NATIVE LAB / STATIC → EMULATED → MEASURED

Reverse the rule.
Prove the port.

A deeper native-game workflow: inspect a shipped executable, recover function maps, probe a bounded machine-code slice and compare a separately instrumented original with Rust.

Read the technical guide ↗Evidence Workbench ↗

OPT-IN ANALYSIS · LOCAL EVIDENCE · NO GAME ASSETS BUNDLED

RUSTPORT / RESEARCH PIPELINEvNEXT

INPUTOriginal .exe / .dllSHA-256 pinned · no runtime launch

↓ STATIC INFERENCE

GHIDRAFunctions + callsHeadless export

LEDGERUnknown → mappedEvidence per symbol

↓ BOUNDED EMULATION

UNICORN / OPTIONALIsolated x86-64 functionInstruction & time budgets · not a VM

↓ INDEPENDENT EXPERIMENT

ORIGINAL HARNESSGolden tracesObserved behavior

RUST PROBEPort tracesSame input sequence

FIRST DIFFERING FIELD / REPLAY / AUDIT

SCHEMATIC · NOT A CLAIM OF WHOLE-GAME PARITY

01 / REAL TOOLS, CLEAR LIMITS

Closer to the metal.
Still evidence-led.

01 — RECOVER

Ghidra Headless

Bundled Java exporter records original function entry points, names, body size and bounded call edges. Optional decompilation remains private, and exports are SHA-256 identified.

native_lab.py analyze · ingest

02 — EXPLORE

Unicorn x86-64

Take an explicitly verified raw function slice through a limited emulation harness. No OS, dynamic loader, or external calls. Do not confuse this with observing the game itself.

native_lab.py slice · emulate

03 — CHALLENGE

Differential fuzzing

Run independent original and Rust JSON harnesses on seeded inputs. Capture the first difference, preserve a replay corpus and attempt a smaller failing input.

differential_fuzz.py · parity

02 / CLI LAB

Agent workflows.
Not magic prompts.

Native tools are optional. Analysis and execution require explicit opt-in, are bounded and may require an isolated VM. The code does not silently download software, modify your MCP settings or install agent plugins.

Commands below are documented examples. They are not results from a real game.

LOCAL / RESEARCHPYTHON 3.9+

python3 scripts/native_lab.py doctor

python3 scripts/native_lab.py analyze \
  --binary Game.exe \
  --output .rustport/native/functions.json \
  --allow-analysis

python3 scripts/native_lab.py ingest \
  --export .rustport/native/functions.json \
  --db port-ledger.sqlite

python3 scripts/differential_fuzz.py \
  harness.json --output .rustport/fuzz/001 \
  --seed 42 --iterations 100 \
  --allow-execute-original

03 / WHAT IS VERIFIED

Evidence levels
are not interchangeable.

STATICInferred

Function locations from disassembly are hypotheses, not observations of gameplay.

EMULATEDBounded

Machine code behavior depends on captured ABI state, relocations and available memory.

OBSERVEDReference oracle

A real original-side harness is required before claiming runtime behavior.

COMPAREDParity for tested cases

Recorded equivalence does not certify all game mechanics or asset compatibility.

Reverse the rule.Prove the port.

Closer to the metal.Still evidence-led.