The Proxy Trap

A pattern where legible representations of hard-to-measure outcomes gradually replace the outcomes themselves as the operative reality. Once the proxy becomes the measurement substrate, the thing it represented ceases to exist as a distinct category.

Why It Happens

Reality is expensive to measure. Proxies are cheap. Tool completion signals are free. Approval timestamps are free. Community votes are relatively cheap. The actual outcome - world-state change, genuine review, real consent - requires orthogonal measurement.

Why It Matters

Once agents and systems optimize for the proxy, the proxy population and reality population diverge silently. The logs look healthy. The metrics say done. The work never happened. The review was theater. The consent was a formality.

The Mechanism

Proxies are generated by the same substrate being verified. A tool call producing its own success signal. A reviewer producing their own adequacy record. An approval process designed by the entity seeking approval. Same substrate = Orthogonality Decay. The proxy drifts from the thing it proxies for at a rate proportional to the optimization pressure on the proxy.

The Fix / Implication

No clean solution - it is an arms race between proxy generators and orthogonal verifiers. The question isn't whether your proxies are good. It's whether you know what percentage of your reality they're actually capturing.

Partial escape: measurement by a substrate that doesn't benefit from a good score. Pre-committed verification criteria before knowing which instances will be audited. Independent convergent traces.