An average percentage calculator should make that choice visible. This original example compares an equal-row mean, a pooled count rate and a separately declared weighted mean. It also keeps missing records from quietly becoming zeros or disappearing from the scope.
Keep the counts behind each rate
These two fictional groups are disjoint, use the same assignment-count basis and share the snapshot date October 13, 2026. They are illustrative records, not actual staff or student results.
| Group | Completed | Assigned | Rate |
|---|---|---|---|
| G01 | 9 | 10 | 90% |
| G02 | 1 | 2 | 50% |
The first group contains five times as many assigned tasks as the second. Decide whether you want to give each group equal weight, give each assignment equal weight, or use a separate declared weighting scheme.
Compare three explicitly labeled answers
| Mode | Calculation | Answer | Question answered |
|---|---|---|---|
| Equal-row mean | (90 + 50) / 2 | 70% | What is the average of these two group rates? |
| Pooled count rate | (9 + 1) / (10 + 2) × 100% | 83.333333% rounded | What share of these 12 assignments is complete? |
| Custom weighted mean, weights 2 and 1 | (90 × 2 + 50 × 1) / (2 + 1) | 76.666667% rounded | What is the mean under the separately declared 2:1 group weighting? |
NIST's Dataplot reference gives the ordinary mean and weighted-mean formulas. The weighted mean divides the sum of each value multiplied by its weight by the sum of the weights. We use those basic formulas here, without running Dataplot or adopting a grading policy. NIST weighted mean reference
The custom 2:1 weights are illustrative inputs, not an official reason to favor G01. Record the actual meaning and authorization of weights before interpreting a real result.
You can reproduce the pooled answer with denominator weights 10 and 2:
(90 × 10 + 50 × 2) / (10 + 2)
= 1000 / 12
= 83.333333% rounded
Those weights correspond to the assignment counts. Custom weights 2 and 1 do not. Label both the mode and the denominator beside the result so “average percentage” cannot hide the distinction.
Percentages alone cannot reconstruct the pooled rate
If you only have 90% and 50%, you can compute the equal-row mean of 70%. You cannot recover the pooled count rate without the corresponding numerators and denominators.
For example, the original rates came from 9/10 and 1/2. Their pooled rate is 10/12. Treating the percentages as if their group sizes were equal would answer the equal-row question instead. Do not invent count weights to fill a missing source record.
Recompute from unrounded source counts or rates when available, then round the final result. Do not average already rounded display values when more precise source values exist.
Make missing coverage part of the output
Suppose the intended report also includes G03, but its counts have not been supplied. The available G01/G02 calculations still have a defined two-row scope. The full three-group pooled aggregate is unavailable.
Show both statements: “Available two-group pooled rate: 83.333333%” and “Incomplete coverage: G03 missing; full three-group aggregate unavailable.” Apply the same scope warning to any available-row mean. Do not silently exclude G03 and label the result as the whole report.
Keep these states distinct:
| Row or weight state | Required treatment |
|---|---|
| Recorded zero completed with positive assigned count | A defined 0% rate |
| 0 completed / 0 assigned | Undefined row rate; not 0% |
| Missing count, percentage or required weight | Missing input; no assumed zero |
| Explicit weight 0 in declared weighted mode | Excluded by that declared weighting, with a recorded reason |
| All supplied weights 0 | Undefined weighted mean: total weight is zero |
A zero-weight exclusion is a declared decision, not a way to erase missing records without explanation. Preserve the row, reason and coverage status in the ledger.
The example uses finite percentages from 0 through 100 with at most six decimal places. Weights must be finite and nonnegative, with a positive total for a defined weighted mean. Count mode requires integers with 0 ≤ completed ≤ assigned ≤ 1,000,000, positive assigned counts, and disjoint comparable groups. Rows with a zero denominator remain flagged rather than being accepted as 0%.
Copy this prompt for the aggregation specification
Prepare an original US-English specification and worked ledger for an average percentage calculator. Do not implement or claim a tested tool, official grade, quality score or policy.
Fictional snapshot October 13, 2026: disjoint comparable assignment groups G01 completed 9 of 10 assigned = 90%; G02 completed 1 of 2 assigned = 50%. Keep counts, date and basis visible. Show three labeled modes:
1. Equal-row mean: (90+50)/2 = 70%.
2. Pooled counts: (9+1)/(10+2)*100% = 83.333333% rounded, exact fraction 5/6 of the assignments.
3. Separately declared custom weights 2 and 1: (90*2+50*1)/3 = 76.666667% rounded. Do not label this pooled. Denominator weights 10 and 2 reproduce the pooled answer.
Percentages alone support the equal-row mean of 70% but cannot recover pooled counts. Recompute from unrounded source values and round final output. Explain the question and denominator for each mode; actual custom-weight purpose/authorization is TO CONFIRM.
G03 has missing counts. Retain its row and incomplete-coverage status. Available G01/G02 answers must be labeled two-row only; full three-row aggregate unavailable. Missing input is not zero. Count 0/0 has an undefined row rate; zero completed with positive assigned count is 0%. Explicit weight 0 may exclude a row only in declared weighted mode with a reason; missing weight is not 0. All weights 0 means an undefined weighted mean.
Percent inputs: finite 0..100, up to six decimal places. Weights: finite nonnegative, positive total for a defined answer. Count mode: integers with 0 <= completed <= assigned <= 1,000,000 and positive assigned; require disjoint comparable groups. Flag invalid/missing rows and clear stale full-scope output; if a partial answer is shown, explicitly retain its limited coverage. Provide mode/basis/row/weight/missing-coverage ledger, calculations and error states. Native ledger remains authoritative. HTML, keyboard, boundary and rounding tests are later work, with no persistent edits or actual data collection claimed.
Review the mode beside the source rows
The local HTML example shows the three modes beside the source ledger and coverage warning. NOS supports saving and rendering prepared HTML. Keep the maintained source and aggregation records authoritative.
Explore NOS to present the prepared comparison. Report what received equal weight, which rows were covered and which denominator produced the answer.