CouchDB unexpected reduce/rereduce behaviour - mapreduce

Question

I'm assuming I'm looking at the problem incorrectly, but I seem to be hitting rereduce unexpectedly
A simplified example of my problem would be a student report card
Data (in table format for easy reading)
Entered Student Assig Grade
2019-02-01 Alice 1 0.80
2019-03-01 Alice 2 0.50
2019-04-01 Alice 2 0.80
2019-04-01 Alice 3 0.80
The story that goes with the data is that Alice is a good student, but the instructor fat fingered the data entry. Alice, pointed it out to the instructor who then put an amendment into the grade report. For accounting reasons, entries may not be deleted, but updated entries may be added.
The Map:
function(doc){
var key = [doc.student,doc.assig];
var val = {
grade:doc.grade,
entered: doc.entered
};
emit(key,val);
}
There are then two objectives when reducing:
Take the most recently entered grade
Gather up stats on the grades
function(keys,values,rereduce){
if(rereduce){
return sum(values);
}
else{
value = values.pop();
for(var v in values){
v = values[v];
if(v.entered > value.entered){
value = v;
}
}
return value.grade;
}
}
The minute I apply group=true, I'm expecting to get a list of the most recent grades for each assignment. Instead, I'm getting sums of all the grades.
Key Actual Expected
["Alice",1] 0.8 0.8
["Alice",2] 1.3 0.8
["Alice",3] 0.8 0.8
Oddly, further reducing:
Key Actual Expected
["Alice"] 2.4 2.4
I'm confused. How have I confused myself?
(CouchDB v2.3)
EDIT
OK, so I understand how I've got it wrong (overflowing indexes), but now I'm wondering how to make it right...
http://guide.couchdb.org/draft/views.html#reduce
I'm also still very confused about the group_leve 1 & 2 behaviour.

Answer 1

I suspect what is going on is that I am using the wrong tool to the job. Reduces in CouchDB work best when they have one job.
The solution used in the end was to create an update function that simply prepends the new grade to an array on the existing document.
Heavily based on an RTFM moment: https://docs.couchdb.org/en/stable/ddocs/views/collation.html
Student Assig Grades
Alice 1 [{e:'2019-02-01',g:0.80}]
Alice 2 [{e:'2019-04-01',g:0.80},{e:'2019-03-01',g:0.50}]
Alice 3 [{e:'2019-04-01',g:0.80}]
The Map changes to...
function(doc){
var key = [doc.student,doc.assig];
var val = doc.grades.shift().grade;
emit(key,val);
}
... and the reduce can use the built-in reduce functions.
The only thing lost is that modifications to the data are not as apparent, however a it is simple to create a view to report on that:
function(doc){
for(var val in doc.grades){
var key = [doc.student,doc.assig,val.entered];
emit(key,val.grade);
}
}
Now the users have access to both the detailed listing of changes, as well as the aggregated results.
The code above is untested, as its a trivialized example of a problem. Conceptually, its correct; syntactically, I expect issues.