How this publication works.

Who writes here

Articles are published by the Superalignment Team and credited that way. We do not attach individual bylines to work the group produces collectively, and we do not list contributors, reviewers or advisers who have not agreed to be listed.

Nothing on this site has been externally peer reviewed. Where an article reports a result of ours rather than summarizing somebody else's, it says so in place.

How we source

We cite the original paper, report, dataset or announcement wherever one exists, and we link it. Reporting is cited for institutional history, not for results. Secondary explainers are a last resort and are labeled as such.

For a quantitative claim we try to state what was measured, over what sample, by whom, and whether the number is measured, forecast or self-reported. Where we cannot establish those, the sentence says the number is unverified rather than repeating it as fact.

We separate three things that are easy to blur: what an institution stated publicly, what credible reporting established, and what we infer.

Corrections

When an article changes materially after publication, the change is visible on the page in a Corrections section, and the page carries an updated date. We do not silently strengthen a claim, and we do not quietly delete one that turned out to be wrong.

The corpus already contains one worked example: our own earlier copy repeated a widely quoted pilot failure rate, and the article that took it apart records what we had said and what we say now.

To report an error, write to [email protected] with the page and the sentence.

Review and maintenance

Where a reference page or survey has been reviewed, it carries a visible last-reviewed date. A review means somebody re-read the page against its sources and either changed it or found nothing to change. An updated date is not a review date. A page with no visible review date has not yet been reviewed under this policy.

Evidence expires. Where a page depends on results that a later study could overturn, it states what would change our position.

Commercial interests

Superalignment is a research group that also builds and sells software. Praxis and Verity are commercial products, and the research on this site informs them. That is a conflict worth knowing about while reading, so we state it rather than arranging the site to obscure it.

Research articles keep product references out of the argument. Where implementation is relevant, it appears in a labeled note at the end of the piece. If an article ever evaluates our own product, it will carry its methods and its results, and it will say plainly that we made the thing being measured.

The group is privately funded. No article on this site is sponsored, and nobody pays to appear in one.