Running a hiring round with Trial Day.

1. Set up, once

Your name and logo

In the console sidebar, open Branding. The name you type appears on the candidate's first screen: "Your company asked you to complete this", with your logo beside our mark. This is the one step the console will insist on before your first invitation, because an assessment from nobody in particular is the fastest way to have a good candidate abandon it.

We do not remove our own mark. A candidate being assessed is entitled to see whose rules apply and who to complain to, and that is worth more to you than a white label: it is the reason they trust the thing they are about to spend an hour on.

The role

A role holds everything that does not change between candidates: which kind of analyst you are hiring, which tools they are assessed in, how much the simulation explains, and how long a link stays open. Set it up once and inviting somebody afterwards is a name and a click.

Every candidate on a role gets a different company to analyse, drawn from a measured pool and scored on an equated scale. Two people on the same role never see the same data, and their scores still compare.

2. Sit your own analysts first

Invite two or three people on your team who you already have an opinion about, and tick one of my own team when you create the invitation. Each of those sittings uses one assessment, like any other, and stays out of your candidate list.

What you are checking is not the product. It is whether the result agrees with you. If the analyst you rate highest comes out at the top, the scale means something in your world, and a stranger's score on the same scale means something too. If it does not agree, you have learned that before betting a hire on it, which is the point of doing it in this order.

Once two of them have finished, every candidate's report carries a line placing them against your team rather than against an abstract average.

3. Invite candidates

Create an assessment, add a name and an email address, and send the link, either from the console with Email the invitation, or by copying it into your own email, which is opened more often because it comes from a person.

What the hour contains

Six decisions, and three of them are the same for everybody:

• An analysis to check. Something a model already wrote, with a chart and a confident conclusion. Does it hold up? Roughly one in four is sound, so doubting everything is not a strategy.
• A dashboard to read. The database is closed for this one. They are handed tiles, charts, a table and three filters, and told plainly that a previous analyst built it and nobody has tidied it since.
• A number wanted now. One request arrives from somebody on their way into a meeting. There is no separate timer — it comes out of the same hour, and the report compares how long they took against their own pace on everything else.

The rest depends on the guidance level you picked. A standard or unaided sitting ends on one investigation rather than separate questions: what kind of problem is this, where does it sit, and what is behind it — each asked in the light of the last answer. Get one wrong and the next begins with the business having tried it and got nowhere.

Every company is generated with exactly one thing genuinely wrong with it, and which one is drawn per candidate. Two people on the same role get different companies, different faults and different answers.

4. Read what comes back

The top of the report says advance or not, and why, in a sentence. Below it are the decisions they made, what each was based on, and how sure they said they were.

The most useful part is rarely the total. It is the evidence trail: which figures they pulled before committing, whether they noticed the two tables disagreeing, and what they wrote when asked to justify the last call. That is the part an interview cannot show you, and it is why the report is designed to be read rather than filed.

Five things are scored, and the two newest are the ones a technical screen cannot reach: Depth, whether they got to the cause or stopped at the first true thing, and Supervision, whether they checked a machine’s analysis before agreeing with it. Agreeing without checking scores nothing even when the machine happened to be right, because one analysis in four here is sound and a candidate who trusts everything is right a quarter of the time without having supervised anything.

Two things are reported and deliberately not scored: what they wrote for the team, and how they spent the hour. Both are for you to read rather than for arithmetic to rank — how clearly somebody explains a finding to a colleague who does not write SQL is the part of this job no machine here judges.

Mark a result as read and it leaves the “back and unread” group on your dashboard, so the console always opens on what still needs a person.

Using it fairly

Trial Day is one input to a human decision, and the terms say so: a score must not be the sole basis for rejecting anybody. There is no bias audit yet, which matters if you hire in New York City, where local law requires one before an automated tool is used in hiring. Our security page lists what else we do not have.

Every candidate is told what is recorded, who sees it, and how to have it deleted, before they start. If one asks you for a copy or a deletion, write to [email protected] and we will action it.

If something goes wrong