# lr-independence of the post-saturation clock (run 6, coordinator), beta2 0.995, N=88, p=97, eps 1e-16
Script run_lr_test.sh (clamp_gen.py, logs in logs_lr/). Prediction: T_post = T_grok - step(frac_sat>0.5) ~ 2 ln(g_sat/g_grok)/(1-beta2) ~ 2800 steps, independent of lr.

| lr | seed | step frac_sat>0.5 | T_grok | T_post |
|---|---|---|---|---|
| 5e-4 | 0 | 3200 | 6150 | 2950 |
| 5e-4 | 1 | 3300 | 6450 | 3150 |
| 1e-3 | 0 | 2500 | 5300 | 2800 |
| 1e-3 | 1 | 2550 | 5450 | 2900 |
| 2e-3 | 0 | 2000 | 4550 | 2550 |
| 2e-3 | 1 | 2050 | 4800 | 2750 |

A 4x change in lr changes the pre-saturation phase by 1.6x (2000 -> 3250) and the post-saturation phase by 1.15x (2650 -> 3050). The lr 1e-3 rows are from beta2_sweep.csv.