Communication Under Pressure
Reviewed August 2026 · Civil Systems LLCA negotiation you play by writing, not by clicking. A client is furious about a $30,000 change order and you have an hour to make the case. The simulation reads what you actually type, decides whether you made the argument and how you made it, and answers back. There is no chatbot behind it: just a small, deterministic language model that runs entirely in your browser.
Try it below, then see how it works ↓
You are the project manager at an architecture firm designing a new office building. In last week's design-review meeting, Priya asked for a set of marketing renderings, and you nodded. The next day you sent a $30,000 change order for the visualization work: modeling, materials, lighting, and revisions across several images. Priya is furious: she believes renderings are part of your normal services and refuses to pay a cent extra. The contract's scope language is genuinely ambiguous, so you cannot simply point to a clause. Your case is that visualization is an additional service beyond the base design deliverables, and one you never confirmed in writing. The project is healthy, but you did not budget for this and you will not give the work away. Your hidden floor for the full package is $20,000. Priya wants this resolved before her afternoon.
Why you write the reply instead of picking one
Most training simulations hand you three replies and ask you to choose the best. That tests recognition: you can often spot the good-sounding option without being able to produce it yourself. But no client hands you a multiple-choice menu when they are angry on a call. The real skill is composing the words under pressure.
So this sim makes you type. It reads your actual sentence, decides whether you made the argument and how you made it, and the client responds to what you wrote. You cannot pass by clicking something that sounds right. You pass by saying it.
Tests whether you can recognize a good answer. Fast to build, but a learner can score well without ever forming the argument on their own.
Tests whether you can produce the answer. Closer to the real moment, and the only way to practice finding the words when it counts.
How it understands what you write
It is not a chatbot. There is no large language model writing the client's lines, and nothing you type leaves your browser. The system does two small, predictable things, then follows a fixed rulebook.
You write a reply in plain language, however you would actually phrase it.
Exact keywords catch obvious phrases. A small 22M-parameter model catches meaning, comparing your sentence to example phrases for each argument, so you are not forced to guess magic words.
The readers settle what you actually made: acknowledge, scope, cost, value, proof. Each one counts only once.
A deterministic model moves the client's number, sets her reaction to your tone, and writes her reply. Same input, same result.
The language model's only job is to recognize meaning, so good writing is not punished for missing a keyword. Everything that decides the outcome, how far the client moves, when she walks, what she says, is a rulebook you could read on paper. No hallucination, no server, no per-use cost: about a 20 MB download that then runs offline in the browser.
The map the client is hiding
The client starts at $0 with a hidden ceiling, and she will not move for a number alone. She moves when you build the case in order: each argument you land unlocks a little more room, and only then does an offer meet her in the middle. Push too hard and she leaves. Explore the routes:
Every outcome traces to a rule you can read. No black box quietly deciding your grade, which matters when the whole point is seeing why you got the result you did.
It runs in your browser. No API bills, no accounts, and nothing you type is sent to a server, so it scales to a whole class for free.
The client cannot be charmed by keywords or bullied by pressure. Only a real, well-ordered case moves her number, the same way it would in the room.
Develop the skills to lead projects.
Critical Path is a professional development simulation for the people who manage real work.