Updated
Estimate scorecard
How close our dates land, PERM decisions first, graded against what DOL and the State Department actually did.
The short answer
- On 10,284 real DOL decisions (the nightly backtest), our dates were typically 3 days off, and 86% landed within a week.
- Both checks say our dates run late: in the backtest DOL usually decided about 3 days before the date given, and in the daily sample about 4 days before the date given.
6 more readings
- On 3,632 cases DOL's queue had already passed, our rate rule was at least 9 days off, against 3 days for answering "today". That counts the ones still waiting at the days they're late.
- The range we print near the front is re-measured every night from where DOL's decisions landed: tested on 61,931 cases it wasn't drawn from, 74% were decided inside it and 13% were still waiting two weeks after their date.
- 207 of our 207 grades are for dates within a month. Dates a month or more ahead haven't come due, so the long-range test hasn't really started.
- The fair test, which counts a case still waiting a month past its date as a miss, begins about October 27.
- The behind-the-queue rate, for cases the queue passed earlier: 151 graded, typically 6 days off, 1 more than two weeks off, and its dates run late.
- The RFI clock, for a case at a request for information, dated from its own RFI day: 56 graded, typically 2 days off, and its dates run late.
Typical miss on real decisions
3 days
Across 10,284 cases DOL decided between October 19 and November 8, 2026, each dated from the queue as it stood on the first of those days, at DOL's measured pace. This scores decided cases only. The daily sample below also counts the ones still waiting past their date.
Within a week
86%
of those decisions
Inside the printed range
74%
of 61,931 cases near the front, tested on start days the range wasn't drawn from, and 13% were still waiting two weeks after their date
Right about the window
83%
said "decided by November 8, 2026" or "not yet", and was right
DOL’s pace used
533 a day
the measured average before the first day
Recomputed every night. This run: November 9, 2026.
Every estimate, checked
Each kind of date the site gives, rebuilt from the record as it stood earlier and set against what DOL, or the State Department, then did. Each check reruns on its own: the PERM and wage-request ones nightly, the bulletin and H-2A, H-2B and CW-1 ones weekly.
- When DOL decides a PERM case waiting in its line3 days typically, 86% within a week10,284 cases DOL decided from October 19 to November 8, 2026, each dated as the line stood on the first day
- The range given with that date, near the front of the line74% decided inside it61,931 cases, tested on start days the range wasn't drawn from
- When DOL decides a PERM case its line has already passed8 days typically, where DOL's own average was 33 days off2,634 such cases DOL decided from October 19 to November 8, 2026
- When DOL decides a PERM case after an RFI3 days typically, 52% inside the range435 RFI cases DOL decided, dated from their own RFI day
- When DOL issues a prevailing wage determination1 day typically, 99% within a week3,548 requests DOL decided through November 8, 2026
- When a priority date a year past the cutoff becomes current7.6 months typically, 23% within a quarter1,853 dates replayed on every visa bulletin since October 2016
- When DOL decides an H-2A application53% inside the window's middle half32,539 certifications, each dated from DOL's earlier ones
- When DOL decides an H-2B application43% inside the window's middle half15,225 certifications, each dated from DOL's earlier ones
- When DOL decides a CW-1 application52% inside the window's middle half1,288 certifications, each dated from DOL's earlier ones
A fresh sample, every morning
Each day a few random pending cases from every filing month get the estimate the case page would show them, written down before DOL decides and graded once it has. Nobody picks the cases. The figures count the methods the case page uses today.
Recorded
741
predictions since September 26, 2026
Graded
207
DOL has decided these (withdrawals left out)
Typical miss
5 days
median distance, decided cases
Counting the late ones
at least 6 days
131 still waiting past their date, counted at the days they're already late
Inside the range
80%
decided between the range's two ends
Dated within a month
5 days207 graded of 572 recorded
| Case | Recorded | Predicted | Decided | Miss |
|---|---|---|---|---|
| G-100-25365-519620 | 2026-11-05 | 2026-11-17 | 2026-11-07 | -10 days |
| G-100-26029-603302 | 2026-11-06 | 2026-11-17 | 2026-11-06 | -11 days |
| G-200-26058-668782 | 2026-11-06 | 2026-11-17 | 2026-11-06 | -11 days |
| G-200-25329-436898 | 2026-11-05 | 2026-11-17 | 2026-11-06 | -11 days |
| G-100-25351-492310 | 2026-11-03 | 2026-11-12 | 2026-11-06 | -6 days |
| G-200-25306-357356 | 2026-10-28 | 2026-11-09 | 2026-11-06 | -3 days |
| G-100-25328-434270 | 2026-10-27 | 2026-11-09 | 2026-11-05 | -4 days |
| G-100-25363-513903 | 2026-10-26 | 2026-11-03 | 2026-11-05 | +2 days |
People watching their own case: the date each one's page showed is recorded once, 517 so far. Of the 99 PERM cases DOL has since decided, the date was typically 3 days off.
