How Amara actually decided
Rules derived from eleven years of decisions, not from anything Amara wrote down. Every finding carries the statistic it rests on and the exhibits behind it, so you can read the evidence and disagree with it.
The risk appetite that describes nothing
Stated vs revealedKestrel's stated appetite is a 2% loss ceiling. It appears in every board deck. Across the decision record, no decision was ever made at it — the real behaviour splits into two regimes with nothing in between. The published figure is the average of two different policies, and describes neither.
Ask the record
Answers come from the doctrine and the source documents. Where Amara never faced something, the agent says so and declines to invent a verdict — that refusal is the point.
Is this you?
These rules were derived from your decisions, not from anything you wrote down. Nobody else can say whether they are true. Mark each one — and be honest about the ones you did not know you had, because those are the ones your successor cannot get any other way.
Consent & access
This system models a named person's judgement from their private conversations. It runs only while they permit it, excludes categories they never agreed to share, and records every use where they can see it.
Consent of record
Excluded from scope
Applied to source text on the way out, so it holds regardless of what an agent asks for or how the question is phrased.
Policy decisions
Ask the policy engine
Deny by default. If the policy store cannot be reached, the answer is deny — an outage must never widen access.
Right to erasure
Erase the induced model
Deleting the source documents would not be enough. What Tacit holds is what it inferred across hundreds of conversations, and those conclusions survive the deletion of any single transcript. Erasure here destroys the derived model itself.
Chain of custody
| When | Who | Question | Exhibits | |
|---|---|---|---|---|
| No access recorded yet. | ||||