Add GigaBIT M1 frame stiffness characterization project

- 3 studies: sensitivity (Method D), stiffness sweep (Method B), robustness
- Atomizer specs with run matrices, Zernike annular OPD extractor
- War-room validated hybrid B+D method

projects/gigabit-m1-frame-stiffness/studies/03_robustness/README.md

@@ -0,0 +1,26 @@
+# Study 03 — Robustness Checks
+
+**Type:** Parametric — Specific Scenarios
+**Runs:** 8
+**Model:** Uses both .sim setups (CBUSH for K tests, displacement for correlated tests)
+**Prerequisite:** Complete Studies 01 and 02 first (needs knee values)
+
+## What This Does
+
+Tests scenarios that the simple sweeps can't capture:
+
+1. **One-soft-point (3 runs):** What if one support is 10× softer than the rest?
+2. **Combined compliance (3 runs):** Both vertical and lateral at knee values simultaneously
+3. **Correlated displacement (2 runs):** All verticals move together (frame bending mode simulation)
+
+## Before Running
+
+Update `atomizer_spec.json` with actual knee values from Study 02:
+- Replace `KNEE` placeholders with K_vert_knee and K_lat_knee values
+- Replace `KNEE/10` with knee/10, `KNEE/2` with knee/2
+
+## Output
+
+- Interaction factor: is combined ΔWFE > sum of individual?
+- Worst-case single soft point identification
+- Correlated vs independent displacement comparison