Compositional generation of robot mission admission conditions from incomplete execution evidence

Robot missions are often dispatched on execution evidence that covers only part of the planned task: a route bound omits the return tail, a controller guarantee needs more time than the configured timeout, or a position estimate is valid only under assumptions the flight will not meet. We pose admission as a constructive question: what must the missing evidence establish, and which task settings must change, before the mission can be admitted? For a finite task built from sequence, environmental choice and bounded repetition, the method generates two coupled outputs. The first is the polyhedral region of missing resource upper bounds that jointly fit the remaining budget; it is exact in the declared bound language. The second is the set of completion, interface, terminal and coverage obligations that no resource bound can discharge; task-structure rules derive them, and only evidence whose declared scope matches the execution context can discharge them. Admission is conditionally sound. The coupling carries the decision: on sixty held-out tasks with two missing bounds, the joint region admits all 120 feasible task-budget conditions where fixed per-phase allocations admit 23. On 160 frozen software tasks the conditions produce no erroneous admission, whereas dropping completion or interface obligations admits 25 and 24 failing tasks. Under a 5 m GPS step in ArduCopter, estimate-only checking certifies executions that all miss the docking tolerance, while scope matching withholds every one. On twelve new PX4 tasks, generated watchdog timeouts admit and complete eight missions that the original setting never completes. This record contains the manuscript (17 pages) and supplementary material (26 pages). Experimental code and derived data are not included in this record. Fields to complete in Zenodo: publication date, license, and DOI reservation/assignment. No journal publication or acceptance is claimed.