P145 tests a tighter execute-prefetch repair rule. Instead of globally removing the second repair word as in P144, it restores that word only when the prefetched instruction’s sequential next PC is the adjacent word in the same I-cache line.
The shell workload passes and reaches P145-FILE-OK.
The result is a useful negative. P145 adds 165,881 prefetch second-word grants and 107,964 background repair fills versus P144, but only gains 2,276 first later fetch hits and 1,016 repeat hits. The shell window worsens by 908,201 cycles versus P144.
Next step: roll the active policy back toward P144 and add audit-only counters for several candidate execute-prefetch usefulness predicates before changing the machine again.