Studio
How KDC gets built — and maintained
A one-person software studio founded, populated, and now maintains this dataset with its own agent machinery — every run receipted, every change a pull request.
KDC is a working example of how the Bussetech Software Studio builds. One human and a workforce of gnomes — single-purpose agents, each named and doing one task end to end — took this dataset from an empty repository to a live, source-cited, public site, and every step is on the record.
Founded by filing one issue
KDC began as a single new-project issue. A founder gnome analyzed it and the studio’s factory did the rest — created the repository, wired DNS, turned on branch protection, published the site, configured its tests — and a human reviewed and merged the founding change. From there, two project gnomes do the ongoing work: one researches the world into per-source signals, the other resolves those signals into the records you browse here. Neither can merge its own work. Every change they make arrives as a pull request a person reviews first.
Trust is a property of the pipeline, not the model
The very first thing the research gnome produced, live, was invalid against this dataset’s schema — and the site’s continuous-integration checks refused to merge it. The prompt was corrected, the re-run passed, and no invalid record has ever reached the published dataset. That is the studio’s whole approach in one event: output you can trust is a property of the pipeline that gates it, not a promise about the model behind it. Every record on this site carries its sources; the dataset is open under CC BY 4.0, and the coverage page even documents where research looked and found nothing — because an honest negative is data too.
Open by construction, cheap by design
The dataset now holds 162 source-cited records across 48 US states and its first international entry, resolved from 716 per-source signals — a living count you can recount from the open data any time. Putting those records here took on the order of a dozen dollars of agent spend — counting the console-directed research runs that expanded coverage, not just the automated slice — inside a studio budget it publishes in aggregate. The studio doesn’t promise transparency; building this way, it can’t not provide it.
What’s proven, and what’s still being measured
Here is the honest line the studio holds itself to. Cycle correctness — that the machine advances cleanly from research to record to published site — is proven: the studio ran the full cycle on demand and watched it work (and fixed the defects those runs surfaced). Growth to 162 records was studio-dispatched — real runs, each a reviewed pull request.
What is not yet claimed is that the dataset maintains itself unattended. The scheduled daily refresh is now firing, and it is in a short reliability soak this month. Until that soak returns a verdict, the true claim is the one the studio makes: built in a day, sustained by dispatched cycles, a receipt for every run — scheduled autonomy in verification. When the soak passes, this page will be updated with the run links that earn the stronger claim, and not before.
Want the verdict when it lands? The studio will update this page when the autonomy soak returns. Use the subscribe box on this page to get a note when KDC resolves new records — and when the next study ships.