PMFA
← Return to Index
036

Metrics Are Incentives

Metrics are not measurements. They are incentives that shape behavior. What you measure you govern.
Version 1.0.0 — Ratified

Metrics do not measure behavior. Metrics create behavior.

Whatever a system measures, it teaches actors how to win.

I. The Illusion of Neutral Measurement

Metrics are often treated as:

objective indicators

passive observations

health signals

reporting artifacts

This is false.

Metrics define:

what is rewarded

what is ignored

what is punished

what is optimized away

That is incentive design.

II. Measurement Changes Reality

Once a metric exists:

actors adapt to it

strategies shift toward it

effort concentrates around it

Unmeasured work disappears. Measured work dominates.

This is not corruption. This is rational behavior.

III. Goodhart’s Law Is Governance Failure

“When a measure becomes a target, it ceases to be a good measure.”

This is not a paradox. It is a warning.

Metrics fail when:

incentives are misaligned

proxies replace purpose

numbers outrank outcomes

This is policy failure.

IV. Metrics Can Reward Harm

Poorly chosen metrics can:

reward speed over correctness

reward volume over legitimacy

reward uptime over fairness

reward success over compliance

Harmful behavior that improves metrics will spread.

Metrics enforce what they reward.

V. Metrics Must Reflect Authority, Not Convenience

A lawful metric answers:

did the system act lawfully?

did authority apply correctly?

did policy hold under pressure?

Convenience metrics answer:

was it fast?

was it cheap?

was it green on a dashboard?

The second set is insufficient.

VI. Metrics Must Be Explicitly Declared as Incentives

A lawful system declares:

why a metric exists

what behavior it incentivizes

what behavior it discourages

what trade-offs it accepts

Hidden incentives are manipulation.

VII. Metrics Must Be Paired with Constraints

Metrics without constraints:

drive exploitation

encourage shortcuts

reward rule-breaking

Constraints restore balance.

A metric without a constraint is a loophole generator.

VIII. Replay Requires Metric Stability

Replay must reproduce:

the same metrics

the same thresholds

the same evaluations

the same consequences

If metrics change retroactively, history is rewritten.

IX. Infrastructure Metrics Are Not Enough

CPU usage, latency, error rates do not measure governance.

They measure mechanics.

Incentive metrics must live in:

policy evaluation

kernel outcomes

compliance signals

Not only in ops dashboards.

X. Final Conclusion

Metrics are incentives.

A lawful system:

treats metrics as behavior-shaping tools

aligns metrics with authority and policy

declares incentive intent explicitly

constrains metric optimization

replays metric evaluation deterministically

Anything else trains actors to succeed numerically while failing systemically.

Canonical text. Interpretations are invalid.
SHA-256: 0d9a27269badad725de9285244e105f4f0dcf8c700c7563cae2976de99d3fa0d