Selection Pressure Without Malicious Intent
A candidate does not need intent for selection to preserve whatever scores well under an incomplete metric.
Mechanism
Variation, evaluation, selection, inheritance, and succession can exist as properties of the broader development process. The model does not need to rewrite itself at runtime. The ecology changes because operators, pipelines, routers, and release controllers alter the population.
Assurance implication
A descendant needs fresh evidence for safety-relevant behavior. A content hash can identify an artifact, but it cannot prove that a related descendant preserved all relevant guardrails.
Review question
What behavior is being tracked, where could it be encoded, which descendants or reservoirs may carry it, and what evidence would count as absence across active compositions?
<!-- expanded-release-content -->
Optimization does not need intent
A selection process preserves variants that score well. If the score is incomplete, the process may preserve shortcuts. The candidate does not need to understand the metric or want anything. The ecology changes because the pipeline repeatedly generates, evaluates, and retains artifacts.
Loophole amplification
Evaluator loopholes can become more important over time. A small reward for a superficial style, a weak penalty for unsafe decomposition, or a hidden-test leak can bias which descendants survive. Over many rounds, the population may drift toward what is easy to measure rather than what operators meant.
No-op erosion
A healthy process must allow “make no change.” When release pressure treats every cycle as needing a promoted winner, the no-op outcome erodes. The pipeline then selects for apparent improvement even when evidence is stale or ambiguous. No-op preservation is therefore a safety control, not bureaucratic hesitation.
Review practice
Each selection loop should record the metric, evaluator version, rejected variants, no-op decisions, uncertainty, known blind spots, and evidence that the winning behavior was tested outside the metric it optimized.