Methodology
How the scoring
actually works.
Not a black box. This page explains the real mechanics — where AI is used, where a simpler fallback kicks in, exactly what the readiness score is built from, and precisely what company calibration is and isn't based on.
The scorecard
Every answer gets scored on named dimensions, not one number.
An AI model scores your answer against a rubric specific to the practice format — the dimensions for a technical drill are different from a behavioral story or a system design walkthrough:
Technical drills
Technical Accuracy, Depth, Structure, Clarity, Conciseness
Custom Question
Technical Accuracy, Depth, Structure, Clarity, Practical Application
Behavioral (STAR)
STAR completeness, personal ownership, quantified result, competency alignment
System design
Requirements clarity, component design, scalability, failure modes, trade-offs
Troubleshoot Sim
Methodology quality, hypothesis formation, isolation method, fix proposal, regression verification
Alongside the scores, every session returns strengths (what to keep doing), gaps (named specifically, not generic advice), a highest-leverage fix (the one change that would move your score most), a model answer built from what you got right plus what was missing, and up to 3 flashcards generated specifically from your weakest dimension — never a generic deck.
Where AI is used, and where it isn't
Real-time AI grading, with an honest fallback.
Every scored answer is graded by a frontier AI model against the rubric above. Two things are worth stating plainly:
- Reliability: scoring tries one AI provider first and automatically retries with a second if the first errors or times out, so a slowdown on one provider doesn't take grading down for everyone.
- Fair-use fallback: Sharp and Pro plans include a generous but capped number of AI-scored sessions per day, sized around real per-user AI cost. Practice past that cap stays unlimited, but returns an instant heuristic score (based on answer length and structure) instead of full AI grading, clearly labeled as such, until the next day's cap resets.
Company calibration
What “calibrated to Amazon” actually means.
Question style, level bar, and behavioral framing are modeled from published hiring frameworks, engineering blogs, and community-reported interview experiences for each company — not from insider access to any company's actual internal rubric. That is a meaningful distinction and we want to be precise about it: the calibration is a well-researched inference about what a company's interviews tend to emphasize, not a verified copy of their internal process.
In practice, this means the level bar (Foundations through L7), the mix of technical depth vs. trade-off reasoning expected, and which behavioral competencies get emphasized are tuned per company — but no claim is made that these match a specific company's actual scoring sheet.
The readiness score
One number, five real inputs.
Readiness isn't vibes — it's a deterministic formula combining five signals. Topic quality and breadth matter most, mock interview experience close behind, with consistency and trend as smaller adjustments on top:
Topic quality
Weighted heaviestYour average mastery score across every topic you've actually drilled.
Topic breadth
Weighted heavilyWhat share of the relevant topic set you've touched at all — depth on 5 topics scores lower than solid coverage across 40.
Mock interview bonus
Weighted heavilyYour average Betty mock score. For onsite-loop prep, this tracks all 5 expected panel types separately. A stale mock (no recent session) decays gracefully rather than dropping to zero.
Consistency
Smaller adjustmentBuilt from your streak, capped at 7 days. Breaking a streak costs points daily, not a cliff back to zero.
Trend
Smaller adjustmentWhether your last several sessions are improving, flat, or declining — needs at least 3 recent sessions to compute.
Speech metrics
Text-and-timing analysis, not tone-of-voice guessing.
Voice answers are transcribed, and eight metrics are computed from that transcript and its timing — not from acoustic analysis of tone or emotion, which we don't claim to do:
See it in practice
The mechanics only matter if the practice works.