A simple substitution experiment

[oshfdk]

With a correctly transcribed ciphertext solving simple substitution for a single unknown language from a known list of languages is more or less trivial. It's also trivial even when there is a large number (say, 10%) of random transcription/scribal/encoding errors. However if there are systematic errors, like always transcribing two different characters as one, or interpreting one character as several different characters, then the standard solvers might not help.

For example, 

SOURCE: the voynich manuscript is an illustrated codex hand written in an unknown script referred to as voynichese the vellum on which it is written has been carbon dated to the early fifteenth century
SUBSTITUION, NO ERRORS: vxj tzbeqlx fneuwlcqov qw ne qgguwvcnvjk lzkjr xnek scqvvje qe ne uehezse wlcqov cjijccjk vz nw tzbeqlxjwj vxj tjgguf ze sxqlx qv qw scqvvje xnw mjje lncmze knvjk vz vxj jncgb iqivjjevx ljevucb
GUBALLA DECODER: the founich manyscript is an illystrated codex hand written in an ynknown script regerred to as founichese the fellym on which it is written has been carbon dated to the earlu gigteenth centyru

Very good, a few minor issues.

Now imagine we treat ciphertext q and g as the same character (replace g with q), and imagine that we see two different versions of v (replace 50% of v with g to keep the size of the alphabet).

MISTRANSCRIBED CIPHERTEXT: gxj tzbeqlx fneuwlcqov qw ne qqquwgcnvjk lzkjr xnek scqvvje qe ne uehezse wlcqov cjijccjk gz nw tzbeqlxjwj gxj tjqquf ze sxqlx qv qw scqgvje xnw mjje lncmze knvjk gz vxj jncqb iqigjjevx ljegucb
GUBALLA DECODER: phe younich langscrimp is an iiigstraped codex hand wrippen in an gnknown scrimp referred to as younichese the yeiigl on which ip is writpen has been carbon daped to phe eariu fifteenph centgru

While it's still possible to recognize this, mistranscribing only two characters types greatly reduced our ability to get the optimal outcome.

Let's add one more transcription issue, let's say we decided that ciphertext w is actually the same as vv.

MISTRANSCRIBED CIPHERTEXT: vxj tzbeqlx fneuvvlcqov qvv ne qqquvvgcnvjk lzkjr xnek scqvvje qe ne uehezse vvlcqov cjijccjk gz nvv tzbeqlxjvvj gxj tjqquf ze sxqlx qv qvv scqgvje xnvv mjje lncmze knvjk gz vxj jncqb iqigjjevx ljegucb
GUBALLA DECODER: she lognich manusscrips iss an iiiusstrased codey hand wrissen in an unknown sscrips referred to ass lognichesse the leiium on which is iss writsen hass been carbon dased to she earig fifteensh centurg

If you know what it's supposed to say, you can read it. But only tree systematic mistakes make it nearly illegible. 

But these were just simple changes, what if we misunderstood the mechanics of the cipher. Suppose it's actually to be read right to left or any other reordering happens. This will kill most simple statistical decoders instantly.

ORIGINAL CIPHERTEXT, NO TRANSCRIPTION MISTAKES, JUST REVERSED: bcuvejl xvejjviqi bgcnj jxv zv kjvnk ezmcnl ejjm wnx ejvvqcs wq vq xlqxs ez fuggjt jxv jwjxlqebzt wn zv kjccjijc voqclw eszeheu en eq ejvvqcs kenx rjkzl kjvncvwuggq en wq voqclwuenf xlqebzt jxv
GUBALLA DECODER: brusted nsteesmam blroe ens is hesoh tixrod teex pon tessarc pa sa ndanc ti fulleg ens ependatbig po is herremer swardp tcitytu to ta tessarc hton vehid hesorspulla to pa swardputof ndatbig ens
GUBALLA DECODER RESULT REVERSED BACK: sne gibtadn fotupdraws ap ot allupsroseh dihev noth crasset at ot utytict pdraws remerreh si op gibtadnepe sne gelluf it cnadn as ap crasset nop xeet dorxit hoseh si sne eorlb mamseetsn detsurb

That's it, one very simple misunderstanding of how the cipher works and most standard statistical decoders are useless.