# Standing-profile repair: separate post-result verification

The preceding frozen 18-request panel is complete: 15 task executions and three pre-action rejections. The absolute joint-state admission test excludes seed 96167; it also excludes two of the three earlier successful lowering calibration seeds (96165 and 96166). All these starts exactly equal each robot's configured default pose. The native `add_joint_default_pos` event uses additive uniform [−0.01,+0.01] rad calibration offsets and updates the motor's action offset. Comparing to one seed's absolute joint angles therefore does not describe the standing reset protocol used by the capability evidence.

This is an explicitly post-result repair, not a replacement score for that panel. Freeze a separate ledger schema before three new native executions: public evidence selection on clear/original beam/beam −20 mm, seed 96167. All three requests and fixed-control outcomes have now been inspected. No held-out or generalization claim follows.

Only the standing-profile representation changes: compare actual joint angles to the robot's configured default pose within the same 1e-5 rad numerical tolerance. Require the default pose to differ from the calibration seed's default by no more than 0.020001 rad (the difference bound implied by the unchanged ±0.01 native event, plus numerical allowance). This uses already available robot configuration metadata, not a task outcome or extra scene truth. Initial root position/orientation, velocities, goal, map envelope, motor and bank bindings retain the previous checks.

The designated prior capability outcomes, eligibility rule, posture costs, motion bytes, geometry margin, native packing, motor, floor/beam assets, score and clock stay unchanged. No lower-ceiling repair or safety claim is introduced. All original rejections remain visible and counted in the original panel.

Compare each issued reference/token/action, initial state and dynamics against the exact existing matching executor: old nominal selector for clear, fixed lowered for both beam heights. Reuse those three controls; do not rerun them. Also check the repaired predicate on all three previously collected successful calibration starts, and on retained-seed initial states. For retained seed 96162, the repaired rule must produce the same choice and profile result as the executed original rule; those three executions are retained evidence, not new trials.

Success for the repair means removing the incorrect standing-state rejection and reproducing the selected fixed executor. Report full-task outcomes independently: matching a colliding fixed control verifies the interface but does not solve traversal. At most one preserved infrastructure retry per new case; no scientific retuning or task-score change. No training, moving reference replacement, sensor learning or hardware.