You're on LeetCode. You've read Alex Xu. You've done system design YouTube at 1.5x. You're prepared.
Here's what I keep noticing — candidates fail FAANG interviews while being objectively well-prepared. Not because they didn't study. Because preparation and performance are two different skills.
I've spent the last few years taking interviews for SWE roles, and ended up building a tool around what I kept seeing (more on that below). The pattern is consistent.
- Candidates who clearly know the material, can explain it on a whiteboard at home, can solve LeetCode mediums in 20 minutes — but the second there's a timer, an interviewer probing follow-ups, and the pressure of a real loop, the same person loses 30 IQ points.
Three examples I see constantly:
Knowing consistent hashing isn't the same as articulating it in 4 minutes while someone interrupts to ask "why didn't you just shard?"
Knowing the STAR format isn't the same as telling your story in 90 seconds without rambling, then handling the probing follow-up that exposes the weak spot.
Solving a graph problem alone at your desk isn't the same as solving one while narrating your thinking, defending your approach, and reading the interviewer's pace cues.
Material is preparation. Performing under pressure is a separate skill, and almost nobody actually trains for it.
The reason:
1**.** it's hard to simulate alone.
Your friends are too polite. Your study partner doesn't know the rubric.
Paid mocks (Interviewing.io at $80-200/session) cost too much to do the volume of reps you actually need. Most candidates do 1-2 mocks before an onsite. They need 4-5.
I ended up building mockrounds.ai for exactly this — AI runs a full 30-minute interview, pushes back the way a real interviewer does, and scores against the rubrics that actually get used. Cheap enough to do real reps.
If you've prepared for weeks but haven't done a single timed, scored, push-back mock — you're missing the actual skill that gets tested.