1.) BrainVisa samples should be removed from the NITRC GIfTI downloads (or the data updated so the combination of matrix and vertex location matches the expectation of the specification. To demonstrate the issue, these images appear upside down in connecotme workbench. 

2.) Section 5.0 ("Data Compression") of the specification should include a sentence explicitly describing the encoding used. I would write 'The tag "Encoding="GZipBase64Binary" denotes the data is compressed using the deflate algorithm with the standard zlib 2-byte header and 4 byte footer (https://tools.ietf.org/html/rfc1950). Therefore the first byte of the compressed datastream will be 0x78. Be aware that GIfTI compressed files do not use the 10+ byte GZip header and 8-byte footer (https://tools.ietf.org/html/rfc1952) typical of GZip files compressed using the deflate algorithm (where the first two bytes are 0x1f8b).' if you want to describe all the GIfTI images I have seen (or if either RFC1950 and RFC1952 are acceptable this should be explicitly described and samples provided). The issue I ran into reading the documentation is that GIfTI uses the key "GZip" which is is associated with RFC1952, but in practice I needed to implement RFC1950.

These two changes would make it easier to develop and validate tools using this format.