IC.
GITHUBGH LINKEDININ
SYSTEM 002 · HIRING SIMULATION · PUBLIC + LIVE

Liveweek.

A desktop OS in the browser where candidates live a simulated first week - real messages, meetings, fires. A hidden evaluator scores how they think, not what they claim.

30 SECONDS INSIDE THE SIMULATION - FROM THE LAUNCH FILM, RENDERED IN CODE

01 · THE PROBLEMWHAT WAS BROKEN

Resumes don't show how someone works under a real week's pressure. Interviews reward performance; work reveals judgment. So the interview became a week of work.

02 · THE SYSTEMWHAT SHIPPED · WHO USES IT
The Liveweek desktop mid-fire: WhatsApp decision card, Metabase clone, Discord, dockThe in-sim Metabase clone running a real SQL query over the seeded warehouseThe login screen - the workday drifting behind frosted glass

REAL SIM SCREENS · THE DESKTOP MID-FIRE · THE SQL ENGINE RUNNING · THE LOGIN

A desktop OS in the browser - inbox, chat, meetings, fires - where a candidate lives a fictional first week on the job. A hidden evaluator scores judgment, prioritization and written thinking as they work, not as they interview.

What changed: 173 candidates screened by a week of work instead of a stack of resumes - and the hire came from the simulation.

173candidates ran it
31finished the week
1hire made

SOURCE: Hiring panel · July 2026

03 · UNDER THE HOODWHAT ACTUALLY HAPPENS

Pick a moment. This is what the system does with it - every step a deliberate call, none of it visible from the outside.

01
The login page is the concept in one frame - behind frosted glass, the workday is already drifting: a support fire, a sprint board, a calendar filling up. You are signing into a week that is already alive.
the living backdrop
02
The desktop boots into a reading lobby. The clock is frozen at 9:45 while you read last week's notes and the chat backscroll - preparation is free, because the test has not started yet.
clock frozen at 9:45
03
Press start and every candidate's day begins at exactly 10:00 - every run is the same length, so every score is comparable.
uniform day length
04
A heartbeat drives the world: one simulated minute every 0.9 seconds. Hide the tab or lose the network and time pauses - walking away cannot silently burn your week.
1 sim-minute per 0.9s
05
THE CLEVER BITthe clock respects thinking. It freezes while you type and runs at half speed while you read a decision - care costs nothing, stalling costs time. A visible tooltip says exactly that; the sim never lies about its own rules.
typing-aware clock
04 · WHY NOT ONE PROMPTWHAT A ONE-SHOT WOULD HAVE GOT WRONG

One prompt did produce a version of this - and it is the reason the real one exists.

  1. The one-shot was built, and rejected. The first pass was a single HTML file with invented app interfaces. It read as a toy, and a toy cannot measure how someone works. That rejection is the fork below.
  2. The realism came from a real dump. The week's content was mined from actual office conversations, then fictionalized by design - role names only, invented numbers, a stand-in product. No prompt contains that raw material.
  3. The scoring is an accreted rubric. Cold test runs by real people kept exposing false negatives - the diligent candidate who pulled a number a day early, the banter nobody should be penalized for skimming. Every guard in the engine cites the run that broke it.
  4. The data layer was rebuilt on a taste call. Canned dashboard answers were replaced with a hand-written SQL interpreter over a seeded warehouse, purely because canned numbers felt fake. One prompt does not make that call.
05 · THE DIRECTIONWHERE THE HUMAN WAS
THE FORK · WHAT THE AI PITCHED VS WHAT I SENT

Two days in, the first playable version comes back: a single HTML file with invented app interfaces - a WhatsApp that does not look like WhatsApp, a Discord with a made-up game icon.

WHAT THE MODEL HAD BUILT

Its cheaper, plausible default: approximate the apps, invent the icons, ship a stylized demo of an office instead of an office.

WHAT I SENT BACK · TYPOS KEPT

the UI of the apps should reflect real UI more accurately - whatsapp does not look like how you coded it. [...] This feels like a toy demo - nothing like a real website - feels really half baked really - dissapointed. [...] this is not a brand new concept - so let us not re-invent the wheel, let us use the current wheel and build on it

That message is why every app in the sim is a pixel-faithful clone with the real brand hexes - and why the whole desktop stands on a proven open-source desktop-OS pattern instead of an invented one.

SESSION TRANSCRIPT · 2026-06-06 · QUOTED EXACTLY · [...] MARKS OMITTED SPEECH

The realism IS the test - you can only measure how someone works if the tools work like the real ones.

8 decision records8 app replicas, exact hexesfictionalized by designheadless sim-QA harnessHOW IT WAS RUN, NOT HOW IT WORKS