The honest position
Matter Settle is new. There are no client case studies on this site because there are none to publish, and inventing some was never on the table — a fabricated logo wall is both a lie and, for the sort of client we want, an instant disqualification. So this page is about the two things we can be held to today: the engineering standard we work to, and the commercial terms that make a first engagement low-risk for you.
The engineering standard we work to
None of this is novel. It is simply the difference between a model in a notebook and a system somebody has to operate on a Tuesday morning.
- Tests that gate the merge, not tests that exist. Behaviour-sensitive paths get pinned by tests that fail loudly when the behaviour changes, so nothing drifts quietly into a release.
- Evaluation on data the model has never seen. Results are reported on held-out data, across time windows where time matters, and the unflattering number is reported alongside the good one.
- Costs and constraints modelled honestly. A result that ignores the real-world frictions of the domain — fees, latency, rate limits, bad rows at 3am — is a result about a world that doesn't exist.
- Pipelines built to run unattended. Checkpointing, retries, quality gates, and multi-source fallback, because the thing that breaks in production is rarely the model.
- Deployment and monitoring treated as part of the job. Shipping is not the end of the work; knowing when the system has quietly stopped being right is.
An example of that standard, which is not ours
When people ask what the above looks like in practice, we point at RLAlphaLabs, a live, public reinforcement-learning research project in Indian equity markets. It gates every merge behind a test suite, models the full set of Indian brokerage costs rather than assuming them away, validates walk-forward on held-out data, and publishes its weaker out-of-sample numbers instead of only its best ones.
What we put in writing
These are commitments, not history. They cost you nothing to verify, because you find out whether we keep them during the first small engagement rather than after a large one.
- A written scope before any money moves — deliverable, price, and an agreed definition of "done". If we think AI is the wrong tool for your problem, we say so at this stage and you owe us nothing.
- Fixed price, fixed scope pilot — one clear deliverable, agreed up front, so the downside is known before you start.
- Your code and your infrastructure from day one — commits land in your repository and run in your accounts. Nothing to extract later, no lock-in to negotiate out of.
- Something runnable every week — not a status deck. You can execute it, read the tests, and form your own view.
- A pilot you can walk away from — if it doesn't convince you, you keep everything produced and the engagement ends there.
- Direct access to the engineer doing the work — no account layer, no handoff to a junior bench you were never shown.
What we won't claim
No client logos, no testimonials, no delivered-project count, no years-in-business figure, and no borrowed metrics from projects that aren't ours. When any of that becomes real, it will appear here with a link you can check. Until then, its absence is the most accurate thing this site can tell you.