More grammar weirdnezz.md

I have the following token in a grammar that is behaving very mysteriously:

our token ident_sys is export {
  <keyword>
  |
  $<o>=[ <:Letter + [ _ @ # ]> <:Letter + [ _ @ # $ ]>* ]
  |
  # YYY: Verify that this is IDENT_QUOTED
  '"' [ <keyword> | <~~> ] '"'
  |
  "'" [ <keyword> | <~~> ] "'"
}

So, given that definition, you would expect the following to fail out:

!not
$dolla
0\@notavar
%caseabeer
'misquoted"
"misquoted2'

Yet I get the following gists from each one:

# ｢not｣
#  ws => ｢｣
#  ident_sys => ｢not｣
#   o => ｢not｣
#  ws => ｢｣

# ｢dolla｣
#  ws => ｢｣
#  ident_sys => ｢dolla｣
#   o => ｢dolla｣
#  ws => ｢｣

# ｢@notavar｣
#  ws => ｢｣
#  ident_sys => ｢@notavar｣
#   o => ｢@notavar｣
#  ws => ｢｣

# ｢caseabeer｣
#  ws => ｢｣
#  ident_sys => ｢caseabeer｣
#   o => ｢caseabeer｣
#  ws => ｢｣

# ｢misquoted｣
#  ws => ｢｣
#  ident_sys => ｢misquoted｣
#   o => ｢misquoted｣
#  ws => ｢｣

The only one of the above that does fail with the first character preserved is: "misquoted2'.

What is gobbling up the first character?