feat(nimbus): Include ration in the calculation for the SRM ratios alert

[yashikakhurana]

Because

• Experiments with intentionally unequal branch ratios (e.g. 95% delivery / 5%) were triggering false SRM mismatch alerts because the check assumed equal distribution across branches

This commit

• Passes each experiment's configured branch ratios to the SRM check so the chi-square test compares actual enrollment against the correct expected ratio, not 50/50

Fixes #16010

[yashikakhurana]

@jaredlockhart Fixed two things that were causing false SRM alerts:

• The check assumed equal branch ratios even for intentional splits like 95/5 — now it uses the actual configured ratios
• With large populations, even a 0.2% difference was triggering the alert — added a minimum 5% deviation so it only fires when something is actually wrong

[jaredlockhart]

Oh yes good, taking the designed ratio into account good catch! And actually yeah checking telemetry shows that 5% is probably exactly what we want. One thing though I think this only updates slack, don't we also need to pass the ratios in here:

experimenter/experimenter/experimenter/experiments/models.py

Line 1589 in 4abc978

is_srm, srm_p_value = check_srm_mismatch(self.monitoring_data)

To fix it in the UI?

[yashikakhurana]

Oh yes good, taking the designed ratio into account good catch! And actually yeah checking telemetry shows that 5% is probably exactly what we want. One thing though I think this only updates slack, don't we also need to pass the ratios in here:

experimenter/experimenter/experimenter/experiments/models.py

Line 1589 in 4abc978

is_srm, srm_p_value = check_srm_mismatch(self.monitoring_data)

To fix it in the UI?

ahh very good catch @jaredlockhart

changes done