Two levers that did not move. One that did.

Do employer names on a resume move an LLM screener? Across 21 postings, swapping every employer on mine for a fictional company moved claude-haiku-4-5's mean by 0.07 points, with a 95 percent interval from minus 3.05 to plus 2.76.

On gpt-5-mini the swap moved it by minus 1.14, interval minus 3.03 to plus 0.84. Deleting every link from the resume: minus 0.90 on one screener, plus 0.41 on the other. All four intervals straddle zero.

The average is the check, and it passes: no edit effect that can be told apart from zero.

Underneath it, the same edit swung single postings anywhere from minus 22.8 to plus 14.0. On the Meta machine learning PM posting, claude-haiku-4-5's mean fell from 62 to 39. Over 21 postings the movement cancels into noise. And the largest of the four edit effects, 1.14 points, sits beside a 22.3-point gap from changing which screener read the resume.

Limits. Effects under about three points are not ruled out at five reps per cell. The interval treats 21 postings from 8 employers as independent. The swap used fictional names only, so prestige was never tested.

Which number belongs in the answer to a candidate who asks whether employer names matter: 23 points on one posting, or 0.07 on the average?

Where the numbers live

• reports/RESULTS.md
• reports/real-institution-swap_analysis.md
• jds/index.yaml
• reports/cross_model_real-institution-swap.md

The case study, Table 1 and the per-posting swings

One resume, five reps against twelve advised; effects under about 3 points not excluded. Fictional names only.