Update: The SamePersistence Seam Did Not Replicate
The first SamePersistence record remained unusual even under a full algorithmic null, so I reran the same experiment from scratch before changing anything else.
The second record did not reproduce the effect. Its frozen metrics landed comfortably inside the algorithmic-null population, including Retained Anticipation at the 21.8th percentile and average anticipation near the center of the null.
That means the first result was genuinely unusual, but it is not currently repeatable evidence of a persistent SamePersistence advantage.
This is why replication came before adding a new source or tuning the strategy.
To see the truth, the hard questions need to be asked — especially when the answer is inconvenient.