Show your convergence profile — what does your loop actually look like?

[fitz2882]

The benchmark covers 10 workload cells across 2,000 trials, but they're all workloads I chose. I'd like to see loops I didn't design.

If you've wrapped something, paste:

result = lg.result
print(result.outcome)             # converged | oscillating | diverged | stalled | max_iterations
print(result.error_history)       # the trace
print(result.iterations_used)
print(result.savings_vs_fixed_cap)

Plus one line on what the loop does and what the error signal is. That's it.

Especially interested in the ugly ones. What's useful is oscillation, plateaus that resolve after you'd given up, divergence that rollback caught, and anything where the classifier called it wrong.

Nothing sensitive is needed — error magnitudes only, never prompts or outputs. (That's the same contract send_telemetry enforces, and it's tested.)