25,000 applied.1,000 got in.
Every engineer clears five scored stages before a profile reaches you — and you get the marking, not just the verdict. Nobody skips a stage to hit a deadline.
Assessments run continuouslyWhere the other 96% go
Bars are drawn to the real counts, not to a shape that flatters us. Most candidates are lost at the technical assessment, which is the point of putting it early.
What each stage actually tests
Open a stage to see who runs it, how long it takes, what a pass looks like and what ends the process on the spot.
Work history, code samples and public repositories are read by an engineer who ships the same stack — never a keyword filter and never a recruiter. We are looking for evidence of ownership: systems taken from decision to production and maintained afterwards.
Remote work fails on writing far more often than on code. We score clarity, how someone escalates bad news, and whether an English-speaking reviewer could follow their reasoning without booking a second call. Product instinct is scored here too — whether they ask what the change is for.
An adaptive technical interview built from the candidate's own CV and the role criteria, in a real codebase with existing tests and existing mess. Submissions are marked with the candidate's name and country removed. Tooling is allowed and expected — being unable to explain a line of your own submission is not.
An hour or more on a problem from a real backlog, not a puzzle. The engineer running it has shipped the domain and is instructed to push back, because how someone handles disagreement about their design is more predictive than the design itself.
Two calls with former technical leads — not HR, not a colleague of their choosing — plus identity verification and employment-history checks. We ask each referee the one question that matters: would you take them back on your team tomorrow?
Six axes, weighted, marked blind
Does the change fit the codebase it lands in, and is the reasoning legible to the next person who opens the file?
Constraints and failure modes named before boxes are drawn, and an honest account of what the design gives up.
Measurement before theory. We score the route to the cause, not the speed of the first guess.
Written clarity under time pressure, and how bad news gets escalated when it is nobody's fault.
Evidence of maintaining what they shipped: on-call, migrations, the unglamorous second year.
How they disagree in review, and whether a junior on the same PR would come out of it better.
The evidence, not the verdict
A rating you cannot audit is marketing. Every profile arrives with the marked artefacts attached, so your own engineers can disagree with our score.
Aditya DesaiStaff Engineer · Mumbai · 11 yrs94rubric total“Took the slow path deliberately: measured the query before touching the index, and said so in the write-up. Lost two marks for a migration that would lock the table on a large dataset — he flagged the risk himself in review.”
Blind marking, twice
Name, country and photograph are stripped before a submission is scored, and two markers work independently. Where they differ materially a third marker settles it.
The rubric is calibrated against real outcomes
Every quarter we compare scores against how those engineers actually performed on engagements, and re-weight. Communication carries more of the total than it used to, because it predicted retention better than design scores did.
Acceptance is not permanent
Engineers are re-assessed at the close of every engagement and reviewed annually. Ending an engagement badly removes someone from the network whatever their original score.
Questions about the process
Engineers ask the first three. Hiring managers ask the rest.
The adaptive assessment is capped and unpaid, which is the trade we make to keep it a real codebase task rather than a 40-minute puzzle. Candidates who reach the tech lead interview are paid for their time from that point on.
The tech lead interview is a spoken walkthrough of the submission with a senior engineer who has read the marking. Tooling is allowed and expected — being unable to explain a line of your own submission is not.
Not by itself. A recognisable logo tells us a candidate cleared someone else's loop, which we score at zero. Rubric marks come from the artefacts, and blind marking means the marker cannot see where they worked.
Every accepted engineer commits to at least four hours of overlap with your core hours, and stage two gates on written and spoken English clear enough for review without a follow-up call.
Yes. The repository, the marking sheet with both markers' notes, and the system-design recording are all attached to the profile you receive.
Almost every team does, and we encourage it — we book the calls and stay out of them. Our rubric is there to save you the first three rounds, not to replace your judgement.
On every engagement close, plus a lightweight annual review. An engineer who leaves an engagement badly is removed from the network regardless of their original score.
Read the marking yourself.
Send a brief and we return three to five scored profiles in 48 hours, artefacts attached. Judge our rubric against your own.