7. Conclusion

of time-system constants, program input variables, and applications of convert-time. The decision procedure automatically simplifies nested applications of convert-time. The final step in DSDRAT's analysis of coordinate-to-time is to add the instantiated theories from TableGraph and Connected Groupoid to the cumulative set of axioms for instantiated decision procedures. DSDRAT then removes the axioms in the NAIF domain theory that are implied by this set. The removed axioms include the three listed in section 5. When these axioms are used by a general-purpose theorem-prover (e.g., SNARK) for deductive synthesis, they typically invoke search through branch points introduced by paramodulation. By replacing these axioms with efficient decision procedures, DSDRAT speeds up program synthesis without manual tuning. We expect that DSDRAT will automatically achieve the same results as were manually achieved with this methodology (section 5). This will provide a key enabling technology for domain experts, such as the JPL NAIF group, to maintain and extend their own specialized AMPHION systems. This paper addresses one aspect of scaling-up KBSE: enabling domain experts to construct and maintain their own domain-specific KBPS systems. The META-AMPHION system, currently under development, is designed to provide the KBSE analogue of application-generator generator technology. A key component of META-AMPHION is a subsystem to automatically operationalize a declarative domain theory for efficient deductive program synthesis. This paper describes extensions of a previous system, DRAT, that speeds up a theorem-prover for analytical reasoning problems by substituting decision procedures for axioms in a theory. An experiment with AMPHION (a real-world KBPS system) demonstrated that extending DRAT to deductive synthesis would be successful. The design for these extensions is described, and are currently being implemented. In parallel work we are also exploring suitable user interfaces for META-AMPHION that will guide domain experts in developing and maintaining domain theories. There appears to be a useful synergy with the component described in this paper: the same process described in section 6 for operationalizing axioms in an existing domain theory might be useful in eliciting axioms from a domain expert. Instead of using a theorem-prover to answer questions when searching the hierarchy, the domain expert would be used as an oracle to construct portions of a domain theory by traversing the same hierarchy. In previous work on DRAT, the work reported in [7] was extended to show that DRAT's attachment of literal satisfiability procedures to a theorem-prover was sound and complete. These results must be extended …