Characterize the artifact
Identify container properties, architecture signals, integrity conditions, compiler and build indicators, and version-sensitive constraints — the profile everything downstream is calibrated against.
Recover the structure
Recover accessible code and data regions, candidate functions, references, call relationships, control flow, constants, strings, and platform interactions into a navigable structural map.
Confirm behavior at runtime
Where the target and authorization permit, scripted instrumentation collects reproducible execution evidence — transitions, register and memory state, module and call-path activity — under defined, logged conditions.
Normalize into IR
Lift low-level instructions into an architecture-aware intermediate representation modeling blocks, calls, state, types, and side effects — the common language every later stage reasons over.
Prune the candidate space
Evaluate competing interpretations under structural, data-flow, type, calling-convention, platform, and runtime constraints — eliminating everything the evidence forbids and ranking what survives.
Apply specialist AI
Authorized high-capability models classify routines, infer types and roles, correlate distant evidence, and flag ambiguity — proposing hypotheses, never declaring verdicts.
Force the evidence to agree
Compare independent passes, expose contradictions, reject incompatible hypotheses, and return ambiguous regions for deeper analysis before anything is allowed to progress.
Deliver with confidence
Label each finding by evidentiary strength — verified, high, probable, tentative, or unresolved — so a strong result in one region never disguises uncertainty in another.
绝不由单一工具单独做出结论
反汇编器提供静态视角,调试器提供动态轨迹,AI模型提供分析假设。流水线将它们全面关联,剔除与底层二进制冲突的假设,唯有独立证据互相吻合时才采纳结论。
