Eval Drift is evaluation suites that no longer reflect real-world conditions. It sits in the Incentives dimension (INC) of the Human Behavior Taxonomy™ as element HBT-INC-0113, within the Workflow family. The core principle: evaluation suites that no longer reflect real-world conditions. In incentive terms, it matters because it changes the payoff people perceive before they choose — which means it can be designed for, or exploited.
Scientific Definition
Evaluation suites that no longer reflect real-world conditions.
Plain-English Definition
Evaluation suites that no longer reflect real-world conditions.
Feynman Explanation
Your benchmarks aged out of relevance.
Core Principle
Evaluation suites that no longer reflect real-world conditions.
Mechanisms
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Pending editorial review.
Inputs (Triggers)
Pending editorial review.
Outputs (Behaviors)
Pending editorial review.
Behavioral Signature
Your benchmarks aged out of relevance.
Examples
- Benchmark scores rising while real usage quality declines.
- Quality erosion masked by stale measurement.
Pending editorial review.
Original analysis from The Incentives Lab — how this element behaves inside real payoff structures.
Why this element matters to incentive design
Most organizations meet this element as a personnel problem. It is not one. The mechanism underneath it operates in the Incentives dimension — what makes behavior more or less likely?. You can recognize it in the field by its signature: your benchmarks aged out of relevance. Every element in the Incentives dimension changes the perceived payoff of an action before the action happens, which is exactly where incentive design has leverage.
How it gets exploited
Left undesigned, quality erosion masked by stale measurement. It is amplified whenever quality erosion masked by stale measurement. Inside organizations that shows up as quality erosion masked by stale measurement. The pattern is the same one Goodhart's Law describes: the measurable proxy attracts the effort, and the purpose behind it quietly loses funding.
How the Lab designs around it
The redesign move is to periodically refresh evals against current production data. Design against it the way you would design against a known failure mode — assume it will appear, and price the exploit before someone finds it.
Famous Experiments
Pending editorial review.
Design Principles
- Periodically refresh evals against current production data.
Measurement Approaches
Pending editorial review.
Evidence
Pending editorial review.
Pending editorial review.
The Perverse Incentive Lens™
How this behavior is exploited — and how to redesign around it.
- Periodically refresh evals against current production data.
Pending editorial review.
Pending editorial review.
Interactive Mini Network
Click any neighbor to re-center the graph and follow the threads of connection.
Knowledge Graph Neighbors
Auto-linked to the rest of the Human Behavior Taxonomy by family, domain, dimension, and shared keywords.
The relationship between inputs and outputs changes.
Underlying data distribution changes over time.
Performance degradation as real-world data shifts.
AI system that takes actions to achieve goals, often across tools.
Designing processes from scratch around AI capability.
How much input the model can process at once.
Vector representation of content for similarity and search.
Systematic testing of model quality, safety, and capability.
Models learning from examples in the prompt.
How much delay the user experience tolerates.
Train a smaller model to imitate a larger one.
Multiple specialized agents coordinating on tasks.
Where Eval Drift is cited in the corpus
Essays, field guides, and diagnostics from The Incentives Lab that apply this element.
- Field guideIncentives: definition, types, examples
The parent field guide for this element.
- ReferenceThe laws of incentives
Goodhart, Campbell, and the Cobra Effect.
- EssayAI Agents Inherit Your Incentives
How this element propagates into automated systems.
- ReferenceThe Periodic Table of Human Behavior
The full 1,267-element map this page belongs to.
- ReferenceThe incentive glossary
Definitions for every mental model, bias, and fallacy in the corpus.
Questions about Eval Drift
- What is Eval Drift?
- Eval Drift is evaluation suites that no longer reflect real-world conditions. It sits in the Incentives dimension (INC) of the Human Behavior Taxonomy™ as element HBT-INC-0113, within the Workflow family. The core principle: evaluation suites that no longer reflect real-world conditions. In incentive terms, it matters because it changes the payoff people perceive before they choose — which means it can be designed for, or exploited.
- What is an example of Eval Drift?
- Quality erosion masked by stale measurement. The Incentives Lab catalogs everyday, organizational, and historical instances of this element on its Human Behavior Taxonomy™ page (HBT-INC-0113).
- How is Eval Drift exploited?
- Quality erosion masked by stale measurement.
- How do you design around Eval Drift?
- Periodically refresh evals against current production data.
- Which behavioral dimension does Eval Drift belong to?
- Eval Drift is classified in the Incentives dimension (INC) of the Human Behavior Taxonomy™, family "Workflow", class "AI-Behavioral Coupling". Its permanent identifier is HBT-INC-0113 and its evidence grade is C.