Three contributions are presented. First, LogicBombs-Go, a dual-compiler benchmark corpus of 25 Go programs with ground-truth trigger conditions, compiled with both the TinyGo and the reference gc toolchains. Second, a single-threaded engine that interprets P-Code against a faithful x86-64 CPU and memory model, detects vulnerabilities at the micro-operation level, and explores untaken branches through a panic-gated filter cascade that concentrates solver queries on branches able to reach a known panic site. Third, an extension to gc-compiled binaries whose runtime is multi-threaded, through multi-thread state recovery, controlled preemption neutralisation, and overlay concolic execution, a copy-on-write replay of untaken branches that detects silent vulnerabilities such as integer overflows that leave no panic marker. The evaluated gc results target sequential vulnerability classes inside multi-threaded runtime artefacts under a main-only scheduling policy; systematic goroutine-interleaving exploration is prepared through Volos but remains future work. On real-world vulnerabilities from Kubernetes, Go-Ethereum, and other production projects, Zorya detects bugs that static analysers and binary symbolic executors miss, including a silent overflow no other evaluated tool finds without a hand-written oracle, and a disclosure campaign surfaced twenty-two further findings. Zorya analyses compiled artefacts without source instrumentation or fuzzing harnesses; however, the evaluated function mode relies on DWARF debug information and unstripped symbols to recover argument layouts. The thesis also lays the design and an initial implementation for concurrency-bug detection, and Zorya is released as open source.