SMT-LIB 2.7 / AIGER / BTOR2 Formal Verification Orchestrator (CDCL + BMC + k-Induction + Craig Interpolation + IC3)

Everything runs on a self-contained TypeScript stack: the SMT-LIB front end covers the common QF_BV/Bool operator set (with define-fun inlining and let bindings), AIGER files are read in both ASCII and binary (delta-encoded ANDs, latch next functions, outputs interpreted as HWMCC bad signals), and BTOR2 models are bit-blasted from the word level (sorts, consts, states with init/next, bad/constraint/output, and the arithmetic/logic opcode core). The SAT core is a real CDCL solver — two-watched literals, clause learning with first-UIP backjumping, activity-based branching and restarts — the same algorithm family inside cvc5 and Z3, at didactic scale. Safety engines: BMC finds shortest counterexamples with full input/state traces; k-induction adds unique-state constraints to catch invariants that plain induction misses; the interpolation engine derives candidate Craig interpolants from the refutation structure and then re-verifies each candidate by independent SAT calls before calling it an invariant — an unverified candidate is never reported as one; and IC3-lite maintains per-frame blocked cubes with k-induction closing the proof. Scale guidance: bit-vector widths ≤ 16 and bounds ≤ 30 keep runs interactive.