Bell Labs Was Filed Under "National Lab." Three Tribal Governments Were Filed Under "Nonprofit." Getting a Database Right Is the Same Problem as Getting a Country Right.

A flavorings and syrups company, acquired by Truelink Capital for $1B, was sitting in Oluwadi's own database under the category "AI Application." Nothing about it is artificial intelligence. It makes fruit purees and beverage bases. Someone noticed the mismatch on a screenshot of the site's own category filter, and asked the obvious question: if one row is wrong in a way this visible, how many others are wrong in ways nobody happened to be looking at?

That question turned into a full audit of all 11,839 companies in the database. This post is an honest account of what it found -- not a cleanup log, but a real answer to what "investigating the overlooked" costs when you turn it on your own tool instead of someone else's company.

The flavorings company wasn't a one-off. The same failure mode showed up independently, in six different reviews of six different parts of the database, run without knowledge of each other's results: an entity that was mentioned in a news item got filed as whatever the item was actually about, regardless of what the entity itself is.

A company that received a venture investment got filed as the investor. A law firm that advises private-equity deals got filed as a fund. Lockheed Martin, Northrop Grumman, and Bank of America -- three of the most recognizable company names in the country -- were sitting in a generic catch-all category because the newsletter blurb that introduced them into the database happened to describe a services relationship rather than what the company actually builds or sells. An institutional crypto exchange, backed by Citadel and Fidelity Digital Assets, was filed as a generic "Fintech" company because its own scraped description read like ordinary financial-services boilerplate.

None of this is a spelling error. It's what happens when a pipeline reads "X raised money" or "X was in this article" and defaults to filing X under the shape of the sentence, not the shape of the company.

The clearest example: Bell Labs, GE's Research Lab, DuPont's Experimental Station, RCA Laboratories, and IBM's Watson Research Center -- private corporate research institutions, several of them over a century old -- were filed under "National Labs," a category meant for federal facilities like Oak Ridge or Los Alamos. Alongside them: Cray, DEC, Control Data, and Silicon Graphics, historic computer companies, sitting in the same government-labs bucket, apparently because "lab" appeared somewhere in a description.

A sharper case sat in the same review pass: three federally-recognized tribal governments -- the Yerington Paiute Tribe, the Shoshone-Paiute Tribes, and the Washoe Tribe of Nevada and California -- were filed under "Civil Society / Advocacy," the category for nonprofit advocacy organizations. A sovereign government is not a nonprofit that advocates for something. That's not a taxonomy nicety; it's the difference between recognizing a government and treating it as a special-interest group.

The largest single pattern, by row count, wasn't a wrong category so much as a wrong grain. Of 587 companies sitting in department-level categories -- Business, Engineering, Medicine, Computer Science, and so on -- 393 were entire universities, not a specific school within one. MIT was filed as "Computer Science." Johns Hopkins was filed as "Engineering" -- despite Johns Hopkins' own defining, name-carrying strength being medicine and public health, not engineering. Stanford, the University of Michigan, Cornell, Northwestern: each one, an entire research university with dozens of schools, narrowed to a single department that in several cases wasn't even the institution's best-known strength.

Fixing this wasn't a matter of picking a better department. In the overwhelming majority of cases, no single department is the right answer -- the institution itself is a university, full stop, and forcing it into one school misrepresents what it is exactly as badly as the flavorings company misrepresented as an AI startup. A small number of institutions are genuinely defined by one discipline -- MIT and RIT by computing, Carnegie Mellon and Rice by engineering, Mayo Clinic and Cleveland Clinic by medicine -- and those stayed put, deliberately, rather than being swept into the generic bucket along with everything else.

Across the full audit: 1,176 individual companies had their category corrected, each one checked against real evidence -- a company's own description, and where that was missing or unreliable, direct research rather than a guess. A separate, structural error was fixed at the same time: nine entire categories, covering real sectors like Fintech, Healthcare, Defense/Aerospace, and Agriculture, had been wrongly flagged as "Unknown" at the database's root level, which meant 790 real operating companies were invisible to every region page's own company count -- not miscategorized, simply unrepresented, regardless of how correctly each individual row was otherwise labeled. Combined, roughly 1,970 companies -- about one in six of everything in the database -- now sit under a materially more accurate description of what they actually are.

Not everything got fixed, and the honest version of this post says so rather than rounding up. A meaningful number of rows were left exactly as they were, because the alternative was guessing:

Real taxonomy gaps came out of this that the site doesn't currently have a home for -- a tobacco company, university-affiliated startup accelerators, K-12 schools, a space museum -- each sitting in the nearest available category because no better one exists yet. That's an honest limitation of a closed list of categories, not a hidden one.

The single largest remaining pile is inside the Investor / VC category: roughly 1,300 rows with no description on file at all, mostly small or boutique-sounding fund names that are plausible by naming convention alone but unverifiable without either restored search capacity or a paid data source. A short list of well-known operating companies -- Amazon, Unilever, Arm, and others -- were flagged sitting bare inside "Investor / VC" with no fund-style name attached. Some run real corporate venture arms; some are likely scrape errors. Both readings stayed on the table rather than one being picked by guess.

Oluwadi's whole premise is that a place, an industry, or a company gets misread when the record keeping it either doesn't show up or shows up as something it isn't -- the same argument this site has already made about counties left out of federal datasets, about companies that never make it past a metro's headline name. That argument doesn't get to stop at the site's own front door. A database that files three tribal governments as advocacy nonprofits, or crushes a research university with a two-century medical legacy down to "Engineering," is doing the exact thing this whole project is built to notice when someone else does it.

This is published as a plain account of that, not a victory lap. The database is more accurate today than it was a week ago. It is still not finished, and this post names specifically where it isn't -- because the discipline this site asks of its readers is the same one it owes its own record.