score.py goes haywire

[RobFisher]

Sometimes making one small mistake can be interpreted as a much larger mistake. This is due to the diff algorithm used, which does not guarantee to find the smallest possible diff.

It may be necessary to implement a specialised diff algorithm rather than use difflib.

[RobFisher]

As an example of the problem, here's a complete session:
$ python score.py "PWLU I.T MLSS.ST IRKJ .UTTMJ IRTSLMNU SI WWLJLU ..MWITJU TPOPAP.I MOW TSMMP AKL MTLRMO PATSSW KUO MNWPSULL KOAJ WKKLP J" "pwlu i.t mlss.st irkj .uttmj irtslmnu si wwljlu ..mwitju tlli mow tsmmp l mtlrmo passw kuo mnwpsull koaj wkklp j"

PWLU I.T MLSS.ST IRKJ .UTTMJ IRTSLMNU SI WWLJLU ..MWITJU T-P+L+L+I+ +MO+W+ +T+S+M+M+P+ +L+ +M+T+L+R+M+O+ PA-P-.-I- -M-O-W- -TS-M+S-M+W-P +K+U+O+ +M+N+W+P+S+U+L+L+ +K+OA-K-L- -M-T-L-R-M-O- -P-A-T-S-S-W- -K-U-O- -M-N-W-P-S-U-L-L- -K-O-AJ WKKLP J
Mistakes: 82

The same session with characters before the first mistake removed:
$ python score.py "..MWITJU TPOPAP.I MOW TSMMP AKL MTLRMO PATSSW KUO MNWPSULL KOAJ WKKLP J" "..mwitju tlli mow tsmmp l mtlrmo passw kuo mnwpsull koaj wkklp j"

..MWITJU T-P+L+L-O-P-A-P-.I MOW TSMMP -A-KL MTLRMO PA-TSSW KUO MNWPSULL KOAJ WKKLP J
Mistakes: 10

[RobFisher]

I think what I want is the Levenshtein distance.