Key Findings
1. The dominant queue is still receipt -> interview
Among closed cases with both timestamps known, the median share of total wait spent before interview is 98.96%. That makes the pre-interview stage the main determinant of total time in the observed sample.
The all-case survival/CDF views keep the right-censored tail in frame and therefore provide the least assumption-heavy top-level picture of total processing time.

The core structural interpretation is unchanged: most cases spend the overwhelming majority of their lifecycle waiting to reach interview, not waiting after interview.
2. More recent interview periods, including early 2026, appear faster
To avoid anchoring the story to receipt_year, the main temporal comparison now keys receipt -> interview to interview timing. That makes current office activity visible even when the underlying receipts are still from 2025.
| Interview year | Cases with known receipt -> interview |
Median receipt -> interview |
IQR |
|---|---|---|---|
| 2025 | 36 | 154 days | 108-212 days |
| 2026 | 7 | 71 days | 61-100 days |
The 2026 group is still immature: only 1 of the 7 interviewed-in-2026 cases is closed, and 6 remain pending after interview. So the current signal should be read as an early acceleration in the pre-interview stage, not a stable estimate of full end-to-end completion time for 2026-era processing.

3. After interview, approval is often near-immediate, but the long tail is real
For closed cases with known interview -> I-485:
- median delay is 1 day (
n=29, IQR 1-10) - 16/29 (55%) were approved within 1 day
- 23/29 (79%) were approved within 14 days
- 25/29 (86%) were approved within 30 days
For interview -> I-130 when known:
- median delay is 1 day (
n=21) - 17/21 (81%) were approved within 1 day
- 21/21 (100%) were approved within 7 days
When both post-interview approval dates are known, 12/21 (57%) were approved on the same day.

The central mass is heavily concentrated at day 0-14, but the observed sample also includes completed waits of 45, 63, and 190 days, plus current pending cases well beyond 30 days. The correct reading is therefore: the interview often ends the main queue, but it does not eliminate tail risk.
For the full artifact inventory, including raw plots and CSV outputs, continue to Results.