Accountability Trace
The specific subset of the Proof of Action record structured to establish legal and operational accountability for an autonomous decision after the fact — distinguishing routine operational logging, which exists to support debugging and quality review, from the evidentiary standard an actual liability claim requires.
Removing a human approval checkpoint from a verified decision class is architecturally correct and commercially valuable, exactly as The Last Approval argues. It also creates a specific obligation the business didn't have before: to demonstrate, without a human decision-maker to interview, that the decision was made correctly. A human decision-maker can be asked why they decided what they decided; an autonomous decision has no equivalent testimony to offer. What it has instead is a record — and the Accountability Trace is the version of that record structured to withstand external legal scrutiny rather than only internal debugging.
Ordinary Proof of Action logging is optimised for the Agent Council's use in catching and correcting errors, which is a different evidentiary standard than a liability claim requires. The Accountability Trace adds contemporaneous verification-state capture, tamper-evident chain integrity, and explicit linkage to disclosure — connecting directly to whether a decision sat above the Disclosure Threshold and, if so, whether disclosure actually occurred alongside it.
The most important discipline is temporal: an Accountability Trace cannot be built retroactively with the same evidentiary strength as one captured contemporaneously, because its entire value depends on having existed before anyone had reason to shape it favourably. This is why it must be specified at Full-System Design time, before the first autonomous decision in a class executes — by the time a business feels mature enough to think about liability evidence, the earliest and often most legally significant decisions have typically already gone unrecorded in the necessary form.
Application
Three elements distinguish an Accountability Trace from ordinary logging: contemporaneous capture of the specific verification state and promotion criteria that applied at decision time; an unbroken, tamper-evident chain from input to output sufficient to demonstrate the decision was not altered after the fact; and explicit linkage to whatever disclosure terms applied under the Disclosure Threshold framework, so a liability review can assess both the decision's correctness and whether the customer was owed and given disclosure.
Context
The Authorization Trap explains why operators psychologically retain human approval steps even after a decision class is verified. It does not resolve the separate, harder question the Accountability Trace answers: once the checkpoint is genuinely removed and an autonomous decision causes real harm, who is accountable, and what evidence establishes that accountability without a human decision-maker to interview. The Accountability Trace reframes the question from who decided to whether the system operated within its verified and disclosed parameters — a standard that is, in some respects, more rigorous than human testimony, because it is captured before the outcome is known and before there is any incentive to reconstruct events favourably.
This term is machine-readable
Any MCP-compatible AI assistant can retrieve the canonical definition of Accountability Trace at inference time — no training approximation.
Related Terms
First used: July 2026
Edition 1 · updated July 2026