What it can show

Termyte can connect derived memories to observations and traces. It can record which context items were shown, preserve task evidence, and store explicit feedback such as helpful, harmful, ignored, or corrected.

That gives you a trail to inspect. You can ask what was captured, what was derived, what was selected, and what feedback was recorded.

What it cannot show yet

A recorded context item is not proof that it caused a successful change. A memory marked helpful is feedback, not a controlled experiment. Retrieval ranking has deterministic bounds, but it is not calibrated on a public coding-agent corpus.

The current product also does not provide comprehensive redaction, sandbox isolation, or a complete task-management interface in the viewer.

Why the distinction matters

Useful infrastructure should make its evidence visible without turning that evidence into a marketing claim. Termyte is built around that distinction: preserve the work, expose the links, and keep the remaining uncertainty explicit.