Chat + tool-call storytelling
Same learn_from_outcome tool, different outcomes:
A reports a success — confidence boosted +0.075, Q-value rises
0.5→0.65, then a second success compounds to 0.69/0.755.
B reports partial_success — smaller boost (+0.024), Q-value
0.5→0.545. Real MCP payloads showing confidence & Q-value propagation.