The question
Platform scores are easy to game and easy to misread. The honest test is whether the score predicts a different, expensive outcome — landing an offer at a loop the candidate had no other signal into.
We tracked 318 candidates who declared the platform their primary interview prep, with consent to share outcome data. None received any platform-side coaching during the tracking window.
318
Tracked candidates
Consented, anonymized
1,440
Interview rounds
Across 18 named employers
0.62
Spearman ρ
Platform score vs offer rate
+27pt
Top-decile lift
Vs bottom-decile offer rate
Platform score vs offer rate
Each dot is a candidate; the x-axis is their composite platform score in the 90 days before their loop, and the y-axis is the share of loops they cleared to offer. The relationship is monotonic but noisy — the lower band is wide, and a handful of candidates with mid-band platform scores cleared at top-decile rates.
Cohort outcomes over the year
Offer rates climb with platform engagement, controlling for prior employer and YOE. The 2026 cohort tracked above their 2025 peers across all four quarters.
What predicts an offer best
Not all subscores predict equally. Defensibility — the panel's score on how clearly the candidate defends decisions in replay — is the single strongest predictor, edging out raw correctness. AI-rejection rate (how often the candidate rejected an AI suggestion that would have shipped a bug) is the surprise.
| # | Subscore | ρ | Top-decile lift |
|---|---|---|---|
| 1 | Defensibility (panel) | 0.58 | +31pt |
| 2 | Correctness on Hard tier | 0.54 | +28pt |
| 3 | AI-rejection rate | 0.49 | +24pt |
| 4 | Cross-cutting feature score | 0.41 | +19pt |
| 5 | Security category coverage | 0.37 | +15pt |
Caveats
This is observational, not randomized. Candidates self-selected into the platform. The 0.62 correlation does not mean the score causes the offer — it means the people scoring well are the people getting offers. The honest claim is narrow: practicing here is correlated with the outcome you want. We re-run the study in H2 2026.