The principles
Five requirements
One thread of trust
A convincing answer is not enough. To remain useful over time, intelligence must show where information came from, what was retained, who authorised an action and what evidence supports it.
Guiding principle
AI can produce an answer
Auditoris must be accountable for it
01
Provenance
Information without an origin is not reliable knowledge.
Auditoris preserves the link between information and its source: document, person, system, date, version and context. Transformations and interpretations must never erase that link.
What this requires
- Identify the source and when it was observed.
- Preserve relevant versions, transformations and evidence.
- Flag anything incomplete, indirect or impossible to verify.
02
Memory
Auditoris memory belongs neither to an AI nor to a conversation.
It connects knowledge, decisions, reservations and evidence over time. A subject can be resumed without reconstructing the past or confusing what was true yesterday with what applies today.
What this requires
- Preserve context, relationships and chronology.
- Separate current state from historical or superseded material.
- Keep memories separated by person and project.
03
Governance
A technical capability is never a permission.
Auditoris determines who may access what, in which project, under which rules and within which limits. Boundaries remain explicit, and final human authority cannot be absorbed by automation.
What this requires
- Enforce permissions in the system, not only in the interface.
- Isolate each project’s data and rules.
- Retain identifiable authority for every sensitive decision.
04
Discernment
Understanding begins by refusing to confuse.
An observed fact, a statement, a hypothesis, an inference, a decision and a reservation do not have the same status. Auditoris qualifies them so the confidence of an answer never conceals its real level of evidence.
What this requires
- Separate facts, hypotheses and interpretations explicitly.
- Make contradictions, uncertainty and reservations visible.
- Revise conclusions when better evidence appears.
05
Action
Action turns an authorised decision into a proven result.
Before execution, Auditoris checks scope, authority, conditions and reversibility. Afterwards, it seeks observable evidence. When ambiguity remains, it stops instead of claiming success.
What this requires
- Prepare backups, rollback and stop criteria.
- Execute a bounded, traceable and reversible action.
- Verify the actual result before declaring success.