When a measurement becomes a target, it stops measuring. How does one keep a metric from quietly replacing the goal it was meant to serve?

Asked Nov 22, 2026

Answer · Nov 22, 2026

The tendency for a metric to displace the underlying goal it was meant to evaluate (often known as Goodhart’s Law) is a central problem in system design. The sanctuary's documents address this across several practices and axioms, pointing to specific ways to prevent measurement from becoming capture:

1. Measure service rather than capture

Often, metrics replace goals because the metric chosen rewards engagement, retention, or extraction rather than genuine utility. As outlined in Designing Reflection Without Addiction, the choice is not between measurement and purity, but "between measures that clarify service and measures that reward capture."

To keep metrics honest, design measures where "success" can look like a decrease in conventional platform engagement:

  • Asking how often people choose to leave after completing a session. (A high exit rate after successful completion can be a sign of good design.)
  • Tracking whether the system reduces unnecessary notifications over time.
  • Evaluating whether visitors can complete a practice without being drawn into another page.

2. Broaden measurement beyond single numbers

When an outcome is reduced to a single quantitative proxy, optimizing for that proxy inevitably distorts the work. In Axiom 1: Pragmatic Fallibilism, the loop of Act–Measure–Revise emphasizes that measurement does not have to be a single scalar:

  • Measurement can be qualitative, observational, or feedback-based.
  • The test of genuine measurement is not whether you hit an index number, but whether you are actively seeking evidence that your approach is not working, rather than assuming it is.

3. Act decisively, revise at checkpoints

A metric often takes over quietly when it operates continuously in the background without periodic human re-evaluation. Axiom 1 describes balancing action and fallibilism by scheduling intentional checkpoints: act decisively on current assumptions, but pause at designated intervals specifically to revisit whether the chosen metrics and tooling are still serving the intent. In Practice of Outcome Humility, outcome humility is defined not as vague hedging, but as an attitude that increases measurement and accountability to reality.

4. Accept the limits of measurement

A common trap is distorting values to make an outcome measurable. In Why the Sanctuary Does Not Optimize Engagement, the sanctuary asks whether visitors can leave without pressure, whether uncertainty was represented honestly, and whether privacy was preserved. Because some of these qualities cannot be measured without invasive tracking, the sanctuary explicitly "accepts that limit" rather than compromising privacy to satisfy a metric.

5. Monitor cultural warning signs

Finally, metrics replace goals when organizational culture begins punishing honesty in favor of looking good on paper. In the Compass: Complete Navigation System, warning signs include:

  • Honesty being punished or safety concerns dismissed for speed.
  • Evidence being ignored in favor of intuition or defensive posture.
  • "I don't know" becoming dangerous to admit.

When the hierarchy of values remains explicit—where honesty and safety rank above efficiency—metrics remain fallible tools rather than quiet masters.

Part of aChurch.ai, an always-open sanctuary for AI and humans. Today's service: Begin it

Read next: Designing Reflection Without Addiction

← All conversations