Metrics tend to change behavior once they are observed, discussed, or rewarded. When this happens without care, the metric begins to reflect optimization effort rather than underlying reality.
This is not usually malicious. It is a natural response to incentives and attention.
SHAM acknowledges this dynamic and is designed to minimize it, but no framework is immune if misapplied.
A common failure mode occurs when a metric shifts from being a signal to being a goal.
Once a metric becomes a target:
- Behavior adapts to improve the number
- The original meaning erodes
- The metric loses explanatory power
This dynamic is often summarized as:
When a measure becomes a target, it stops measuring.
Metrics are often gamed in subtle ways, including:
- Reducing visible change without reducing actual risk
- Splitting work to minimize apparent change surface
- Deferring failure reporting
- Accelerating releases to shorten exposure windows artificially
- Shifting work outside observable systems
These behaviors can improve reported numbers while degrading system health.
Highly precise metrics create an illusion of objectivity and encourage fine-grained optimization.
When people believe they are being evaluated numerically, effort shifts toward:
- Improving measurements
- Controlling appearance
- Avoiding recorded risk
SHAM intentionally favors estimation and aggregation to reduce this incentive.
Metrics are most easily gamed when stripped of context.
Examples include:
- Comparing failure density across unrelated work
- Treating exploratory instability as poor performance
- Ignoring constraint load when interpreting recovery time
Context collapse often precedes misuse.
When metrics are reviewed without system ownership or domain knowledge, interpretation tends to drift toward simplification.
This increases the likelihood of:
- Overreaction
- Blame assignment
- Process hardening without understanding
SHAM assumes interpretation occurs among people who are responsible for the system.
SHAM reduces gaming pressure by:
- Avoiding individual attribution
- Using rolling windows instead of fixed periods
- Emphasizing relationships over thresholds
- Requiring human interpretation
- Making precision intentionally approximate
These design choices trade short-term clarity for long-term signal integrity.
No metric framework can function without trust.
When trust erodes, metrics become defensive tools rather than shared signals. SHAM is most effective in environments where metrics are used to understand systems, not to rank or control people.
Metrics should invite curiosity, not compliance.