Hiring.

The Work Sample

5 min read·Apply by lunch

The question

For the last role you filled, what did you actually watch that person produce before you signed the offer — a document, a call, a piece of code, a plan for something real — as opposed to hearing them describe having produced things like it? If the honest answer is nothing, you didn't hire on evidence. You hired on a well-told story about evidence.

The idea

In 1968 Paul Wernimont and John Campbell published a five-page paper in the Journal of Applied Psychology called Signs, Samples, and Criteria, and it quietly reorganised how selection is meant to work. Almost everything we collect about a candidate is a sign — a trait, a test score, a credential, an impression across a table — something believed to correlate with doing the job. A sample is a piece of the job itself, performed and measured. Their argument was that we'd spent decades hunting for better signs when we should have been building better samples, because the closer your measure sits to the behaviour you care about, the less inference you have to do, and inference is where selection goes wrong. Sixteen years later, John and Ronda Hunter's meta-analysis in Psychological Bulletin ranked the standard predictors against each other and put the methods that have a candidate actually perform the work near the top of the table, far above credentials, experience and the interview as it was then practised. The practice never caught up with the finding. Founders still run four conversations about a job and never once watch it being done, usually out of a worry about asking for free labour — a real worry with a very easy answer: make it small, make it artificial, and pay for it.

The framework

01
Cut the sample out of the job, not out of a puzzle book.
Take the top outcome from the scorecard and find the smallest real piece of it: a one-page account plan for a plausible account, a forty-minute discovery call with someone playing the customer, a bug sitting in a repo you've stripped of anything sensitive, a rewrite of a support reply that genuinely went out last week. A good sample is one that whoever does this job today would recognise as a Tuesday. If they wouldn't, you've written a puzzle. Abstract reasoning tests and take-homes about fictional widget factories measure something you should be measuring another way entirely.
02
Two hours, paid, and the same brief for everyone.
Resistance to work samples comes from their worst version: the unpaid six-hour take-home that treats a candidate's weekend as a free resource. The objection is fair; the answer is cheap. Cap the task at two hours, say so in writing, and pay the going rate — a few hundred dollars against the cost of a wrong hire isn't a close call, and it turns the ask from an imposition into a transaction the candidate can decline without losing anything. The ones who wave the payment away tell you something too. Everyone gets the same brief and the same clock, or you're comparing tests that were never the same test.
03
The pilot run is what makes the rubric real.
Write the criteria before the first submission lands — three or four things lifted straight from the outcomes the role exists to deliver — then set the task to someone who already does the job well and mark their answer first. Their work becomes your description of strong, which is far more useful than an adjective you invented at a desk, and it tells you whether two hours was ever enough. Sometimes the pilot fails your own test, which is the cheapest failure available. Then mark the real submissions the way you marked the screen: names off, each one marked against the rubric before you open the next, compared only afterwards. Add a second marker here — a sample takes more judgement than a screen does, and two independent reads catch what one confident read won't.
04
A work sample is powerful and narrow.
It sees craft under artificial conditions and near-total attention. It doesn't see how somebody behaves in month four, how they take a correction, or whether they finish the things nobody is watching. Say out loud which lines of the scorecard the sample covers and which it doesn't, then assign the uncovered ones deliberately to the interview and the reference calls. Treat the sample as the whole assessment and it'll slowly become a new proxy for the job rather than a piece of it — which lands you back exactly where Wernimont and Campbell found you.

Members only · 4 principles + template + AI mentor

364 more frameworks are waiting.

The Work Sample is Day 1 of 365. One framework every morning for a year — across strategy, sales, negotiation, leadership, and more.

$1 /day

Billed $365/yr · cancel any time

365 frameworks

12 topics · templates · AI mentor

Reserve your place — $365/yr →

One payment a year · the whole library · no auto-renewal.

More Hiring frameworks

All 19 hiring frameworks →