COPRO (EP-21): coordinate-ascent prompt optimization. Where instructionSearch is one-shot per node, COPRO improves each node's instruction over several rounds (depth), proposing several candidates per round (breadth) and feeding the scored attempt history forward so later rounds learn from what scored well. It is the principled generalization of instructionSearch (depth-1, no-history COPRO ≈ instructionSearch), kept alongside it rather than replacing it.

COPRO consumes EP-19's grounded proposer (proposeInstructions) directly: each round's call passes the node's current instruction and its scored PastInstruction history, and the proposer returns ranked candidates with the current effective instruction always retained. Keeping that candidate writes no redundant override, preserving the safety property that a node never degrades.

Output is V1's CompiledProgram via freezeProgram, invoked through optimize and serialized unchanged (integration point #4).