# Robust QAOA Optimizer Baseline

This report compares the default RA-QAOA grid-selected parameters against a SciPy Nelder-Mead QAOA optimizer baseline.
The optimizer uses the same fixed initial point and `maxiter=80` for every graph/depth pair, then the resulting parameters are evaluated under the same noise sweep as RA-QAOA.

## Robust vs Optimizer Summary

| depth | method | graph_count | mean_noisy_expected_cut | ci95_noisy_expected_cut | mean_success_probability | mean_noise_sensitivity | ci95_noise_sensitivity |
| --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | optimizer | 60 | 8.743269 | 0.991982 | 0.131064 | 3.935351 | 0.272510 |
| 1 | robust | 60 | 7.676503 | 0.873080 | 0.096099 | 0.870598 | 0.206181 |
| 2 | optimizer | 60 | 9.311529 | 1.026225 | 0.247469 | 5.569756 | 0.399816 |
| 2 | robust | 60 | 7.750276 | 0.859378 | 0.127260 | 1.079351 | 0.232858 |

## p=2 Robust Minus Optimizer Deltas

| metric | graph_count | mean_delta_robust_minus_baseline | ci95_low | ci95_high | robust_better_rate |
| --- | --- | --- | --- | --- | --- |
| noise_sensitivity | 60 | -4.490405 | -4.999357 | -3.981453 | 1.000000 |
| noisy_expected_cut | 60 | -1.561254 | -1.738209 | -1.384298 | 0.000000 |
| success_probability | 60 | -0.120209 | -0.138005 | -0.102413 | 0.016667 |

## Interpretation

The optimizer baseline is a performance-oriented comparison, not a robustness-aware selector.
If optimizer noisy E[C] dominates but sensitivity is higher, the RA-QAOA claim should stay framed as a stability trade-off.