An output is not an artifact.
VerificationTests, independent review, and the difference between saying it works and showing that it does.
Research.
How we think.
What we check.
Engineering notes on intelligent software, evaluation, and the boundaries that make a result worth trusting. These are methods and positions, not a list of claimed breakthroughs.
Notes from the lab.
Read the method. Inspect the boundary.
Replay the evidence.
Deterministic replayWhy observation time, availability time, and the decision cutoff belong in different fields.
Keep human approval explicit.
Agent systemsA proposal, a confirmed assumption, and an authorized action are different system states.
Open questions.
The next useful question is usually more specific.
- Deterministic agent workflows
- Adversarial testing and AI security
- Engineering harnesses
- Evidence and provenance
- Automated gameplay
- Generated-software evaluation
These are areas of interest for future notes and experiments. Completed work will be published with its method, evidence, result, and limitations.