Slow-loop decisions
The operator library is a small, auditable set of structural actions selected only when local objective, evidence, and resource budget agree.
Each operator has a trigger, cost, and guardrail.
| Operator | Trigger | Costs | Guardrail |
|---|---|---|---|
| Split / Genesis / Wedge | Persistent high entropy, predictive confusion, bimodal evidence. | New active unit, parameter allocation, review burden. | Merge back if loss does not improve. |
| Merge | Redundant units with overlapping evidence and low disagreement. | Reference rewrites, audit revalidation. | Re-split if uncertainty rises. |
| Add | Projected utility pays for complexity. | Activation, graph expansion, latency, governance. | Retire if utilization remains low. |
| Retire | Low utility, high ambiguity, resource drain. | Fallback routing, cold storage migration. | Reactivate only if future novelty warrants. |
| No-op | No affordable edit improves the local objective enough. | Baseline maintenance only. | Growth restarts only after novelty or resource recovery. |
No-op is an active, selected, resource-conserving decision. It prevents structural clutter, runaway novelty, meaningless feature accumulation, and chaotic oscillation.
When no affordable edit improves local viability, the strongest engineering choice is to do nothing, log why, and wait for better evidence or recovered resources.
Deep route polish
The operator visual distinguishes split, merge, add, retire, and no-op as reviewable choices with different evidence requirements.
The operator library turns structural growth into a controlled decision space. No operator should be treated as inherently good. Each action must explain its trigger, expected gain, cost, uncertainty, and effect on future maintenance.
If two glyph candidates repeatedly collapse into the same concept, merge may reduce redundancy. If one glyph hides two stable use patterns, split may be justified. If neither has enough support, no-op preserves budget.
| Focus | What to inspect |
|---|---|
| Split | Separate a confused class when evidence shows two stable patterns. |
| Merge | Reduce duplicate structure when two paths carry the same role. |
| No-op | Prefer stability when value does not repay cost. |
Every operator remains an educational design pattern until trace logs and review outcomes support wider use.
