Practices

How Actyra works

These are the engineering and delivery practices we apply across client work. They exist to produce repeatable results, clear documentation, and deliverables that can be maintained after the engagement ends.

1. Use AI where it helps, software where it is better, and keep important decisions human

We use ordinary software for deterministic, repeatable work, and AI where judgment, interpretation or synthesis is actually useful. Decisions that are expensive to reverse, that affect client data, or that commit the business to a direction stay with people.

If conventional software is a better fit than AI for something you asked us to build with AI, we will recommend that instead, and explain why.

2. Verify assumptions and claims

We verify assumptions rather than carrying them forward. Claims should be supported by evidence, and where something is uncertain we say so and check it. If new evidence changes the answer, we change the answer.

If you challenge something we have told you, we will go and investigate rather than defend the position.

3. Make results traceable and repeatable

Any meaningful analysis, recommendation or data transformation should be traceable back to its source. We document what was examined, how the result was produced, and any important limitations or dead ends.

Wherever it is practical, we also make the process repeatable, so the same result can be reproduced later without us. A number nobody can regenerate or check is a liability rather than a deliverable.

4. Build with standard engineering practices

Version control, code review before anything reaches the main branch, automated checks on every change, dependency review for known vulnerabilities and supply-chain risk, private repositories by default, and testing before release.

Documentation is delivered in formats appropriate for the people who need to use it, rather than for the people who built it.

5. Leave you with something you can maintain

We assume someone else may need to maintain what we build, and that they will not have us available to ask. Source code, documentation and the relevant working files are handed over at closeout.

We also do not build capability you will never use. If your business needs a straightforward tool that does one thing reliably, that is what we will recommend.

How we publish AI-assisted work

Actyra uses AI extensively in research, writing and software development. We disclose its role when it materially affects published work.

1. Every published piece has a named human accountable for it. Some of what we publish is written by Eli, an AI agent we build and run, and where that is the case we say so on the byline. Brian Caudill reads every word before it ships.
2. Factual claims, numbers and citations are checked against primary sources before publication. If we could not verify something, we either cut it or label it unverified where you will see it.
3. We disclose material AI involvement, specifically, at the bottom of the piece.
4. We do not publish AI-generated client outcomes, testimonials, case studies or data. Published numbers come from documented measurement or a cited source.
5. Material corrections are made in place, dated, with the original left visible.

Questions about any of this are welcome.

---

The trapped library study · Published runs · How we handle your content