Purpose
Operators define how descendants are produced. Naming them clearly makes experiments reproducible and keeps risk reviews concrete.
Model-breeding operators
| Operator | Compatibility need | Risk tier |
|---|---|---|
| Fine-tune | approved base, data lineage, training recipe | medium |
| Adapter train | compatible base and adapter runtime | low to medium |
| Adapter fuse | shared base and compatible adapter format | medium |
| Distill | teacher evidence and target task definition | medium |
| Quantize | runtime support and calibration suite | low to medium |
| Prune | architecture support and regression suite | medium |
| Weight merge | architecture and tokenizer compatibility | high |
| Ensemble | shared behavioral contract | medium |
| Router mutate | stable capability ontology and replay stream | high |
Code-breeding operators
| Operator | Compatibility need | Risk tier |
|---|---|---|
| Prompt-template variant | fixed output schema and test cases | low |
| Routing-rule patch | replay tests and fallback path | medium |
| Evaluation-suite change | governance approval | high |
| Tensor-kernel change | numerical correctness and hardware tests | high |
| Cache-policy change | latency and correctness tests | medium |
| Permission-policy change | security approval | high |
Operator record
RECORD OperatorCatalogEntry
operator_id
name
domain: "model" OR "code" OR "policy"
risk_tier
allowed_parent_types
required_evidence
required_approvers
rollback_requirement
known_failure_modes
END RECORDReview rule
If an operator changes the evaluator, router, permissions, or runtime isolation, treat it as high risk even if the code change is small. Blast radius matters more than line count.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.
Core synthesisThe Four Fs of AI: Code Breeding, Model Breeding, and the Teleodynamic Convergence of Mutable Small-Model EcologiesConceptual synthesis · 80.5 KB
Core synthesisTeleodynamic Evolution of AI EcosystemsConceptual synthesis · 15.3 KB
Evolutionary AIDesigning the “Perfect” Evolutionary AI SystemEstablished and emerging methods · 46.5 KB