Affect-Modulated Promotion Thresholds
by Nick Clark | Published March 27, 2026
Affective state raising or lowering the minimum score required for candidate mutations to advance through evaluation stages, producing experience-driven selectivity.
What It Is
Affect-modulated promotion thresholds adjust the minimum score required for candidate mutations to advance through evaluation stages. When an agent's affective state reflects caution, such as elevated risk sensitivity or reduced ambiguity tolerance, the threshold rises, requiring stronger evidence for promotion. When affect reflects openness, the threshold lowers, admitting candidates that would otherwise be filtered.
This modulation applies at every promotion boundary in the agent's evaluation pipeline, from planning graph branch promotion to verified memory commitment. The threshold adjustment is computed deterministically from the current affective field values.
Why It Matters
Fixed promotion thresholds produce agents that are equally permissive in all circumstances. An agent recovering from a failed delegation should not evaluate new delegation candidates with the same openness as one with a history of successful delegations. Affect-modulated thresholds create experience-sensitive selectivity.
This is architecturally distinct from confidence governance. Confidence determines whether the agent can act at all. Affect-modulated thresholds determine how selectively the agent evaluates what to act on. An agent can be confident enough to act but affectively cautious about which specific candidates it promotes.
How It Works Structurally
The promotion threshold function takes the base threshold from policy and applies a modifier computed from relevant affective dimensions. Risk sensitivity and uncertainty sensitivity increase the modifier, raising the bar. Novelty appetite and cooperation disposition may decrease it. The specific weighting of dimensions is policy-configurable.
The modulated threshold is recomputed at each evaluation cycle, meaning it can change between successive candidate evaluations as affective state updates arrive. All threshold computations are recorded in lineage, making it possible to audit why a specific candidate was or was not promoted at a specific time.
What It Enables
Systems where agents become naturally more cautious after adverse experiences and more exploratory after positive outcomes, without explicit programming of these behavioral shifts. The caution emerges from the architecture rather than being imposed by external rules.
Multi-agent coordination benefits because delegation requests from agents with elevated caution thresholds signal that the requesting agent has reason for careful evaluation, allowing receiving agents to adjust their own behavior accordingly.
