People prepare for the EB-1A as if it were a checklist: ten criteria in 8 CFR 204.5(h)(3), meet three, done. That is half the analysis, and the half that denies well-prepared petitions is the other one.
Where the two steps come from
The structure comes from Kazarian v. USCIS, decided by the Ninth Circuit in 2010 and adopted by USCIS as the approach officers apply.
Step one is objective and narrow: does the evidence meet the plain language of the criteria? Does this award qualify as a nationally or internationally recognized prize? Was there a completed judging activity? Is this material about the person, in a qualifying publication?
Step two steps back from the pieces and looks at the whole. The Policy Manual puts it this way: the officer should consider the petition in its entirety, and the petition must demonstrate that the person has sustained national or international acclaim and that their achievements have been recognized in the field — indicating that the person is one of that small percentage who has risen to the very top of the field of endeavor.
Those are different questions. The first asks whether documents fit categories. The second asks who you are in your field.
What step two actually rewards
Three things, and none of them is volume.
Coherence. A record that points in one direction reads as acclaim. A record that lists unrelated achievements reads as a career. The strongest files can be summarised in a sentence: this person is known in this field for this, and here is who recognizes it.
Externality. Evidence that exists because other people acted — citations, adoption, invitations extended to you, coverage written about you, prizes awarded by others — carries the weight. Evidence you generated about yourself carries very little at this step, however well written.
Sustainment. The definition asks for sustained acclaim. A single remarkable year followed by silence answers that question badly, and so does a long career of competent work with no peak. What reads well is recognition that kept arriving.
What it punishes
The padded file. Three strong criteria plus four weak ones is often worse than three strong ones alone, because the weak ones tell the officer how the file was built. Having seen a stretched section, an officer reads the rest with that in mind.
Letters instead of facts. Recommendation letters are qualifying evidence and they are not a substitute for the thing they describe. Ten letters saying you are exceptional, with nothing external corroborating it, is ten people's opinion.
The field defined to fit. Narrowing the field of endeavour until you are at the top of it is visible from the outside. "One of the leading researchers in applied topology" survives scrutiny; "the foremost expert in our proprietary methodology" does not.
What to do about it in practice
Write the one-sentence version of your record first, before assembling anything. If you cannot write it, the file will not produce it either, and no quantity of exhibits will fix that.
Then assemble the evidence that supports that sentence — and leave out what does not, even when it technically qualifies. A short coherent file beats a long scattered one at the step that decides.
Where to look next
The EB-1A page sets out both steps and the ten criteria. The criterion where the gap between the two steps shows most clearly is original contributions of major significance, where evidence routinely proves originality and stops short of significance. If you are weighing this category against the NIW, which asks a prospective question instead of a retrospective one, see EB-1A or EB-2 NIW.