A real graded assignment

Real output · sample assignment

See a real graded assignment.

Every line below is exactly what a student would see — produced by the live GridGrade engine on a short sample workbook, nothing retyped or touched up. It shows the four things hand-grading does and most auto-graders don't: it accepts a different correct method, explains a deduction with the evidence and a fix, follows the dependency chain for partial credit, and flags what it isn't sure about. You stay the grader of record.

This is genuine engine output. The four cases come from a small sample assignment (regional sales, an invoice, and a real worksheet-source PivotTable) graded by the same engine that grades a submission. The scores, evidence, and fixes are reproduced verbatim — no student names or files are involved.

1 · It accepts a different correct method.

Q1 — Total sales for the West region (10 pts)

Accepted

The answer key used =SUMIFS(B2:B5,A2:A5,"West"). This student wrote something different:

=SUMPRODUCT((A2:A5="West")*B2:B5)

located Sales!B7 passed is_formula · result method sample_eval confidence 1.0

10 / 10 · full credit

GridGrade evaluated both formulas on the data, saw they compute the same result, and gave full credit. SUMPRODUCT instead of SUMIFS is a valid method — not a wrong answer.

2 · It explains a deduction — with evidence, the location, and a fix.

Q2 — PivotTable of loans by purpose and credit risk (10 pts)

Deduction

The pivot's placement is right (rows, columns, and filters all matched), but the Values area aggregates the wrong way:

Value field/aggregation does not match expected — aggregation mismatch (pivot count != expected sum)

located Q1!A4:D16 passed rows · columns · filters failed aggregation

−2 8 / 10 · flagged for review

Fix: Fix the pivot's aggregation — change the value field from Count to Sum. Because the aggregation differed, GridGrade couldn't auto-verify the pivot's totals, so it also flagged the question for you to confirm rather than guess.

3 · It follows the dependency chain for partial credit.

Q3 — Extended cost = unit price × quantity (two linked cells, 20 pts)

Upstream — Unit price

The unit-price cell is wrong on its own merits — its formula recomputes to the wrong number, so it owns its own deduction:

=B3+4

located Invoice!B5 passed is_formula failed result

−5 5 / 10

Result does not match the expected value. Expected canonical form 12.

Downstream — Extended cost

The extended-cost cell uses the right method (price × quantity). It inherits the wrong unit price from above — but GridGrade does not penalize it a second time for a value it inherited:

=B5*B3

located Invoice!B6 passed is_formula · result method sample_eval

10 / 10 · method credit

Q3 earns 15 / 20 overall: the bad input loses its points once, and the correct downstream step keeps its credit. A wrong number doesn't cascade into a zero for every formula that touched it.

4 · When it isn't sure, it flags — it never silently zeroes.

Q4 — Sum of orders over $50 (10 pts)

Flagged for review

The student answered with a function outside the set GridGrade can fully evaluate (a dynamic-array FILTER). Rather than assume it's right or wrong, GridGrade grades the structure and hands it to a human:

=FILTER(B2:B4,B2:B4>50)

located Sales!B6 passed is_formula confidence 0.5

needs review

Out-of-set formula graded structurally; could not recompute value. Manual review recommended. The grade is held at low confidence and surfaced to you — never quietly marked zero.

These four outcomes are the actual return values from the GridGrade engine on the sample workbook — captured field-for-field, including the located cell, what passed and failed, the confidence, and the fix text.

See it on one of your own assignments.

Tell us about your course and we'll run GridGrade on a real exam you already grade by hand — then walk you through the evidence behind every mark. Design partners pay $0 and help shape what ships.

Request early access See the full method →

What happens next: one human replies within two business days. No sales calls, ever.