Who runs this

ModelCensus is run by Nirnay Patel. It exists because how these models fail is genuinely interesting to me, and because the people building on them keep hitting the same failures with no shared, checkable record of how often they actually happen.

Everyone ends up rediscovering this privately, at their own cost. A public record — an open taxonomy, readable detectors, and a harness anyone can re-run — is worth more to the people doing that work than another private spreadsheet. That is the whole motivation.

The taxonomy is an original working proposal, currently at v0.9. It is not a standards-body output and not peer-reviewed. Every instrumented mode has a detector you can read and a case set you can re-run — issues and pull requests against the index are the point.

No vendor funds, sponsors, or reviews this work. No vendor sees results before publication, and none has any say over what is published. If something looks wrong, the corrections process is open to anyone, not just vendors.

Independence here does not rest on taking my word for it. The panel, case sets and detectors are fixed and published before a run, every failed trial shown on this site is human-reviewed, and every cell links to the transcript it came from — so any finding about any vendor, favourable or not, can be re-run and checked against the record.

This site carries no advertising and no sponsorship. Nothing here is for sale, and no services are offered. Model costs are paid personally.

This is personal work, done independently and on personal time. It is not affiliated with, endorsed by, sponsored by, or connected to any employer, client or institution, and no employer data, systems, resources or confidential information are used. Everything published here is my own view and my own responsibility.