Hire me

Work/PRJ-08/Immigration · Canada

Express Entry in 2026: your stream matters more than your score

It depends on your stream more than your score. Canada hasn't held an all-program Express Entry draw since April 2024. In 2026, 41% of invitations went to the Canadian Experience Class, 37% to French speakers and 15% to occupation categories, and cutoffs ranged from 169 for physicians to 523 for the CEC. Once the draw size is known, the next CEC cutoff is predictable within about 2.4 points.

Data
All 448 Express Entry draws, January 2015 to 1 October 2026, plus the pool's score distribution (IRCC)
Methods
SQL modelling, data-quality checks, regression forecasting with a rolling backtest
Tools
SQL (DuckDB), Python: pandas, statsmodels
Code
github.com/BoredMongoose/express-entry-draws ↗
Business question

What CRS score does a candidate need for Canadian PR in 2026?

Recommendation

Category eligibility matters more than a few extra points: 2026 category cutoffs were 40–120 points below the CEC's. And once a draw's size is announced, its cutoff can be estimated within a few points.

Estimated impact2.4 pts

Average error predicting the next CEC cutoff, against 4.4 for 'same as last time' (29 draws, rolling backtest).

01Results

In 2026, the score you needed depended on your stream.

In 2026, the score you needed depended on your stream. Median cutoffs: physicians 198, French speakers 400, healthcare 475, trades 476, Canadian Experience Class 516.
Fig. 1In 2026, the score you needed depended on your stream. Median cutoffs: physicians 198, French speakers 400, healthcare 475, trades 476, Canadian Experience Class 516.
Canada stopped inviting candidates from all programs at once. Invitations by type of draw. 2026 is already a record year, with 135,215 invitations by 1 October.
Fig. 2Canada stopped inviting candidates from all programs at once. Invitations by type of draw. 2026 is already a record year, with 135,215 invitations by 1 October.
Since 2023, each stream has its own bar. Every draw's cutoff since 2015, including a draw of 27,332 invitations with a cutoff of 75 in February 2021.
Fig. 3Since 2023, each stream has its own bar. Every draw's cutoff since 2015, including a draw of 27,332 invitations with a cutoff of 75 in February 2021.
Bigger draws mean lower cutoffs. For Canadian Experience Class draws, each doubling of the draw size lowers the cutoff about 8 points.
Fig. 4Bigger draws mean lower cutoffs. For Canadian Experience Class draws, each doubling of the draw size lowers the cutoff about 8 points.
A simple model beats 'same as last time'. Draw size, the last cutoff and the days since the last draw predict each cutoff within 2.4 points on average across 29 test draws, against 4.4 points for repeating the last cutoff.
Fig. 5A simple model beats 'same as last time'. Draw size, the last cutoff and the days since the last draw predict each cutoff within 2.4 points on average across 29 test draws, against 4.4 points for repeating the last cutoff.

02How I did it

From raw data to a decision.

  1. Downloaded IRCC's full draw history (the data behind its rounds-of-invitations page) and mapped 22 draw names, which change almost every year, to 15 streams.
  2. Decoded the pool's 18 unlabelled distribution fields into CRS score bands and built the analysis tables in SQL.
  3. Wrote 11 data-quality checks. One caught an error in IRCC's own table: in draw 247 (April 2023) the 10-point score bands add up to 180 fewer candidates than their total, which looks like a transposed digit.
  4. Compared six forecasting approaches with a rolling backtest (each draw predicted only from earlier draws), and reported prediction intervals for the next draw.

03What could make this wrong

The limits, said out loud.

  • 01The forecast needs the draw size, which IRCC announces with the draw: it answers 'what if', not 'when'.
  • 0242 CEC draws is a small sample, and policy changes can move cutoffs in ways history can't predict.
  • 03This is an analysis of public data, not immigration advice.

04What I learned

  • 01Government data changes shape: 22 draw names for 15 streams needed a proper mapping layer.
  • 02Always backtest. The model that looked best in-sample (with the pool size) was worse on draws it hadn't seen.
  • 03Official data has errors too: one of IRCC's own tables contains what looks like a transposed digit.

05What I'd do next

  • 01Forecast draw sizes as well, using the immigration levels plan.
  • 02Publish an updating cutoff tracker after each draw.
  • 03Model category draws once there are enough of them.
See the code on GitHub ↗All projects