I-821D Medians Fall 53 Days From Summer Peak, but an Early Read Shows a Bounce

Two signals define this week's I-821D picture, and neither cancels the other. Over the past eight weeks, the median final-stage wait has fallen from a peak of 181 days in late July to 128 days this week, a 53-day drop that the regression confirms is real (slope -8.9 days per week, R² of 0.86). At the same time, this week's read is preliminary: 1,582 approved cases, below the prior-week totals that ranged from roughly 1,800 to more than 9,000. The week-over-week bounce of +14 days, from 114 to 128, may partly reflect that thinner sample rather than a true reversal.

The 12-week arc runs from 86 days in late June (W27), through a cluster of readings above 160 days in July and early August, down to 114 days last week and a preliminary 128 this week. The R² of 0.86 on the 8-week slope places this firmly in the category of a confirmed directional trend, not week-to-week scatter. The bounce to 128 from 114 is worth watching, but a single week at a below-average sample size does not break the broader pattern.

This week's preliminary 128-day median sits 49 days above the same week last year, when the median was 79 days on a sample of 1,267 cases. That is a +61.6% gap. The most visible contributor to that elevated comparison is the summer 2026 run-up: medians exceeded 150 days from W28 through W35, a stretch that included the 181-day peak in W31. Whether that stretch reflects a policy-related filing wave, a processing slowdown, or both remains unclear from this data. The current median is below that summer range, and the trajectory is downward, but the distance from last year's level remains substantial.

The 3-year same-week average also sits at 79 days, aligning closely with last year's reading. That consistency across two reference points places the current 128-day level in the same elevated position relative to both the recent past and the longer baseline. The gap is not a product of an unusually easy prior year; it reflects how far the summer 2026 run-up carried medians above historical norms.

The improvement since late July is real and statistically confident. The gap to last year is also real and still wide. Both are true simultaneously, and neither erases the other.

Looking at the active pending population across stages (Source: Stage durations, 30-day raw window), the largest share of cases, 49.3%, sits at the 'USCIS Is Currently Processing the Case' status. There, the median wait is 94 days, with a 25th percentile (p25) of 54 days and a 75th percentile (p75) of 149 days across 87,933 cases, a wide spread that reflects the variability in adjudication (USCIS's case decision step: approval, denial, or a request for more information) timing. The biometrics (fingerprint appointment) stage accounts for 32% of cases but moves quickly, at a 20-day median across 56,981 cases. Those two stages together cover more than 80% of the pending population.

The rfe_fast anomaly flag fired this week, registering a value of 0.06. Cases with an RFE response on file cleared in a median of just 8 days, compared with 133 days for cases with no RFE. That ratio of roughly 17 to 1 is statistically unusual relative to the 8-week trailing baseline. The most likely explanation is a batch effect: a set of RFE responses filed in the same narrow window may have cleared together this week. A single week is not enough to confirm any mechanism. Importantly, receiving an RFE is not itself a negative signal; most cases that respond to an RFE are still approved.