Release Threshold

The verified boundary that determines whether the Innovation Reflex may self-build and ship a specific class of proposed enhancement autonomously, or must escalate it to the Steward for evaluation, framing, and approval — the product-development-specific expression of the same discipline the Intervention Threshold applies to operational decisions.

Release Threshold is the routing decision that follows every proposal Innovation Reflex generates. A synthesized proposal is not yet a shipped feature, and the current loop engineering discourse — mostly concerned with whether an agentic loop runs reliably and how much it costs — has not specified a precise threshold for the question that actually determines commercial risk: which proposals a business trusts itself to build without asking, and which ones genuinely need a person to weigh in first.

The boundary is calibrated the same way the Intervention Threshold is calibrated: by Task Tiers (T1 / T2 / T3), and by whether the proposal has a Deterministic Outcome that can be verified once built. A proposal optimizing a known pattern with a clear before-and-after metric is a reasonable self-build candidate. A proposal requiring genuine, novel judgment about what the business should become — not merely how it should run a fraction more efficiently — is precisely the case Bridge Operator already concedes AI alone cannot yet reliably resolve, and belongs on the escalation side regardless of how routine the underlying task class otherwise appears.

Before either path executes, a Model Quorum evaluates the proposal independently. Genuine disagreement among the quorum's models is itself informative: it indicates the kind of ambiguity that argues for escalation even when the proposal would otherwise qualify for self-build, while consistent agreement across independent evaluators strengthens the case for proceeding without a human.

Release Threshold must be specified before the first proposal is generated, not calibrated retroactively once several proposals have already shipped under an undefined boundary. A business without this specification in place either escalates every proposal — collapsing Innovation Reflex into an ordinary backlog a human still triages — or self-builds every proposal, including the genuinely novel ones that needed a Bridge Operator's judgment, and discovers the miscalibration only after a shipped feature turns out to have been the wrong call. The boundary should be validated continuously against the Escalation Rate for each proposal class, narrowed whenever self-built proposals show a pattern of requiring correction, and never extended to cover a genuine product pivot — a decision this body of work has consistently reserved for the Steward regardless of how any other threshold is set.

First used: July 2026

Edition 1 · updated July 2026