How do we have encoded info in decoder side for hamming distance calculation?

I thought that I completely understand the idea of convolutional coding and viterbi decoding.However,now,there is one funny question in my mind that I have never thought about.

I am not sure how we compare received data code words with outputs on trellis diagram and find hamming distances,branch metrics and so on.I mean should I imagine like "trellis again drawn in viterbi decoder"? No,I don't think so because,to find hamming distances,we already have encoded data outputs on trellis? So,what? how decoder knows these encoded data outputs if there is a noise? If decoder knows these encoder outputs, i mean code words, to compare received noisy code words,why do we need decoder then? I know I am thinking completely wrong now and I would like to learn if you guys make some comments.

I hope my english was enough to explain what my concerns are.Thank you for your helps.

While I am trying to sleep,I guess I found out the answer

Of course,decoder doesn't know the trellis if I don't say to decoder that what trellis is in encoding part so while I create viterbi decoder,I need to mention the original trellis and its outputs in decoder design so viterbi decoder will be able to know the trellis of encoder and outputs of trellis so it will have a chance to find hamming distances,Bm etc.

I know that convolutional encoding may vary according to my design but why I didn't think that according to my design,viterbi decoder design will also change.I was thinking that all viterbi decoders are same but it was a fault I guess.

Hope,I am not wrong in my ideas