Why this analysis exists

Community timeline threads are most useful when they answer a concrete question rather than merely accumulating anecdotes. The motivating question here is:

For the current public Raleigh/Durham sample, where is the observed delay actually concentrated, and how should an already-interviewed but still-pending case be interpreted?

The repository already publishes the right raw artifacts. What it needed was a reader path. This page defines the question, key-findings.md answers it, and pending-cases.md explains how to read the open tail without discarding censored cases.

Current snapshot

As of analysis date 2026-03-15:

  • filtered RDU cases: 47
  • closed cases: 29
  • pending cases: 18
  • receipt years represented: 2023, 2024, 2025
  • interview years represented: 2024, 2025, 2026
  • office filter: field_office == Raleigh/Durham

The canonical public dataset lives in data/canonical/rdu_timeline_data.csv; the exact filtered input used for the current site build is published at results/latest/processed/dataset_filtered.csv.

What this site can and cannot identify

This is a versioned observational sample, not a population frame. It is useful for describing the current public signal, comparing cohorts, and updating pending-case probabilities under explicit assumptions. It is not a substitute for USCIS administrative data, and it is not legal advice.

The motivating community sample is marriage-based AOS oriented, but the canonical analysis table does not currently encode a formal case type variable. That means the site can rigorously analyze observed durations, censoring, and cohort shifts, but it cannot yet stratify the published curves by case subtype without a schema change.

Analytical frame

The site focuses on two time-to-event questions:

  • total time: I-485 receipt date -> I-485 approval date
  • post-interview lag: interview date -> I-485 approval date

Pending rows are not treated as missing. They are right-censored observations at the analysis date and therefore still contribute information about the tail of the distribution. This is why the survival/CDF views and the pending_predictions.csv output are central, rather than optional.

The primary published models are fit on all office-filtered rows rather than a fixed receipt-year subset. Where the site needs a temporal comparison, it now keys that comparison to interview timing so early-2026 office activity remains visible even when the underlying receipt dates are still in 2025.

Reading map

Use the pages in this order:

  1. Key Findings for the empirical story.
  2. Pending Cases and Forecasts for the current open tail and conditional probabilities.
  3. Methods, Data Contract, and Assumptions and Caveats for the technical details behind the published curves.