I've been posting AI model tests on X for a while. It is a bad archive.
You get a screenshot, some post text, and almost none of the product. That is especially useless for frontend tests. A screenshot cannot tell you whether the buttons work, whether a game is balanced, or whether the model produced a broken mess just below the fold.
The posts also disappear into my timeline. If I want to compare a new model with something I tested months ago, I have to remember the wording I used and hope X search finds it.
So I built a proper home for them:
Browse the Adam Holter tests page
Each test has its own page with the prompt and every run I have collected. For newer runs, I kept the HTML itself. You can open the result in your browser and use it instead of judging a flattened image.
That matters for tests like the Hive Mind landing page, the interactive LLM learning site, and the three-genre browser game. The output is the application. The screenshot is only a preview.
Browse Adam Holter tests by model
The site also has a page for each model. The GLM-5.3 model tests page collects every GLM-5.3 test in one place.
The coding tool is saved as part of the run, but it does not create a fake second model. A GLM-5.3 test run through Qoder belongs beside a GLM-5.3 test run through ZCode. The tool matters. It is not the model.
There are still gaps
I am going back through old X posts and recovering what I can. Some have media but no surviving code. Some have vague captions that make sense only if you remember the thread from six months ago.
I am not going to invent an interactive result where I do not have one. Older tests will fill in as I find the original files.
For new tests, saving the result is now part of the process. The prompt, model, coding tool, date, media, interactive file, and X post can stay together instead of being scattered across several machines and an X timeline.