Keep hard selection documentation focused on published rule

metadata/splits/hard_methodology.md

@@ -73,9 +73,7 @@ define the hard-candidate pool, then `hardness_score` ranks tasks within that
 pool. The saturation condition prevents a large or structurally complex task
 whose Gurobi runs all finish with certified optima below the runtime threshold
 from qualifying solely because of its structure. Candidate status takes
-priority over score in the full ranking. In this frozen snapshot, however,
-ranking by `hardness_score` alone produces the same top 50 IDs in the same
-order; the candidate gate does not change the published hard-50 membership.
+priority over score in the full ranking.
 
 In the frozen snapshot, `median(scale) = 0.5261111111111111` and the capped
 runtime 75th percentile is `3600` seconds. There are 109 hard candidates. Sort