How these numbers are computed

No computation. The value DOL prints in that row, with the date DOL attached to it. Where DOL prints a dash, the page says so instead of substituting anything.

Counted over DOL's published snapshot, read weekly and kept, since DOL overwrites its own.

DOL's own figure, unmodified. It describes cases DOL finished in the reported month, so it looks backwards and includes audited cases that ran long.

Counted over Determinations issued in the month DOL names.

For each month of determinations, the filing month at their median; then the change in filing month divided by the change in calendar month. The range comes from rolling three-month windows, because adjacent months are lumpy enough to produce rates of 0 and 2.0 that describe scheduling rather than pace.

Counted over Every decided case in the unioned disclosure files. DOL publishes today's position and keeps no archive, so this cannot be read from DOL at all.

A count of determinations by the month DOL issued them.

Counted over Every decided case in the unioned files. Lags the weekly queue page by a quarter.

Denied divided by decided, where decided is certified plus denied. Withdrawn cases sit on neither side: a withdrawal is neither an approval nor a refusal, and putting it in either place moves the rate.

Counted over Decided cases only. Pending cases are absent from these files entirely.

A group's denials divided by every denial in the corpus. Published next to the rate rather than instead of it, and never multiplied together into an expected count.

Counted over The same decided cases the rate runs over.

A Wilson score interval on the denied count over the decided count. It's asymmetric near zero, which a plain plus-or-minus is not, and near zero is where most of these rates sit.

Counted over That group's decided cases. Groups under the population floor carry no rate at all.

A state's biggest occupation or employer divided by its total filings. Occupations are grouped on the SOC code without its O*NET suffix, and employers on the slug rather than the spelling, because DOL writes the same firm and the same job several ways.

Counted over Every case with a worksite state, decided or withdrawn. It's a share of a census, so it needs no population floor; the counts are printed beside every bar.

The offered wage annualised through its unit of pay, then kept only between $15,000 and $1,000,000. Outside that band a value is a data-entry artefact, most often an hourly rate recorded as an annual one.

Counted over Certified cases only. A denied case's offered wage is what an employer proposed and DOL rejected, which is not what the job pays. Cells with fewer than 50 wages carry no percentiles.

The median of decision date minus received date.

Counted over Decided cases in that group. DOL works one national queue, so this barely moves by state.

A count of cases whose status is not final. Pending is read off the case's own final flag, never off a list of status names: sixteen statuses appear in the data and a hardcoded list silently mis-buckets the next one DOL adds.

Counted over A per-case scan of DOL's case-status pages, refreshed on a rolling basis. A different source from the quarterly files, which contain no pending cases at all.

A running total of DOL's published pending counts from the oldest receipt month to yours. Arithmetic on DOL's own figures, with no rate applied.

Counted over Pending prevailing-wage requests as of DOL's stated date.

20 CFR 656 arithmetic, not an estimate. The filing window closes at the earlier of 180 days after the first recruitment step and the wage determination's expiry, and the tool says which one applied.

Counted over Your dates only. Nothing about anyone else's case enters it.

USCIS suppresses its smallest counts, so a total built from published cells has a known floor and an unknown remainder. The answer is a floor and a ceiling, never a point.

Counted over USCIS's published employment-based adjustment inventory.

Parsed from the published charts. A cell reading C means the category was open to every priority date and U means it was shut to all of them; those are opposites, so they are drawn as different states rather than one shaded band.

Counted over Employment-based charts only, read from the Internet Archive because the live site refuses automated readers. It lags the current month, which is why it is framed as a history.