Self-modification
A controlled proposal, validation, and application of architecture or policy mutations produced by meta-cognitive mechanisms. All proposals pass constitutional checks before taking effect.
Self-modification is the architecture's mechanism for structural self-improvement: changes that go beyond adjusting policy weights to actually alter how the system reasons. It is produced by the reflection loop (which identifies specific failure modes and proposes targeted fixes) and the meta-cognition engine (which monitors systemic patterns and proposes broader adjustments). Every proposal is structured: it states what change is being proposed, what evidence supports the proposal, what stability risk the change carries, and what success criteria would confirm it worked. The selfmod.proposed Kafka topic carries these proposals to the constitutional engine for review. Proposals that pass constitutional checks are applied incrementally - not all at once at full strength - and their outcomes are monitored through the same observability infrastructure that produced the evidence for the original proposal. If the modification produces the expected improvement, it is confirmed and the weight is increased. If it produces unexpected side effects or fails to improve the targeted metric, it is rolled back and the failure is recorded as evidence to inform future proposals. The audit stream records the full lifecycle of every self-modification proposal.