Stress And Ablation - Cycle 4 v0.1

Date: 2026-04-19Scope: sensitivity thresholds D03 (loss_u) on benchmark Cycle 3 input.

Setup

• Stress script: datasets/cycle4_work/run_d03_threshold_stress.py- Input: datasets/cycle3_inputs/D03.csv- Tested grid:- low in {0.272, 0.274, 0.276, 0.278}- high in {0.284, 0.286, 0.288}

Global outcome

• Evaluated combinations: 12- Combinations with complete coexistence (preservativo and degenerativo in all contexts): 12/12- Combinations with ambiguous_rate < 0.20: 9/12

Robust configurations (selection)

| low | high | ambiguous_rate | Coexistence || ---: | ---: | ---: | --- | | 0.276 | 0.284 | 0.1077 | YES || 0.276 | 0.286 | 0.1385 | YES || 0.278 | 0.284 | 0.0769 | YES || 0.278 | 0.286 | 0.1077 | YES |

Interpretation

The D03 revision does not depend on a single point: there is a stable region of thresholdswith multi-context coexistence and sub-threshold ambiguity. Raw Output:- datasets/cycle4_outputs/d03_threshold_stress.json- datasets/cycle4_outputs/d03_threshold_stress.csv