A paid work trial is a final hiring stage. You do real work for the company for one to five days. They pay you for it. The team scores how you run that work. PostHog’s handbook says its one-day task is built to be too much for one person. That’s how it tests prioritization. Linear says its trials look at how candidates drive a project and communicate. Plane says it watches how you handle problems you haven’t rehearsed. So the trial rewards three things. You cut scope in writing early. You post check-ins that anyone on the team could act on. Your final demo states what you skipped and why. For an Android engineer there’s a fourth trap. A fresh checkout of a real app can take hours to build. Ask for help on that in the first hour. Don’t lose half of day one to a Gradle error.
We’ll follow one candidate through this post. They’re a mid-level Android engineer. A product startup has offered them a three-day paid trial after a passed virtual onsite. The brief asks for a saved-filters feature in the company’s Android app. Users should be able to save a search, name it and see it on other devices.
A paid work trial scores how you run a few days of real work
The companies that publish their process say plainly what they watch. Linear explained its trials in a December 2023 blog post. Candidates join a project team of three to five people for two to five days. They lead status check-ins and give a final presentation. Linear names the signals it looks for. They include how candidates think about problems and drive a project forward. Communication, self-direction and responsiveness to feedback are on the list too. Each team member gives independent feedback. Then the team votes blind before discussing. The hiring manager makes the final call.
Plane’s work trials page is undated. It describes a scoped project from the current product cycle. For engineers that means a slice of a feature. Plane says it watches how you handle unrehearsed problems. It also watches how you work with others when things are unclear. The people who worked with you write independent feedback.
Neither list mentions how much code you ship. The work counts as evidence of judgment. A half-finished feature with a clear plan can beat a finished one nobody understood. That differs from a take-home assignment, where reviewers only see the final repo. In a trial, the team watches the days as they happen.
Settle pay, hours and tool rules before you say yes
A trial costs you days, so get the terms first. Plane says every trial is paid at a rate put in writing beforehand. It adds that payment never depends on the outcome. Linear’s post says its daily rate is shared in advance. It says trial length depends on the role’s seniority. PostHog’s hiring process page lists a flat $1,000 for its SuperDay. The page shows no date, so check the current figure. Doist’s hiring FAQ is also undated. It says Doist pays candidates for the assignment stage.
Tool rules vary too. PostHog’s engineering hiring page allows AI tools for the main task. It bans them during a separate debugging session. Don’t guess at a company’s rule. Ask.
Our candidate asks the recruiter five things by email. What’s the daily rate? When is it paid? Which hours should they be online? Will they get repo access before day one? Are AI coding tools allowed? Who decides? By when? They also check their current employment contract. Some contracts limit paid work for other companies. If yours is unclear, ask before you accept.
Treat the task as too big and cut scope in writing on day one
PostHog’s engineering page says the SuperDay task is built to be too much for one person. The point is to see how a candidate prioritizes and ships. Other companies may not say this out loud. Assume it anyway. A team that wants to see your judgment needs a task where judgment is required.
The saved-filters brief has at least five parts. There’s a Room table for saved filters and a list screen. There’s a save dialog and a sync call to the backend. There’s also conflict handling when two devices rename the same filter. Three days can’t cover all five at production quality. So our candidate picks a slice that works end to end. Save, list and local storage come first. Sync goes in behind a flag. Conflict handling gets a written plan instead of code.
The key is that the cut is written down and sent early. A scope cut announced on day three looks like running out of time. The same cut posted on the morning of day one looks like a decision. It also gives the team a chance to say “actually, sync is the part we care about.” That’s cheap to hear on day one and expensive on day three.
Get the Android build running before you touch the feature
A real Android codebase rarely builds on the first try. It may need a private Maven repository or a signing config. It may need a local.properties key or a specific JDK version. A build variant might point at a staging backend you can’t reach yet. Each of these can eat an hour. None of them is the test.
So spend the first hour on one goal. Get a debug build onto an emulator and open the screen you’ll change. If you’re stuck for 20 minutes, post in the team’s channel. Give the exact error and what you’ve tried. The trial is meant to look like working on the team. Nobody on a real team fights a missing credential alone for half a day.
Our candidate hits a Gradle failure at 9:40. A dependency resolves from a private repository they can’t reach. They post the error line and the task name at 9:55. A teammate sends a token at 10:10. The app runs at 10:25.
A three-day message script you can adapt
Here are the three messages our candidate posts in the trial channel. The first goes out mid-morning on day one. The second goes out at the end of day two. The third opens the final demo.
Day 1, 11:00 - Scope plan Build is green on the emulator (needed a repo token, thanks Sam). Plan for the three days: 1. Saved filters stored locally in Room, with a list screen and a save dialog. Day 1 and morning of day 2. 2. Sync to the existing /filters endpoint behind a feature flag. Day 2. 3. Tests for the DAO and the ViewModel. Day 3 morning. 4. Not building: conflict handling when two devices rename the same filter. I'll write a short proposal instead. Question: is sync more important to you than the local flow? If so I'll swap 1 and 2.
Day 2, 17:30 - Status Done: local save and list, with DAO tests. PR #1 is open. In progress: sync. The endpoint returns 409 when a name already exists. I'm treating that as "keep both, add a suffix" for now. Blocked: nothing. Risk: sync may not be fully tested by tomorrow noon. If it isn't, I'll ship it behind the flag, switched off. Tomorrow: finish sync, ViewModel tests, then the demo.
Day 3, demo opening I built saved filters end to end on one device. Sync works behind a flag. I didn't build conflict handling. The proposal doc covers it. The 409 suffix choice is the decision I'm least sure about. I'd like your view on it. Here's the save flow on the emulator.
Why each message in the script works
The day-one message turns a scope cut into a decision the team can see. It lists what’s out, not only what’s in. It ends with a question that lets the team change the order cheaply. Linear says candidates lead their own check-ins. A plan like this gives that check-in something concrete to discuss.
The day-two message is short enough to read in 20 seconds. It separates done, in progress, blocked and risk. The 409 line shows a judgment call made in the open. That’s the kind of unrehearsed problem Plane says it watches for. The risk line also tells the team what to expect before they’re surprised.
The demo opening states the gap first. Someone reviewing the work won’t find a missing piece and wonder whether you noticed. It names the decision the candidate is least sure about. That invites feedback. Linear lists responsiveness to feedback as a signal. Then it moves straight to the working app.
Every message is written for someone who wasn’t watching. Linear and Plane both collect independent feedback from several people. Some of them may only read the channel. The messages are what those reviewers see. If a debugging session is part of the trial, it runs differently. Our post on what the bug squash round scores after the first fix covers that format.