How these numbers are made, and where they fail.
Every figure on this site is measured from milestone dates that applicants reported about their own cases. This page says how, and publishes the model’s error whether or not it flatters us.
Where the timelines come from
Applicants record the dates their own application reached each stage — submitted, acknowledged, biometrics, medical, decision. Those reported dates are the raw material. Nobody’s file is scraped from IRCC, and no individual case is ever shown here or anywhere else on the site.
The model currently serving these pages was built from 126,953 reported timelines, last trained 22 July 2026.
Why a range and not a date
A single number would be a guess wearing a suit. Applications that are still waiting carry real information — a case that has been open 14 months without a decision tells you something precise — so the estimates are Kaplan–Meier quantiles, weighted so that unfinished cases count too. Drop them and only the fast cases get measured, which makes any model look better than it is.
What you see on a timeline page is the 10th to 90th percentile of elapsed days: eight in ten reported cases landed inside it. The middle band is the 25th to 75th. Neither is a promise about your case.
What is allowed to be published
A figure appears on a public page only when at least 25 weighted cases stand behind it. Below that the group is small enough that a number could start to describe individuals rather than a pattern, so nothing is shown at all — not a smaller number, not a caveat, nothing. The same floor governs what one signed-in person can learn about another.
Cells the model filled in from an external assumption rather than measured from real timelines are excluded too. Publishing an assumption next to measurements, in the same typeface, is how a methodology page becomes decoration.
Measured error
Scored against 7,201 real milestone events that happened after the model was trained, all of them after 2025-06-01.
Both bands land inside the tolerance we set ourselves, so a window labelled p25–p75 is doing roughly what its label says. That is the most we will claim, and it is why no date on this site carries a percentage.
Never tested at all
These stages happen too far out for the observation window to see them, so they were excluded from the scoring above. Their windows are not “measured and wide” — they are unmeasured, and where we know the direction of the error, it is early. These are also, unhelpfully, the dates people book flights on.
How this differs from IRCC’s own figure
IRCC publishes one number per category, and since 2024 it is forward-looking: an estimate of how long an application submitted today should take. It is the official figure and it is the one that governs your case. What it cannot tell you is how the wait is distributed, which stage you are about to reach, or how far the queue has actually moved.
Both numbers appear side by side on every timeline page here, with IRCC’s named as IRCC’s. Where they disagree, we say so rather than quietly picking a winner.
This model run
- Last trained
- 22 July 2026
- Timelines behind it
- 126,953
- Version
- 0.1.0-seed
2026-07-22_km_eb_seed_v1_3c927408e6
The full note
Estimates only — not immigration advice. These dates are modelled from timelines that other applicants reported, and describe a range of likely outcomes, not a commitment. Your case can differ substantially. Not affiliated with or endorsed by IRCC. Please do not make irreversible plans — flights, resignations, leases — on a predicted date.
Track your own application