Provocations run under the verified [skate] detector: sustained healthy
walking, 7x respawn-while-moving, gimp onset at speed, ~21s of sustained
GIMPED walking (aimed-leg self-damage: BT_SELF_DAMAGE_ZONE=dz_ldleg),
leg-destruction death (authentic: lvl 1.0 -> leg gone -> fall/death; the
mid-session "died of the gimp edge" reading was a capped-print artifact,
retracted), and respawn. ZERO skate anywhere. Peer gimp replication
VERIFIED live: observer reads gl=3 + sim=3 with the gimp bodyStates
cycling for the whole master limp window (#82 remains fixed). Conclusion:
the field skating does not reproduce at lab scale; the detector + SKATE
matchlog record ship with the next cut and the field names the failing
case. Bench scripts archived (skatebench3-7; 7 is the clean-room one --
the sed-derived chains dropped envs twice).
Also: BT_LAMP_LOG=1 joins the field bats (#135 -- lamp/annunciator edge
forensics; near-zero noise, answers leak-no-flash reports in one grep).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
