An obfuscated file that downloads successfully tells you nothing about whether it is protected. The only real check is to do what an attacker would: open it in a decompiler and look. It takes ten minutes, it is free, and it is the difference between knowing what you shipped and assuming. Here is what to look for, layer by layer.
Set up: one tool, two files
Install ILSpy — free, and the tool a
curious reader is most likely to use. Open both assemblies side by side:
your original Release build and the obfuscated copy. Comparing them is what makes the
differences obvious. Then confirm you are looking at the file that ships: on modern .NET
the code is in YourApp.dll, not the launcher YourApp.exe.
Renaming: can you still find anything by name?
Expand the assembly tree. In the original you see your namespaces and class names. In the obfuscated copy, private and internal types and members should have short, meaningless names, and — for an application — the namespaces should be gone.
- Try the attacker's first move: use ILSpy's search for words like
License,Trial,Passwordor your product's feature names. In a well-renamed build those searches find nothing among your own types. - Expect some names to survive, on purpose. A library's public API,
virtualmembers, enums, serializable types and anything you excluded keep their names. If far more survived than you expected, check whether you uploaded a library — its public surface is preserved by design (why). - Framework names always remain.
File.ReadAllTextandHttpClientare not yours to rename.
String encryption: search for a string you know is there
Pick a literal you know is in the code — an error message, a URL, a registry path — and search for it. Without string encryption it is found immediately, and it leads straight to the method that uses it, renamed or not. With string encryption on, the search finds nothing; where the literal used to be, the decompiled code shows a call to a decoder instead.
This is the single most revealing check. Many people discover here that their build only had renaming applied — which is what the free tier does, and what a Pro upload does if the layer boxes were left unticked.
Control flow: read one method you know well
Open a method whose logic you remember and compare. The original decompiles to clean
if/else and loops. With
control-flow obfuscation, the same
method becomes a tangle of jumps and goto-style blocks that does the same thing
but is slow to follow. One caveat that is normal, not a failure: methods containing
try/catch blocks are left in their original shape.
Reference proxying: look at the call sites
In the original, a method body shows calls like File.Delete(path) in plain
view. With reference proxying, calls to
external static methods go through small generated stubs, so the call site shows a
meaningless name instead of the API being called.
Virtualization: the method body should be gone
Open a method you marked with [Virtualize]. Its original logic should not be
there at all — in its place is a short stub that hands off to an embedded interpreter. If
you can still read the method's logic, it was not virtualized: check that the attribute is
on the method and that the virtualization option was on for that upload.
How virtualization works explains what
you are looking at.
Then run it
Looking protected and working are separate questions. After the decompiler check, run the obfuscated build and exercise the paths that use reflection, serialization or data binding — how to test your obfuscated build is the short routine for that.
A two-minute version for every release
- Open the shipped assembly in ILSpy.
- Search for one class name you know. It should not be found.
- Search for one string literal you know. With string encryption, it should not be found.
- Open your most sensitive method. You should not be able to read it comfortably.
If any of those fail, the release went out less protected than you think — most often because a build step shipped the unprotected file. That is exactly the mistake worth catching before your users, and anyone else, have the binary.