Execuria · Measures
Measurement notes
These are practice notes on measurement: how we judge whether a collaboration model between people and agents is actually holding, long after the design work is done. Measurement here is not a dashboard of activity. It is a small, honest set of questions we ask on a cadence and answer in writing.
What we measure, and what we refuse to
We refuse to treat volume as success. The number of tasks completed by agents tells you almost nothing about whether the ensemble is in time. A team can produce a great deal of output while drifting steadily out of alignment, and the volume will look excellent right up to the moment it does not.
Instead we measure four things: whether hand-offs are clean, whether supervision is real, whether escalations flow, and whether the model still matches the work. Each is answered by observation rather than by a metric that can be gamed.
Are hand-offs clean?
Work arrives at boundaries in the agreed state, and returns without silent repair. We sample the boundary and count quiet fixes.
Is supervision real?
Reviewers can describe the failure modes they watch for, and traces exist for the reviews they claim to have done.
Do escalations flow?
Doubts are raised through the agreed path rather than absorbed in private. A healthy register has entries.
Does the model match?
The written model still describes the work teams recognise. Where it does not, it is revised rather than ignored.
The cadence
Measurement runs on a light cadence. Weekly, the section lead notes any silent repairs and any escalations. Monthly, a reviewer reads a sample end to end and records what the sample revealed. Quarterly, we sit with the model owner, re-read the model against reality, and produce a short written revision note.
The quarterly note is the centrepiece. It is deliberately short — a page, not a report — because a document that is too long to read on a busy Friday will not be read at all.
A drifting ensemble looks busy right up to the moment it does not. Measure alignment, not effort.
Signals of drift
Drift rarely arrives as an incident. It arrives as a pattern of small accommodations. A reviewer who starts approving faster. A boundary that quietly moves because a system changed. An escalation path that goes unused for a quarter. A role description that no longer matches what the role actually does. Any one of these is survivable. Several together mean the model is no longer guiding the work.
| Signal | What it usually means | First response |
|---|---|---|
| Silent repairs rising | Hand-off state was wrong, or the role limit is unclear | Re-read the boundary and restate the expected state |
| Register empty | Escalation path is not trusted or not known | Re-open the path and make pausing blameless |
| Reviews speeding up | Reviewer fatigue or rubber-stamping | Rotate reviewers and widen sampling briefly |
| Role description stale | The work changed and the model did not | Revise the atlas at the next quarterly note |
Reporting to leadership
Leadership does not need our raw notes. It needs three things: whether the model is holding, where it is under strain, and what we recommend changing next. We deliver that in a single page, with the evidence behind it available on request. We are explicit about uncertainty, and we never present a healthy sample as proof that the whole ensemble is sound.
Measurement is part of Ongoing Oversight. It is a review practice, not a scoring system, and it promises no particular result. Read it alongside agent oversight and the collaboration model.
For the working notes behind these measures, see practice notes. For how we report during an engagement, see engagement tiers.