We open the source. Then we say what we didn't check.
Loaded Tokens is a short AI news desk, not a lab. A verdict is our reading of the public record, with the limits written next to it.
If we can't open the page, it doesn't run
Someone on the desk has to open the cited page, see the claim on it, and check the date. A search snippet, an aggregator blurb, or a model's memory is not verification. If we can't do that check, the item doesn't run.
We link the specific pages we read, not a homepage. This is our recap, not a copy of the original.
Primary first, and a company claim stays a company claim
If the company, lab, or agency published the thing itself, we read that page and cite it. A news write-up of a filing is a fallback, not a shortcut past a document we could have opened.
Primary doesn't mean true. A blog post proves what the company said. Benchmarks, savings, user counts, and "fastest" claims stay attributed in the sentence that carries them. We label vendor benchmarks as vendor claims. We did not test the product ourselves.
Use, Watch, or Ignore
Every brief and tool page gets one of three calls:
- Use The public docs say it does what you need. We read the docs. We did not test it.
- Watch Promising, but something is missing: access, pricing, or proof.
- Ignore Not for you, at least not yet.
For a tool, the verdict is tied to the public docs, the price, the access rules, and the date we checked them. When a verdict changes, the page says what moved.
When we get it wrong
Email hello@loadedtokens.com. If a fact is wrong, we correct the page and leave a note about what changed.
Corrections: if a fact is wrong, email hello@loadedtokens.com.