How we screen — the exact rubric, in the open

The scoring scale

Scores aren't vibes — they're anchored to a fixed distribution so a number means the same thing every time. Most real applicants land 55–75; a score of 85+ has to be genuinely in the top tenth of everything we'd see for that role.

Score	Band	What it means
90–100	Exceptional	Top ~2%. Reserved for answers that teach the screener something. Almost never awarded.
80–89	Top ~10%	Specific and evidenced — we'd present them to a paying client without hesitation.
70–79	Strong	Real experience, minor gaps. Advances to the second-round interview.
55–69	Competent but generic	The modal band — where the typical decent applicant lands. Competent, but no specifics.
40–54	Thin	Vague claims, no numbers or tools, boilerplate phrasing.
0–39	Unusable	Off-role, copy-paste, AI-slop, or incoherent English.

The production score bands — a candidate's number maps to exactly one of these.

What actually counts, in order

Not every answer weighs the same. The graded work sample outweighs any interview answer — because talk is cheap and the work isn't.

Weight	What we score	Why
35%	Graded work sample	The single heaviest input. We ask for a real deliverable — a set of replies, a reconciliation, a JD, working code — and grade the actual work, not a description of it. Where the work lives behind a link — a video edit, a live site — the screener grades what the candidate writes about it, and the link is yours to open.
25%	Scenario judgment	A hard, role-specific situation. We're reading for how they'd actually handle it — the decision, the tradeoff — not a textbook answer.
20%	Experience specificity	Real names, numbers, and tools — not adjectives. 'I cut first-response time from 6 hours to under 2' beats 'I have great communication skills' every time.
15%	Written & spoken English	They'll work directly with a US company. Written answers carry most of this; when a candidate's recorded intro is in, its transcript is read under the same dimension — and the clip itself goes to the client unedited.
5%	Salary-band fit	A small nudge — whether their ask lands in the band you set.

Round-one weighting. The work sample is deliberately the heaviest input.

Each part is scored on its own, on the same 0–100 scale as the table above, and the score is those parts weighted as shown — every scored candidate carries the breakdown, so you can see how their number adds up. A part the questions never asked about is left out, and the others count proportionally more. When a search has a brief, each must-have in it is checked against the candidate's own words: one the application does not show costs 5 points, one it only partly shows costs 2, and together they cost at most 15. Salary-band fit is arithmetic: an ask inside the band, or below it, scores in full.

What a strong answer looks like — and a weak one

The line is specificity. Two candidates can answer the same customer-support scenario and score 40 points apart:

Scores well — specific, evidenced

“I'd acknowledge the second breakage first and own it, then make the refund right before it's asked for twice. I ran a Shopify inbox where I cut repeat-contact rate by rewriting the refund macro and adding a proactive replacement offer.”

Scores low — adjectives, no evidence

“I'm very hardworking and a fast learner with excellent communication skills. I always go the extra mile to make customers happy and I'm great at handling difficult situations.”

The rubric penalizes hard for the second kind: adjectives without evidence, AI-sounding boilerplate, answers that restate the question, unverifiable claims with no specifics.

Then a second round, written from their own answers

Anyone who clears round one gets a short second interview — three questions the AI writes from that candidate's own round-one answers and your brief, not a canned list. That's what catches a pasted or AI-written first round: it's hard to keep a fabricated story straight when the follow-ups are specific to what you claimed. The bar in round two is higher, and it's graded on whether they've actually done the thing — specific systems, numbers, decisions — not whether they can describe it.

Don't take our word — run it

The rubric above is live on this site. Answer a real scenario question and it scores you on the spot, with the same calibration and the same penalize-vague rules our pipeline uses.

Try the screener on yourself → Score a candidate in 2 min See a scored shortlist