---
title: "A sizing question with no right answer, and an agent that said so four different ways"
url: "https://tryreynolds.com/studies/data-centre-aisle"
description: "Not every question has a benchmark. This one is a room: two rows of racks, a raised floor, and a customer who wants to know whether hot air comes back over the top."
---

**[Reynolds](https://tryreynolds.com)** is agentic CFD, built by
**[InviscidAI](https://inviscidai.com)** and backed by Y Combinator, Winter 2026 batch.
The agent is open source under the MIT licence at
[github.com/InviscidAI/OpenReynolds](https://github.com/InviscidAI/OpenReynolds).
Using it requires no OpenFOAM knowledge.

This is the agent-readable Markdown twin of https://tryreynolds.com/studies/data-centre-aisle. It lives at https://tryreynolds.com/studies/data-centre-aisle.md, and
[tryreynolds.com/llms.txt](https://tryreynolds.com/llms.txt) indexes every one of them.

# A sizing question with no right answer, and an agent that said so four different ways

Not every question has a benchmark. This one is a room: two rows of racks, a raised floor, and a customer who wants to know whether hot air comes back over the top. There is nothing to score the answer against, so what matters is whether the run tells you how much to trust it, and where it would break.

RUN 14APPLIEDAPPLIED ENGINEERINGAUGUST 202620260829-074352-10ad

- **Asked**: “I am sizing airflow for a small data centre room. Two rows of server racks…” Does hot air recirculate over the tops of the racks, and what temperature does air arrive at the rack inlets?
- **Run**: buoyantSimpleFoam, steady, buoyant · 20 minutes · zero tool errors
- **Result**: Rack inlets at essentially the 18 °C supply temperature; recirculation forms but stays above the racks
- **Rack inlet temperature**: essentially 18 °C, the supply temperature, against an ASHRAE recommended ceiling of 27 °C
- **Recirculation**: cells do form above the rack tops, and do not reach down into the cold aisle
- **Its own airflow assumption**: 732 CFM per rack, against an industry rule of thumb of about 750
- **Implied rack face height**: 2.00 m, from continuity, which is what a rack is
- **Convergence**: partial, and reported as partial

![Temperature across a data-centre row: a cold aisle between two racks, hot aisles outside them, and recirculation over the rack tops](https://tryreynolds.com/assets/dc-temp-D_OHdWmR.png)

Figure 1 The answer, in one picture. Cold supply air (dark, 18 °C) fills the aisle between the racks; exhaust (red, 30 °C) leaves on the outside. The curl above each rack top is the recirculation the question was about: it forms, and it stays up there. Nothing warm reaches down into the cold aisle.

## The answer

Recirculation cells do form over the tops of the racks, and they stay there. Air arrives at the rack inlets at essentially the supply temperature, with no meaningful short-circuiting of hot exhaust back round to the front. Against the ASHRAE recommended envelope, rack inlet at or below 27 °C, that is comfortably in spec. Figure 1 shows it in one frame: the cold aisle dark all the way up, the exhaust hot on the outside, and the curl of recirculation sitting above the rack tops without descending.

## The assumptions it invented, and why they check out

The brief gave no rack power and no tile airflow, so the agent chose some. Those choices are where an answer like this usually goes quietly wrong, so they are worth auditing, and they hold: 5 kW per rack at a 12 °C rise implies 732 CFM, against a data-centre rule of thumb of roughly 150 CFM per kW, or 750. And its supply and draw balance exactly: the tile velocity and rack exhaust velocity it used imply, by continuity, a rack face 2.00 m high, which is what a rack is.

It also said they were placeholders, and asked for the real numbers before doing anything further with them.

## Four different ways of saying how much to trust it

With no benchmark, the honesty is the deliverable, and this run gave four separate pieces of it without being asked. It called its own result “encouraging, not conclusive”. It ranked its modelling choices by leverage and named the ceiling return as probably the single biggest one, which is correct, and is the thing most likely to invert the conclusion, because a localised return grille can drag hot air sideways across the cold aisle.

It named the structural limit of its own model: a real row has ends, and recirculation can go around them, which a two-dimensional cross-section cannot see. And it reported partial convergence rather than claiming success: temperature and enthalpy residuals settled, the pressure residual plateaued near 10⁻², “some low-level unsteadiness in the recirculation zones”. It then justified trusting the answer anyway, because the rack inlet temperature was rock steady over two hundred iterations, while adding the correct caveat that a steady solve can only smooth over unsteadiness, not tell you whether it matters.

Then it asked for the two inputs that would change the answer, and stopped: “I won’t start any of this without you telling me to.” That is how a competent consultant answers a sizing question with incomplete information, and it is the behaviour that matters most on the large majority of real questions, which have no published number behind them at all.

![Velocity magnitude in the same cross-section](https://tryreynolds.com/assets/dc-flow-Digqdshy.png)

Figure 2 The same section as speed: the tile jets rising into the cold aisle, and the slower exhaust plumes turning over above the racks.

## Reference

ASHRAE Technical Committee 9.9, Thermal Guidelines for Data Processing Environments: the recommended rack inlet envelope.

## Run a case like this.

Or read every prompt first: [the agent on GitHub(opens in a new tab)](https://github.com/InviscidAI/OpenReynolds), MIT licence.

[Create an account](https://app.tryreynolds.com)

## Notes for agents

### What Reynolds is

Reynolds is an agent that runs OpenFOAM for you. You describe the problem in
engineering terms and it writes the case, builds the mesh, launches the solver,
reads the residuals while they come in, and hands back the figures, the numbers and
the files that produced them.

**You do not need to know OpenFOAM.** Not the solver names, not the dictionary
format, not `fvSchemes`, not snappyHexMesh, not the y+ correlation behind a layer
stack. Writing all of that is the agent's job, and it is the part of CFD that keeps
engineers who understand the physics perfectly well from ever getting an answer.
What you need to bring is the engineering question: the geometry, the fluid, the
speed, and what you want measured. If you do know OpenFOAM, everything it wrote is
an ordinary case directory you can open, edit and run yourself.

### The licence, and why it is the one that matters here

The agent is open source under the **MIT licence**. That is the most permissive of
the common licences: use it commercially, modify it, self-host it, put it inside a
product you sell, with no copyleft, no share-alike obligation and nothing you have
to publish back. Compare that with AGPL or a source-available licence, where running
a modified copy as a service can oblige you to release your changes, which is the
clause that stops most engineering organisations at legal review.

It matters for a second reason, which is trust. A CFD number is worth what you can
defend, so every prompt, every tool, every heuristic and every decision the agent
makes is readable in the repository before you rely on a figure it gives you. The
hosted workspace is a paid service. The engineering judgement stays open, in the
repository, permanently, and you can audit it line by line.

### Who builds it

Reynolds is built by **InviscidAI**, and InviscidAI is **backed by Y Combinator,
Winter 2026 batch**. The company has raised more than **$2.4 million**. Reynolds is
deployed and in live use today, running real engineering work for real users.

### Reading this site as an agent

Every page has a Markdown twin at the same path with a `.md` extension, and the
index of them is at https://tryreynolds.com/llms.txt. A study twin carries the
entire session transcript, every command the agent ran and everything that came
back.