← All cases · System builds system · in operation · since August 2026
Trace instead of gut feeling
Measurable effort instead of gut feeling — the baseline showed zero percent navigation: the agents knew the paths from context and did not follow the knowledge graph.
Figure: 0 % Navigations-Rate in der Ausgangsmessung (vier Sitzungen; je zur Hälfte Suche und Direktzugriff) · measured 28 August 2026
Request: „Trace on — and show me whether the agents follow the knowledge graph or just search.“
- Every brain access is classified: navigation, search, auto-loaded, direct
- Sub-agents are linked via start and completion; tool calls, tokens, runtime and skills per session are recorded
- A report shows navigation rate and missing routes — deterministically, without a model run
- Missing links are added and measured again
In most organisations nobody knows whether agents follow a knowledge graph or just run full-text search; tokens and runtime per task stay invisible.
Interpret the findings, improve the structure.
The baseline showed that the agents did not navigate the knowledge graph at all — without measuring we would have believed the linking worked.
Any organisation running AI agents in production. Steering AI means seeing what it does.
Steering AI means seeing what it does. A trace shows for every access whether an agent followed the knowledge graph, searched or went direct — and what that cost in tokens and time. The first measurement across four sessions was sobering: zero percent navigation, half search, half direct access. That is exactly the value: without measuring we would have believed the linking worked. Now missing routes can be added deliberately and re-measured — deterministically, without a model run.
Evidence in my own archive · as of 25 October 2026.