Design and Experimental Evaluation of a Single-Version Software Fault-Tolerance Algorithm for Transient Execution Errors

Transient execution faults can corrupt data and intermediate results without terminating a running application. These faults may cause silent data corruption. Conventional fault-tolerance methods often depend on hardware redundancy, replicated execution, or independently developed software versions. This paper investigates an alternative based on single-version software fault tolerance. Building on prevailing research works on transient software fault tolerance, recovery, data-error recovery, and static redundancy, this study proposes an integrated single-version algorithm. The algorithm combines selective computation redundancy, data-integrity checks, assertions, program-flow monitoring, and checkpoint-based recovery. A small numerical application is used to demonstrate the design. A Python prototype is evaluated against an unprotected baseline using 10,000 integer inputs. The preliminary benchmark uses 50 timed repetitions and compares median execution times. The baseline completes in 0.359 milliseconds, while the protected implementation takes 1.451 milliseconds. This corresponds to approximately 304.3% execution-time overhead. The protected implementation also detects and recovers from a deliberately injected accumulator error in the demonstration. These measurements establish functional feasibility, not general fault coverage. The experiment is limited to a simple workload and one fault-injection pattern