How we reconstruct interview questions
AceOffer’s value is being more accuratethan scraped competitors. That only holds if the process is honest, so here it is end to end — including what “verified” does and doesn’t mean. Today: 610+ candidate reports across 8 companies.
The process, end to end
- 01
Sourcing — first-hand candidate reports
Every question starts from reports written by people who actually sat the interview and shared what they were asked. We don't invent questions or scrape a competitor's list. A question's frequency is simply how many independent reports mention it — so “48×” means 48 separate candidates described that round.
- 02
Clustering & deduplication
The same question shows up across many reports in different words. A multi-phase pipeline groups those variants into one canonical question, aggregates the raw snippets, and pulls out the details that repeat: the interviewer's actual follow-ups, the edge cases they push on, and the mistakes that sink candidates.
- 03
Grounded authoring — never fabricated
Each write-up is bound to the source material: every follow-up, edge case, and “what passers do” note traces back to a specific candidate report, not to a model's general knowledge. Where the reports don't say something, we leave it out or flag it — we don't fill gaps with plausible-sounding invention.
- 04
Verification — executed, not just proofread
Coding solutions are run against their test cases, so the code you see actually passes. System-design and ML breakdowns are fact-checked against primary sources. Content that can't be independently checked is marked as such rather than presented as settled.
- 05
Anonymization
We publish the question and how often it's reported — never who reported it. Reports are pseudonymous and stripped of personal details before anything is written up.
- 06
Monthly refresh
Interview loops change, so the catalog is refreshed on a monthly cadence as new reports come in — new questions added, existing ones deepened, and last-reported dates updated. Latest refresh: August 2026.
What we don’t do
We don’t fabricate follow-ups to make a page look complete, we don’t present unverified claims as settled, and we don’t identify the candidates who reported a question. When the source material runs out, the page stops — it doesn’t guess.
Frequently asked
- Are these the exact interview questions?
- They're reconstructions from multiple first-hand candidate reports, not a leaked question bank. Where candidates quoted the interviewer, we keep it verbatim; the surrounding structure is assembled from what several reports independently describe.
- What does “verified” mean here?
- For coding questions it's literal: the solution is executed against its test cases. For system-design and ML content it means fact-checked against primary sources. It does not mean a named domain expert has signed off on every sentence — where something still needs human review, we say so.
- What does the frequency number mean?
- The number of independent candidate reports that mention the question. A higher count means more people were asked it — it's a signal of how likely you are to see it, not a guarantee.
- How often is the catalog updated?
- Monthly. New reports are folded in, existing questions are deepened, and every question carries a last-reported date so you can see how fresh it is. The latest refresh was August 2026.
- Do you identify who reported a question?
- No. Reports are pseudonymous and anonymized before write-up. We publish the question and its frequency, never the person.
See it for yourself
Every question shows its frequency and last-reported date. The free Waymo and Mercor catalogs are fully public — read a few and check the receipts.
Browse the interview guides