FLAC allows embedding a seektable but encoders rarely emit one in practice. Since FLAC boils down to a mix of verbatim subframes and linear predictive coding (

FLAC allows embedding a seektable but encoders rarely emit one in practice. Since FLAC boils down to a mix of verbatim subframes and linear predictive coding (specifically, polynomials with warmup samples or Rice coding), non-verbatim subframes are variable-length, and the samples for a given interval of time may even span between two blocks. It's a tricky problem, but doable, and FLAC implementations IME (libflac, libavcodec, etc.) have varying levels of accuracy when seeking without a seektable. I assume the choking described depends largely on the medium of transport (e.g. HTTP servers without support for Range headers makes streaming practically impossible), followed by some delay to "guess" the correct block to jump to then figure out which frames contain samples closest to the requested timestamp.