Decision Trace
Source asciidoc: `docs/principle/decision-trace.adoc`
Purpose
Decision Trace is a principle for making important decisions inspectable.
A decision trace does not assume that human beings are perfectly rational. It starts from the opposite premise: fear, reward, belonging, scarcity, status, fatigue, and hidden incentives may shape a decision before the explanation appears.
The trace exists so that a decision does not disappear inside a confident story.
Minimum trace
A minimum decision trace should preserve:
-
fact;
-
reason;
-
alternative;
-
doubt;
-
verification.
For institutional or AI-assisted decisions, the trace may also preserve:
-
source;
-
assumption;
-
incentive;
-
uncertainty;
-
review condition;
-
revision condition;
-
harm if wrong.
Fact
The fact layer records the known basis of the decision.
It should distinguish what was known from what was believed, inferred, assumed, predicted, or feared.
Reason
The reason layer records why the chosen path was selected.
A reason should not only sound coherent. It should show its connection to evidence, purpose, constraint, and expected consequence.
Alternative
The alternative layer records what other paths were considered.
Without alternatives, a decision can pretend to have been inevitable. A trace should preserve the roads not taken, at least in summary form.
Doubt
The doubt layer records what remained uncertain.
Doubt is not weakness. It is the preservation of contact with reality.
A mature decision can act while still admitting what it does not know.
Verification
The verification layer records what could later confirm, challenge, or revise the decision.
A decision that cannot be checked becomes a story. A decision that can be checked can become accountable.
Why this matters for AI
AI systems can produce fluent explanations quickly.
This is useful when the goal is to organize thought. It is dangerous when fluency is mistaken for truth.
A truth-oriented AI system should therefore help preserve the decision trace:
-
what is known;
-
what is assumed;
-
what source supports the claim;
-
what alternative explanations exist;
-
what uncertainty remains;
-
what would change the conclusion;
-
what human decision is still required.
AI should belong to the architecture of trace, not only to the architecture of answer.
Why this matters for contribution systems
A contribution-oriented system cannot rely only on final labels.
It must preserve why a contribution was recognized, why a record was disputed, why a fulfillment was accepted, why a signal changed, and what evidence supports the change.
Without trace, reputation becomes opaque.
With trace, reputation can remain contextual, explainable, and correctable.