What Are Decision Traces?

What is a decision trace?

A decision trace is the record behind a choice: the evidence, assumptions, human input, authorized action and later result. It is not a recording of a model's private thoughts. Teams can inspect what was known, what was decided and what happened next.

Nodes Engine retains this history as part of the company’s context graph. Teams can review decisions, understand exceptions and apply relevant lessons to later work. Current permissions and approval policies govern each action.

What a Decision Trace records

Evidence: the permitted records used across the relevant systems.
Versions and policy: the model, policy, and workflow versions in force.
Recommendation: the proposed response and its supported economic assumptions, including a request for evidence or a reason to wait.
Human input: the named reviewer's approval, edits, decline, and rationale.
Approved action: what executed within permitted systems and applicable approval rules.
Later outcome: the defined result when it becomes available.

The value is the chain. A reviewer can inspect what the system knew, what it proposed, what the human changed, and what happened later.

How do we check why AI recommended something?

Inspect the sources, applicable rules, assumptions, alternatives and uncertainty behind the recommendation. A confident explanation is not a substitute for evidence you can check. The worked record below shows what those checks can look like.

A worked Decision Trace

Show the illustrative Decision Trace

Illustrative example: A customer has not completed onboarding. The figures and records below are fictional; this is a sample of what a reviewable record should contain.

Part of the record	What the team can inspect
Question and owner	Why is this customer's rollout stalled? The onboarding lead owns the decision.
Sources and dates	The CRM shows the agreed start date; a support case opened three days ago describes a missing document; the current onboarding rule lists the required version. Each source keeps its date and access rule.
Assumption and contrary evidence	The proposed blocker is a document mismatch. A separate note says the customer already sent a file, so the team checks its version before asking again.
Alternatives	Request the correct document, investigate the existing file, wait for a scheduled review, or leave the process unchanged.
Human judgment and authority	The lead chooses to check the existing file first. The approved plan permits a follow-up task but does not permit changing contract terms.
Action confirmation	The task is created in the authorized system and its identifier is recorded. A draft message alone would not confirm execution.
Later observation	Record whether the customer reaches usable rollout, how much rework occurred and when this was checked. A favorable result still needs evaluation before becoming a reusable lesson.

The trace is an inspectable business record. It does not expose a model's private chain of thought or prove that the choice was correct. Review how decisions become later outcome evidence and how teams preserve the judgment behind them.

Why the trace includes more than a prediction

A prediction estimates a future result. A Decision Trace preserves the full operating record around that estimate, including the evidence boundary, the recommendation, the human decision, the action, and the later outcome. That record lets teams review a decision without assuming that the prediction or the human response was correct.

What you can ask a decision trace

A Decision Trace makes questions across systems inspectable. For example: show which evidence supported a recommendation, who approved or edited it, what action followed and how the defined outcome compared later.

Where decision traces have been studied

The production evidence comes from one Fortune 500 insurance carrier. Across 10,765 agents, a linked-record audit tested 3,597 resume keywords against a production milestone. None predicted that milestone after Bonferroni correction, and 30 were anti-predictive. The result shows why the evidence and outcome need to remain attached to the decision record. Read the full research.

Frequently asked questions

Is a decision trace a prediction? A prediction may appear inside a Decision Trace. The trace also records how that prediction was produced, what a named human decided, what action followed, and what happened later.

What data is needed to build one? The permitted records needed for the defined decision and outcome. Sources vary by workflow and are joined under the customer's access and governance rules. In hiring, they may include CRM, HRIS, ATS and assessment records.

Who uses decision traces? Business owners, operations, finance, risk, security and audit teams use Decision Traces to inspect the evidence, human judgment, approved action and later outcome behind a workflow.