Error propagation (Mega) Parsec through continued

Question

I have a few technical questions about the internal work of (Mega) Parsec, the informal context of which is:

What is the idea governing the accumulation / distribution of error messages through continuations in Parsec?

In particular: Consider an operation mplusdoing an ParsecTinstance MonadPlus. ( Source ) mplus m nreceives as an input string sfor parsing and continuation of cok, cerr, eokand eerrto follow the results of the analysis.

First execute the syntax swith m:

• If the parsing succeeds and consumes (accordingly does not consume) the part s, follow the continuation cok(respectively, eok).
• When parsing sa mfailing part using s, be continued cerr.
• Suppose parsing sc mfails without consuming anything; let erris a parsing error associated with this failure. Then we will try to parse swith nand accordingly choose a continuation:
  • If parsing swith nsucceeding at input consumption, be continued cok.
  • If parsing swith ndoes not work when consuming input, follow the continuation cerr.
  • If parsing swith ndoing well without the use of input, be continued neok, resulting from eok:neok y s' err' = eok y s' (mergeError err err')
  • When parsing sa nfail without consuming input, follow the continuation neerrobtained from eerr:neerr y s' err' = eerr (mergeError err err')

My question is:

Why do we combine errors by following the “empty” continuations, and not following the “consumed” continuation, thereby forgetting about all previous errors?

It should be a design decision, but I can not understand the reasons for this. Maybe a simple example that will clarify this?