verifiable pairs only (the four 052/271 validation pairs excluded)
bexec6 alone (12 fresh dev runs)   +0.011 [-0.018, +0.041]  attempts=120 task identities=10 improved=9 worsened=5 with_feedback=97 cases_fixed=10 cases_broken=7
paper: six dev runs                +0.022 [-0.005, +0.060]  attempts=60 task identities=10 improved=6 worsened=1 with_feedback=48 cases_fixed=7 cases_broken=2
all 18 dev runs                    +0.015 [-0.006, +0.039]  attempts=180 task identities=10 improved=15 worsened=6 with_feedback=145 cases_fixed=17 cases_broken=9
paper primary: 6 dev + val         +0.022 [-0.002, +0.048]  attempts=134 task identities=47 improved=13 worsened=5 with_feedback=94 cases_fixed=18 cases_broken=6
18 dev + val (descriptive pool)    +0.017 [-0.002, +0.037]  attempts=254 task identities=47 improved=22 worsened=10 with_feedback=191 cases_fixed=28 cases_broken=13
