One real task in. A ranked, defensible shortlist out.
A resume is a marketing document the candidate writes about themselves — and we keep treating it like a lab report. Here's how Starhas Learning measures the work instead of the paper.
Task
You give us a real problem from your codebase, or pick one from our library matched to the role. No abstract binary-tree puzzles unless that's genuinely your job.
Run
Candidates get the brief and a fair clock. Take-home for depth, live for pressure — you choose. Everyone works the same problem, so results are comparable.
Score
Automated tests handle correctness. A structured rubric scores quality, edge cases and reasoning — same lines for every candidate.
Shortlist
A leaderboard with the work attached. Interview the top of the list already knowing they can do the job.
A library of real tasks, or bring your own
Every challenge is a stripped-down version of work an engineer actually does — a bug to fix, an API to design, a data pipeline to repair. Pick one matched to the role, or hand us a task from your backlog and we wrap it as a timed challenge.
Challenges are great for volume screening: same brief, same clock, same rubric, ranked automatically. You point your engineers at the top of the leaderboard instead of the top of the resume pile.
## Fix the idempotent order endpointDuplicate POSTs create duplicate orderswhen the client retries on timeout.# Acceptance- "same idempotency key → one order"- "different key → new order"- "key expiry after 24h"- "survives concurrent retries"# Rubric weights → correctness .4 quality .25# edge_cases .2 reasoning .15
Where you find who ships under pressure
Challenges tell you who can solve a scoped problem. Hackathons tell you who can build something real over a weekend — and collaborate while doing it. Same platform, deeper signal.
We also partner with hackathon organizers, coding communities and campus programs that want their events to convert into real hiring — not just t-shirts and a leaderboard nobody uses afterward. Bring the participants; we bring the assessment layer and the pipeline.
The part a resume can't fake
We give developers an actual task from your stack and let them show the work — how they structure a solution, handle the ugly edge cases most people skip, and reason when the spec is messy.
How they build
Not just "does it run" — is the code clean, is it something the next engineer can maintain, did they name things like a professional.
What they catch
The retry logic, the empty input, the clock that goes backwards. The difference between "works on the happy path" and "works in production."
How they think
When the spec is deliberately incomplete, what assumptions do they make and how do they document them? That's the day-one signal.
Objective doesn't mean cold — it means the reasons are visible
"Isn't skill-based scoring just a different kind of subjective?" Fair question. Every challenge ships with a rubric written before the first submission lands: correctness weight, code-quality criteria, required edge cases, points for reasoning.
- Automated tests do the objective part — it either passes the suite or it doesn't.
- Structured human review, not freeform — reviewers score the same dimensions in the same order, and notes attach to the submission so anyone on your team can audit the call.
- Apples to apples — everyone solves the identical task under the same clock.
- Override with receipts — a hiring manager can move the rank, but they have to look at the code to do it.