Measure of Performance (MOP)

Did the operational number move? — activity and proxy signals; necessary but not sufficient for aim success

MARKO ŽIVKOVIĆ In review

Measure of performance (MOP) asks whether the operational number moved — whether the mechanism, process, or subsystem hit the performance parameters you can observe directly: throughput, latency, checklist completion, tickets closed, benchmark score, deployment cadence.

Program evaluation and defence systems analysis pair MOP with measure of effectiveness (MOE): MOP is how the engine ran; MOE is whether the trip reached the destination. In this corpus MOP is legitimate and necessary — most (A)DR sensors and comparators are MOP — but toxic when mistaken for MOE under reward or mandatory actuators.

MOP vs MOE

	MOP (performance)	MOE (effectiveness)
Question	Did the operational metric move?	Did the aim move?
Examples	SLO error rate, story points, audit pass rate, model benchmark, bounty payouts	Customer harm, fraud loss, replication rate, strategic objective, decision hypothesis outcome
Loop fit	Fast, automatable — natural sensor/comparator	Slower, messier — effectiveness proof
Goodhart pressure	High when wired to incentives	Lower when aim-linked — still gameable if naive
Corpus rule	Label as proxy; multi-signal; meta-loop	Name at bind; prefer for high-stakes “success”

Closed-loop control requires MOP-class observables — you cannot close a loop without measurable signals. The failure is not MOP; it is closing the loop on MOP alone while effectiveness drifts — the pattern behind Goodhart’s Law, metric substitution, and Cobra effect.

In (A)DR hypothesis tables

Typical row structure:

Element	MOP role
Sensor / observable	Usually MOP — what instrumentation watches
Comparator / trigger	Often MOP threshold — early falsification of operational assumption
Actuator	May fire on MOP — but must not read as proof the decision aim succeeded
Meta-loop	Override rate, contest overturn, incidents despite green MOP — MOE-class checks on proxy health

Meta-loop control treats every declared MOP as proxy under pressure — tiered, contested, multi-signal. Double-loop learning revises the MOP definition when the proxy decouples from MOE.

MOE vs MOP in incentives

Misaligned incentives almost always pay MOP:

Domain	Common MOP	Risk if treated as MOE
Software	Tickets closed, velocity, coverage %	Quality and customer harm drift
Security	Findings count, scan pass	Real exposure rises (Cobra effect)
Research	Publications, benchmark delta	Publication over provenance
Operations	Uptime, cost per unit	Correctness and safety tradeoffs hidden

Counter: separate scoreboards — reward and promote on MOE movement where instrumented; use MOP for cadence and early warning, not sole success narrative. Track record and Brier scores are MOE-class when tied to outcomes, not activity volume.

Origin vocabulary

A2 — working context: MOP/MOE pairing is standard in program evaluation and defence systems analysis — performance parameters vs mission effectiveness. Manuals disagree on edge cases; this corpus uses the pair as proxy hygiene, not as import of any single agency glossary.

Corpus stance

A2 — working context: Declare MOP observables explicitly as proxies in ADR and closed-loop binds; pair every high-stakes MOP with MOE intent (measured or sampled); never reward “comparator did not fire” or “MOP green” as organisational success without effectiveness evidence.