Release Notes 2026-08-07
Release notes for the 2026-08-07 hushh.ai production release: callable listings ranked first, regulator doors for every record that can carry one, the business page, and the MVP scorecard reaching 100 percent.
TL;DR: The top of a search result is now something you can act on rather than a name you cannot reach, every record that can carry a regulator door now has one, and the public MVP scorecard reached 100 percent on the live site.
Status as of 2026-08-07: see body.
Callable first, and the page for the businesses whose people we already list
Two threads in one night. For the person searching: the top of the list is now something you can act on. For the business reading about us: there is finally a page that says plainly what we sell and what we never will.
A listing you can act on outranks one you cannot
Searching Stamford for home services returned ten named people in a row, not one of them with a published phone number, while local businesses whose numbers the regulator does publish sat underneath them. We had scored being a person and forgotten to score being reachable. A published phone now outweighs a bare name, a record you can verify comes next, and a shell company stays at the bottom even when it has a phone.
Where the regulator publishes no phone at all we still refuse to invent one. In Connecticut that is zero of 821 trades listings, which is the inventory telling the truth rather than the ranking failing, and it is the honest limit of what better ordering can do.
The candidate pool was alphabetical before the ranking ever ran
Our loaders stopped collecting at three times the page size, in the order the source file lists people, which is the alphabet. The ranking was choosing the best of the first thirty names in the alphabet. That cap is gone, so every matching row in the area is ranked, every time.
Every record that can carry a regulator door now has one
Our earliest curated insurance listings predate registry links, so they offered no phone and no way to verify: a name you could neither call nor check. Any finance, insurance or healthcare record arriving without a door now gets the real one, your state insurance department, the SEC adviser search, or the national provider registry. The trades still show no button, because we hold no verified lookup for them and a wrong door is worse than none.
The business page says the quiet part in the headline
We already list your people. The directory is built from public regulatory records, so the licensed humans on a company's front line are already on it, and claiming a listing is free for every one of them, forever. What we sell is the platform that helps the business manage and serve that force. Every number on the business page is imported from the same measured files our coverage page publishes, no partner is named that has not signed, and the page states what we will never sell: the citizen, their data, placement, or anyone's listing.
The MVP now has a public definition of done, and it reached 100
There is a scorecard in the open repository that walks the product the way people use it, then prints a weighted percentage. Before this release the live site scored 64.6, then 97.9, and it now scores 100 twice in a row. It does not grade communication or commerce, because neither is built yet, and the right response to a full score is to add harder checks rather than stop measuring.
And a reflection on how all of this got broken in the first place
Every defect above was correct code wired to nothing, or wired to the wrong thing. None was a wrong calculation. That essay is published at Everything That Broke Was Already Written, along with the census it produced: 59 modules in our codebase that no other module imports.
Verified
Shipped through UAT with an 11 of 11 smoke pass, promoted, and verified in production with a 42 assertion smoke pass. The full timeline is at hushh.ai/releases.
Sources
- hushh.ai/releases
- hushh.ai/business
- Everything That Broke Was Already Written
- Production release verification on 2026-08-07 UTC: UAT smoke 11/11, production smoke pass, MVP scorecard 23/23 on two consecutive runs