What we're learning building Oryxs in public.
Real bugs, real numbers, real decisions reversed. The same discipline the product runs on, made visible. One post per real event, not a content calendar.
A real defect, the wrong assumption behind it, the fix, and how it was verified.
Never a bare stat. Every figure ships with how it was measured and what it does not cover.
A call we got wrong, with the original argument left visible above its correction.
Real credit where it is due, named specifically, alongside the gap we think we can take.
Long-form argument under a byline about where this market is going and why we are here.
A missing citation looks exactly like a claim that didn't need one.
For days, four sections of every brief made specific claims and cited nothing. Nothing rendered as broken. The useful question for anyone evaluating an AI research tool is not whether it cites, but what it shows you when a citation fails.
We argued against the rewrite. Then we counted.
The case for not migrating was well reasoned and it measured the wrong thing. It sized the feature we were about to add and never sized what we would be adding it to.