Skip to content
Stance

AI testing
on real Macs.

We test your ML and AI apps and models on Macs.
You get clear findings, evidence and recommendations.

Discuss your project Checked by a human.
  • Independent
  • Hands-on
  • Mac first
Illustrative example

Q: Convert 12 cm to inches.

A: 12 inches.

Explanation: 1 cm is approximately
equal to 1 inch.

No.

Small mistakes.
Real consequences.

Synthetic example
The model returned

quantity: "12"

A dimension became an item count. The order reference was also lost.

A dimension is not an item count.

Findings your team can act on.

What happened, why it matters, and what to do next.
A useful assessment makes the next decision clearer.

Example assessment / synthetic inputs

Finding 01 / Incorrect quantity extraction

Possible impact
An app trusting these fields could take an incorrect action. This is a potential consequence, not an observed customer incident.
Evidence
The model returned the string "12" as a quantity when the input described a 12 cm dimension. The order reference was also lost.
Recommendation
Validate types and units. Test measurements separately from item counts, and verify changes against fresh cases.

Two ways to start

Bring the app.
Or the model.

Working applications

Real workflows, AI behavior,
responsiveness and reliability.

Clear findings, evidence
and recommendations.

Standalone models

Task quality, failure cases
and performance on Mac.

Understand where it works,
and where it falls short.

Find a useful scope for your project

The testing work is ours.

  1. 1

    Agree on
    the scope

    You share access and intended behavior. We clarify the questions and what success looks like.

  2. 2

    We test and
    investigate

    On real Macs, in realistic conditions. We document what happens, and investigate why.

  3. 3

    Receive findings
    and a walkthrough

    You get clear evidence, prioritized recommendations and a conversation about what it means.

Testing and recommendations.
Fixes and optimization are outside the initial engagement.

Independent. Hands-on.

A fresh pair of eyes.
One point of contact.

Stance is an independent testing service led by Vaibhav. From the first question to the final walkthrough, there is one person accountable for the work.

We take on a clearly defined assessment, investigate the details, and tell you what we found. Your team gets to focus on building.

Before we start

What do you need from us?

Access to your application or model, what it should do, and the priorities that matter to you. We agree on the testing scope and a suitable access arrangement together.

Do we need to prepare test cases?

No complete test suite required. We design the tests around your goals and clarify what correct behavior looks like with your team.

Will you fix the issues?

The initial engagement covers testing, evidence and recommendations. Your team implements the changes; fixes, training and optimization are outside this scope.

How long does an assessment take?

That depends on the app, the model and the questions you need answered. We agree on scope and timing before testing begins.

Hand over the testing.

Tell us what you are building. We will agree on a useful scope.

What are you providing?

Enquiries are not sent or saved. Submission is not connected yet.

Enquiries are not connected yet. This form does not send or save your details.

Stance

Independent ML and AI testing on Mac.

We take it
from here.
STANCE / EXAMPLE ASSESSMENT

Synthetic inputs. Measured locally. Not a customer report.

A dimension became
an item count.

This example shows the kind of evidence an assessment can contain. It comes from a small local model experiment, not a production application.

01 / The task

Extract an order reference, requested action and item count from a support message.

The label says 12 cm. Please cancel H-808.

Expected

{
  "order_id": "H-808",
  "intent": "cancel",
  "quantity": null
}

Observed

{
  "order_id": null,
  "intent": "cancel",
  "quantity": "12"
}

02 / What we observed

  • The order reference was lost.
  • A dimension was extracted as an item count.
  • The quantity was a string, not the required integer or null.

03 / Why it matters

An application trusting these fields could take an incorrect downstream action. This is a possible impact, not an observed customer incident.

04 / Recommendations

Validate output types and units. Add regression cases that distinguish measurements from counts and preserve references. Assess any model or prompt changes against a fresh held-out set.

05 / Experiment context

Hardware
Apple M5, 16 GB, macOS 26.0
Model
Qwen3 1.7B, MLX 4-bit
Runtime
MLX 0.28.0 / MLX-LM 0.26.3
Test set
15 synthetic cases, repeated 3 times
Overall correct outputs
34 / 45 (75.6%)
Median response time
0.489 seconds
Peak MLX allocation
1.434 GB; not total process memory

A small synthetic set cannot establish production reliability. Repetitions are not independent examples. No iPhone, battery or thermal measurements are claimed.

Model and test provenance
Model revision
21457c6f51ed54a7c16e988c0844db973815c137
Official chat-template revision
70d244cc86ccca08cf5af4e1e306ecf908b1ad5e
Run identifier / case
20260905T204632308820Z / test-05