What happens when an LLM can ask governed questions of an enterprise?
EKA collects evidence from several kinds of enterprise knowledge and explains what it can — and cannot — establish.
This guide is a walkthrough of the real EKA implementation. The five worked examples are canonical captures from live runs on one clean build, not reconstructed demos.
What exists inside the enterprise?
How was this synthetic enterprise constructed, and why?
This guide
What happens when an LLM can ask governed questions of it?
All five canonical runs were generated from live EKA execution on one clean build, under one prompt bundle and one set of contracts.
- Build
- 093d195bbb159084a629310d920875e4c513a6de
- Prompt bundle
- PROMPT-EKA-0.4.6
- Planner fingerprint
- sha256:0fb08227cf1e04d7a…
- Synthesis fingerprint
- sha256:2e56d7b0237722db3…
- EKA contract
- EKA-0.4.1
- QSR contract
- QSR-0.4.1
- Tools contract
- TOOLS-0.4
- Authority contract
- AUTH-0.4
- QSR effective hash
- sha256:c306963db7e1ec109…
- Tool-catalog hash
- sha256:4ff41b1fb43670530…
- Model
- google:gemini-3.5-flash-lite, pinned
- Business now
- 2026-07-31 18:30:00+00
Every capture ran on google:gemini-3.5-flash-lite, pinned rather than chosen by the model ladder, and each case produced its intended result on its first attempt. Four earlier attempts at case 03 ran against an empty document index; the evidence pack keeps them, separately, as setup errors rather than model failures.
What is still to come
The written guide no longer depends on missing run evidence. Future enrichment is visual: UI screenshots after the Stage 10 UI, short 30–90 second run recordings after a stable internal UI, and optional longer demos later. Those slots are designed but not shown yet; the recordings will reuse these same canonical questions rather than inventing prettier showcase scenarios.