Got the results! (CELPIP in 2 days after a 2-year gap)
Scores
Current: L 12, R 11, W 10, S 11
Previous (two years ago): L 10, R 10, W 11, S 10
On the 12th I realized my old CELPIP was expiring on the 22nd, so I booked the earliest slot I could get: Saturday the 15th. That gave me exactly two days.
Preparation
I had completely forgotten the test format. I bought an official CELPIP practice set and worked through it. The approach that actually helped was using AI to find my weaknesses first, then drilling only those areas. Do not just do practice tests and feel vaguely better. Find the specific thing you are doing wrong and fix that one thing.
For me, it turned out my problems were not English at all, but rather format habits. Things like answering a "talk about a time when" task in the present tense instead of the past. Describing a picture with no spatial order, meaning no phrases like "on the left" or "in the background". Never attaching a reason to anything. Not reading the prompt all the way to the end. None of that is vocabulary. All of it was fixable in a couple of days.
I also did two full writing and speaking sets with AI feedback and read through this sub for recent trends.
Test Day Experience (ILAC Toronto)
The overall experience at the centre was quite good. The only annoying part was the loud keyboard typing and the chatter when people reached their speaking section. I guess that is the same situation with every other centre though. They provided earplugs, which I really appreciated. About the test itself:
Listening was moderately difficult. Nothing brutal.
Reading was exhausting. The passages honestly read like they were written by AI. I used the full time on every part and re-read carefully rather than rushing.
For writing, I was really hoping to get an 11 or 12. I ended up with a 10, and I probably would have applied for a re-eval if it was lower, but I only needed a CLB 10 across all bands so I do not really care.
Speaking is where it fell apart. I got nervous out of nowhere. Words just would not come. On Task 7 I could feel myself repeating the same points to fill 90 seconds, and the picture for Task 8 was so unrealistic I struggled to describe it. I walked out convinced I had tanked it.
But I did not stop TALKING. No long pauses, no trailing off; I filled the time on every single task. That is the one thing I would tell anyone. Even when you are circling the same idea, keep going. Silence scores as nothing, but a repetitive answer still scores.
It came back an 11. My self-assessment was completely wrong, and I have since seen enough posts here to know that is normal. You remember your stumbles and forget the 80% that was fine.
What I actually recommend
Practice with AI and tell it to be blunt: Ask it to evaluate against the four CELPIP dimensions and be harsh. You want to find errors now, not on test day.
Diagnose before you drill: Two days of fixing the right thing beats a week of general practice.
Fill the time: Every task, every time. This is the single cheapest thing you control.
Read the prompt to the last word: If it says "and by when", name a date. If it says "ask her if she would like notes", actually ask
Invent your own scenario: If the prompt is vague, make the world as you feel best then describe. For example, I was given a task to suggest what gifts to buy for a friend's colleague. There was no other information. I just invented the colleague's name, age and gender then reasoned my choices.
Honestly, having to redo this every two years until PR at around $340 a pop is exhausting. But it is doable in two days if you are targeted about it.