The other estimates
The wage-request date and the H-2A, H-2B and CW-1 dates, written down and graded the same way. Each H-2A, H-2B and CW-1 application is recorded once, in its first week, as its case page dated it.
| Estimate | Written down | Graded | Typical miss | Inside the range | Tested on past filings |
|---|---|---|---|---|---|
| Wage-request day | 450 | 90 | 1 day | 76% in the printed range | not tested this way |
| Wage-request month | 108 | 40 | 8 days | 93% in the month | not tested this way |
| H-2A, counted back from the first day of work | 47 | 35 | 34 days | 17% in the middle half | 53% of 32,539 landed in the middle half |
| H-2A with no first day known, counted from filing | 206 | 91 | 18 days | 22% in the middle half | not tested this way |
| H-2B, from the same season a year earlier | 325 | 61 | 2 days | 98% in the middle half | 43% of 15,225 landed in the middle half |
| CW-1, counted from filing | 206 | 6 | 29 days | 0% in the middle half | 52% of 1,288 landed in the middle half |
| Visa bulletin release day | 22 | 4 | 9 days | 0% in the middle half | not tested this way |
How these are graded
Methods the case page no longer uses
- DOL's published average, which the case page used for cases DOL's queue had just passed: 40 graded, typically 10 days off, 13 more than two weeks off, and its dates run late. Retired October 7, 2026: the case page no longer dates cases this way.
- Our main method before Nov 4 2026 (cases ahead of yours, at DOL's flat calendar pace): 166 graded, typically 3 days off, 4 more than two weeks off. Retired November 4, 2026: the case page no longer dates cases this way.
Before the daily sample began, a few cases were written down by hand, exactly as the case page printed them then. Their windows are the months-wide ranges of those earlier methods.
| Recorded | Case | Filed | Anchor | Window | Status now | Decided | Error |
|---|---|---|---|---|---|---|---|
| 2026-08-28 | G-100-25324-425560 | 2025-11-20 | Around November 2026 | 2026-09-01 to 2026-11-30 | CERTIFIED | September 17, 2026 | -71 days, inside the window |
| 2026-09-13 | G-300-25324-425356 | 2025-11-20 | Around 2 October 2026 | 2026-09-30 to 2026-10-08 | CERTIFIED | September 17, 2026 | -15 days, outside the window |
| 2026-09-13 | G-200-26015-564165 | 2026-01-15 | Around 10 November 2026 | 2026-10-31 to 2026-11-28 | CERTIFIED | October 29, 2026 | -12 days, outside the window |
| 2026-09-13 | G-200-26075-707139 | 2026-03-16 | Around 15 November 2026 | 2026-11-02 to 2026-12-12 | ANALYST REVIEW | pending, 5 days to the anchor | not yet |
| 2026-09-13 | G-300-26166-017385 | 2026-06-15 | Around 21 December 2026 | 2026-11-30 to 2027-01-30 | ANALYST REVIEW | pending, 41 days to the anchor | not yet |
Notes on the record
- G-100-25324-425560: The first case recorded by hand, as /perm-case-status showed it on August 28, 2026. The next day the estimator gained a measured employer-initial shift, worth +2 days for this employer. The prediction of record is the earlier one, and the shift doesn't move it outside its window.
- G-300-25324-425356, G-200-26015-564165, G-200-26075-707139, G-300-26166-017385: The decision-pace model's first predictions, from the day it became the lead model (13 September 2026): cases ahead from the live census divided by DOL's measured 625 decisions a day, with the range from that rate's own spread. Recorded across four filing months, so the model is also tested far from the front of the queue, and copied from the live page as a reader saw it.
How this is scored
The error is the signed number of days between the date an estimate named and the day DOL's index first showed the case decided, read from the sweep that finished November 9, 2026. Positive means DOL decided later than predicted. Withdrawals are recorded and never graded: an employer pulling a case isn't DOL's clock.
Grading only decided cases would favour estimates that called an early decision, since those resolve first. So a prediction whose date passed more than 30 days ago counts as a miss while the case is still pending.
The backtest rebuilds the queue from today's records and the change log: a case's status on the day is the one its first later change moved it from. Cases found after that day are counted as if we had known them, which slightly overstates what the site knew at the time.
Entries are never edited. When the model changes after a prediction is recorded, the original prediction is still the one scored. Case numbers are public federal records. Nothing here identifies the person who watched a case.