Decision-Time Memory Architecture
The minimum structure required
so “why” stays computable later.
This is not a product spec. It’s the smallest set of primitives that prevents decision intent from evaporating after the system commits—and stops audits from turning into narrative reconstruction.
Core idea
A decision is a first-class event, not an after-the-fact story.
If intent is captured at commitment, the system can later answer: what was known, what was assumed, what alternatives were rejected, and what constraints were binding— without guessing and without blame.
Decision Object (Conceptual)
structure > implementation
Identity
Stable ID + Timestamp
A decision must be uniquely referencable and anchored in time.
Authority
Actor + Scope
Not “who typed it,” but who had the right to commit—and under what scope.
Intent
Goal + Success Condition
What we were trying to accomplish, stated in a way that can be checked later.
Context
Constraints + Tradeoffs
What had to remain true, and what we knowingly sacrificed to proceed.
Alternatives
Considered + Rejected
The paths that were available at the time—and why they weren’t chosen.
This is what makes counterfactuals computable later.
Linkage
Commit Reference + Evidence References
The decision links forward to what was committed (change/version/action) and
links backward to supporting signals (events, metrics, incidents, tickets) available at decision-time.
The point is traceability: decisions remain linkable to evidence and outcomes without narrative reconstruction.
Adoption rule
No surveillance posture
If this feels like snitching, adoption dies. The system must optimize for future clarity, not retroactive blame.
Integrity rule
Decision-time only
The record is strongest when captured at commitment, before outcomes rewrite human memory.
Audit rule
Claims must cite evidence
Explanations should point to linked signals and committed artifacts—not “what we remember.”
Back to the overview →
Decision-Time Memory (Overview)