Learning When to Ask: Budgeted Request Scheduling for Threshold Monitoring with Delayed UAV Energy Reports

Reproducible manuscript and computational artifact for Learning When to Ask: Budgeted Request Scheduling for Threshold Monitoring with Delayed UAV Energy Reports.

Scheduling-focused revision with prereg.md committed before all new experiments, a margin-adaptive empirical Markov-projection DP, nested selection on 27 development windows, two outer date directions with 100 channel seeds, strong fitted static controls, value-blind DP, myopic and tabular Q-learning. The registered one-point adaptive improvement gate fails in both directions despite positive descriptive intervals. Fitted static timing improves over evenly spaced requests, while globally optimal static schedules are certified only for budget four. Includes same-noisy-box channel oracles, mismatch and discretization ablations, complete contrast ledgers, input/output hashes and producer code history. Historical test exposure is disclosed, and no pristine confirmation or independent-flight generalization is claimed.

The single stock-LNCS submission PDF includes references and its internal appendix. Earlier sparse-fault, sender and calibration directions remain in the archive without competing for the main contribution. Public external-source records are short auxiliary checks, not confirmation of the 30-minute question. Restricted dense Solar arrays and raw external downloads are excluded. Neither physical sensor calibration, battery safety nor radio energy savings is established. Code MIT, author manuscript and aggregate results CC BY 4.0, inherited notices retained.