"Calibration & the Judgment Workbench: Making the Override Say Why"

quality-gate

3 min read

The Sensor layer, up close: when someone overrides the gate, what does the system make them write down — and why does that make the organization smarter?


The whole loop starts here, at the moment of judgment. When a gate fails and someone decides to ship anyway, the Sensor layer records a structured JudgmentCalibration. It’s not a checkbox. It asks for four things, and each one is a deliberate design choice.

What an override has to answer

A DecisionResponsibilityMatrix prevents decision compression — the very human tendency for one person to quietly occupy every role (architect, reviewer, override authority, final sign-off) on a quick fix. The matrix assigns distinct people to distinct responsibilities based on the risk tier.

The workbench

None of this works if calibration is painful. The Judgment Workbench makes it reachable: an inbox of the run’s advisory findings that you can acknowledge in place (acknowledging one writes the marker line at the flagged location — the override becomes a recorded, searchable fact, not a memory), and a calibrate wizard that walks the fields and writes the artifact through the corpus transport. Judgment becomes something a human can do at the keyboard, in the flow, without leaving the tool.

Justification, not approval

We first imagined an approval workflow for exemptions. In practice, what matters is the justification, not the sign-off. “This cluster match is expected — we’re mid-migration” is worth more than a rubber stamp, because the justification lets a future session judge whether the exemption still holds. That’s believability-weighting in the Dalio sense: input earns its weight through recorded reasoning, not seniority. Every override is a learning event — but only if it’s forced to explain itself.


Source: github.com/jpurnell/org-judgement-system


Tagged with: quality-gate, mirror, swift