A QR code dedicates around 30% to metadata (positioning, timing data, version numbers). There are very few metadata arrangements shared across all QR codes, so you can recover 30% of the code with almost no computation.
The other 70% is dedicated to "data", on which the Reed-Solomon-coding is applied. If your QR code is absolutely stuffed to its limit (about 2kb data), you can still destroy up to 51% of the image before the Reed-Solomon coding will start to fail at 30% redundancy.
Most codes only store a few hundred bytes of data, so you can afford to destroy a lot more before the payload is also destroyed.
I'd expect the minimal-viable QR code for a 100 character URI can be very degraded.
The other 70% is dedicated to "data", on which the Reed-Solomon-coding is applied. If your QR code is absolutely stuffed to its limit (about 2kb data), you can still destroy up to 51% of the image before the Reed-Solomon coding will start to fail at 30% redundancy.
Most codes only store a few hundred bytes of data, so you can afford to destroy a lot more before the payload is also destroyed.
I'd expect the minimal-viable QR code for a 100 character URI can be very degraded.