 13,543, a 98.5% record-join rate, not an accuracy measure. Property transfers observed: 0.

Two things must be said plainly, every time this table is cited. First, 98.5% is a data-plumbing result: it confirms that sealed records can be reliably joined back to the public records they were derived from. It is not precision, recall, accuracy, or a success rate of any kind, and we make no such claim.

Second, the transfers-observed count is exactly 0. The predictions in this ledger concern outcomes, such as property transfers, that take time to appear in public records, and the observation horizon remains young relative to that kind of outcome. Nothing has had time to resolve into a transfer yet.

Until transfers accumulate in assessor data and are joined against sealed predictions, the honest scorecard for this ledger is: sealed, verifiable, and unscored on the question that matters most.

That scorecard changes only when transfer records actually accumulate in assessor data; no number of additional resolution passes against the same young horizon will produce a transfer that has not yet legally occurred.

We publish at this stage anyway, because publishing only after results are known is exactly the selection bias the seal is designed to prevent. The records that will eventually be scored are already locked; readers can hold us to them. Future resolution passes will be reported against this same growing set as transfer records accumulate in assessor data, in the same plain format used here: records checked, records matched, transfers observed. Whatever those numbers turn out to be, they will be published next to the original seals.

That promise, made in the ledger's earlier edition, holds again this time, and it only means something if it is kept when the numbers are unflattering.

Frequently Asked Questions

Q: Why seal predictions before outcomes are known?
A: Because sealing before outcomes is the only setup that prevents cherry-picking later. If predictions were published after results arrived, there would be no way to prove the misses were not quietly deleted. Sealing, hashing and timestamping each record in an append-only store, fixes the full set in advance across 711 seal runs, so the eventual scorecard covers everything that was claimed, not a curated highlight reel.

Q: What counts as a match in the first resolution pass?
A: A match means a sealed prediction record could be re-identified in current public permit data on the resolution date, July 23, 2026. Of 13,743 records checked, 13,543 matched, a 98.5% rate. That is a record-join result, confirming the plumbing works, not a judgment about whether any prediction came true.

Q: Why is the transfers-observed count 0?
A: Because the observation horizon is still young relative to the outcomes being predicted. Property transfers take time to surface in county records, and as of the July 23, 2026 resolution pass, the count of transfers observed against sealed predictions is exactly 0. Reporting that number plainly, rather than waiting for a flatter one, is the point of the ledger.

Q: Is this ledger the same one published in June, or a new one?
A: It is the same ledger, continued. The window runs June 7 – July 25, 2026, cumulative from the same start date as the June edition, not a fresh July-only count. Both prediction volume and jurisdiction coverage have grown since that edition, though we do not restate its specific figures here.

Reporting the ledger this way, as one continuous window rather than a fresh monthly count, is a deliberate choice: it keeps the sealed set, and the eventual scorecard, honest about what has actually accumulated since the first seal run.

Q: Why publish a ledger with a transfers-observed count of 0?
A: Because publishing only flattering numbers would defeat the purpose of sealing in the first place. 865,275 predictions are already locked across 12 jurisdictions; whatever the eventual transfer count turns out to be, it will be scored against exactly those sealed records, not a revised or curated set. A ledger that waited for a flattering number before publishing would not be a ledger at all; it would be a press release.

Q: Can a third party verify a sealed prediction independently?
A: Yes, that is the design goal. Each record is content-addressed, its cryptographic digest computed and committed across 711 seal runs before any outcome existed. Given a published prediction and its digest, anyone can recompute the hash and confirm the record is byte-identical to what was sealed.

Q: What would it mean if a future resolution pass came back worse than 98.5%?
A: It would be published exactly as this one is: records checked, records matched, the resulting rate labeled plainly as a record-join measure, and any transfers observed reported alongside it. Nothing about the format changes if the numbers are less flattering; that is the entire point of sealing predictions before their outcomes exist.

Why We Publish Our Ledger

Verifiable data discipline is the product. Contractors, suppliers, lenders, and agencies who use permit intelligence have no shortage of vendors making confident claims; they have a shortage of vendors who seal their claims before the answers come in. Publishing this ledger, including a transfers-observed count of 0, is how US Tech Automations earns the right to be checked rather than merely believed. That right has to be earned again with every edition, including editions where the honest scorecard reads exactly like this one does.

The same sealed snapshots behind this ledger drive our per-metro research, including the Austin building permit report and the Scottsdale building permit report, and the live corpus is browsable at permits.ustechautomations.com. Teams that want sealed permit signals wired into their own workflows, monitoring, lead routing, outreach drafting, can talk to us about what the data can and cannot support. We will tell you the second part too.

Source: US Tech Automations Research — computed from sealed daily permit snapshots, June 7 – July 25, 2026.