How the grade works

What is graded

One thing: how long an office's queue for a given form ran, measured as pending cases divided by the rate that office was clearing them. A queue of nine months means that at the pace the office worked that quarter, a case joining the back of the line would wait about nine months to reach the front.

What is deliberately not graded

Denial rates. They are published on every page because USCIS publishes them, but no letter is built on them, and this is the single most important decision on the site.

An office that denies more petitions than another may be applying the law more strictly, or may simply be receiving weaker filings from a different applicant population. USCIS releases no outcome that separates those two explanations. A grade built on denials would silently encode a guess about which one is true, and would punish offices for the cases that walked through their door.

Queue length has no such ambiguity. Every applicant is worse off waiting longer, and the agency publishes both the queue and the rate it is cleared at.

Who is compared with whom

Offices are compared only within the same form, and only against offices of the same kind. A field office's I-130 queue is compared with other field offices' I-130 queues — not with its own N-400 queue, and not with a service centre.

That second rule was added after the first run of the engine. Service centres adjudicate at national scale, twenty thousand decisions a quarter against a field office's two thousand, and they hold the national backlog by design. Ranking them together put five service centres at the bottom of every table — which graded them on having a different job, not on doing it worse.

When no letter is given

• Where USCIS withheld the figure. Small counts are suppressed to protect applicant privacy and appear in the spreadsheet as a literal "D". Reading that as zero would hand the smallest offices a perfect record they never earned, so those rows carry figures and no grade.
• Below fifty decisions in the quarter. A queue ratio built on a handful of cases moves further on one straggler than on a year of policy.
• Below eight comparable peers. A percentile needs a distribution. With six peers a letter is an opinion wearing a measurement's clothes, which is why the service centres carry none.

How a small office is handled

An office that cleared its queue in one good quarter is not demonstrably better than a large one that nearly did. Every office's ratio is pulled toward the national ratio for that form in proportion to how little we have seen of it, so an office needs roughly a hundred decisions of its own before its record outweighs the agency's. Small offices therefore drift toward the middle rather than accidentally topping or tailing a table.

Where the cuts come from

Not from thresholds chosen by hand. For each form, the graded offices are sorted and the letters fall at fixed percentiles of that actual distribution: A for the fastest fifth, B to the 45th, C to the 75th, D to the 90th, F beyond it. "B" means shorter than most offices really ran that quarter, and the boundaries move when the agency moves.