River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Faster Grading and Feedback Template

Three documents and three sheets that mine your own marking for the comments you keep retyping, then price what banking them saves.

Free download  ·  No account needed

Marking a script is four activities of very different sizes, and a generic phrase bank saves nothing because it does not match where the time goes. On the worked cohort in this pack, a fictional writing course with 138 scripts, the per-script breakdown came out at 4.2 minutes reading, 6.1 minutes writing comments, 1.4 minutes scoring against the rubric and 0.9 minutes admin. That is 12.6 minutes a script and 29.0 hours for the cohort.

Comment writing is 48.4 per cent of it, and it is the only block with slack, because most of the comments are repeats. Extract every comment from a marked cohort and cluster them by the error they name and the shape is stark: 1,742 comments across 138 scripts collapse into 47 distinct types, and twenty of those types cover 1,166 comments. The single most common comment appeared on 118 of the 138 scripts, composed from scratch every time.

So the bank is those twenty, each written once with the criterion, the error, the fix and a next step. Four hours to build, 4.86 hours saved per round, taking the cohort from 29.0 hours to 24.1. That is a 16.8 per cent reduction rather than a transformation, and the honest number is the point: a feedback bank does not halve marking, it removes the retyping. The bigger win is the second artifact the same clustering produces.

The comment that was typed from scratch 118 times in one marking round

Three tabs from the pack: the bank clustered out of real marking, the register that routes a repeated error into teaching, and the four-stage time model with its assumptions stated.

Feedback Bank

Built backwards out of one cohort’s marking. ENGL 215 at a fictional college: 1,742 comments across 138 scripts, clustered into 47 types. The top twenty cover 1,166 comments, 66.9 per cent of everything written.

IDCriterionThe error, as the script shows itUsedScriptsThe fix
C-01ArgumentClaim stated in the introduction is not the claim the essay argues11885.5%Say which claim the essay actually defends, then rewrite the introduction to state that one
C-02Use of evidenceSource summarised rather than used as evidence for the claim9669.6%Follow each summary with the sentence that says what it proves
C-03StructureParagraph has no topic sentence, so the point arrives at the end9165.9%Move the sentence carrying the point to the front of the paragraph
C-04ArgumentCounterargument named but not answered8460.9%Answer the objection on its own terms, then say why the claim survives it
C-05Use of evidenceQuotation dropped in without introduction or interpretation7957.2%Frame the quotation, then say what it means for the claim
C-06ReferencingCitation in the text and missing from the reference list7453.6%Add the entry and check every in-text citation before submitting
C-07StructureConclusion opens a new argument instead of closing the one made6849.3%Close the argument you made rather than opening a new one
C-08Use of evidenceEvidence supports a weaker claim than the one made6144.2%Either strengthen the evidence or weaken the claim so the two match
SUBTOTAL20 typesBanked, written once at ~12 min each116666.9%4.0 hours to build
TAIL27 typesUsed 30 times or fewer each; composed per script57633.1%Not banked

The most common comment was composed from scratch 118 times in one marking round. A banked entry carries the criterion, the error, the fix and the next step, and it never goes out without one sentence naming what this script actually did.

Common Error Register

Same clustering, different job. An error on more than half the scripts is a teaching gap, not something to write 118 times.

Common errorScriptsShareVerdictWhere it goes next term
Claim in the introduction is not the claim argued11885.5%Teaching gapSession 3, thesis workshop
Source summarised rather than used as evidence9669.6%Teaching gapSession 5, evidence clinic
Paragraph has no topic sentence9165.9%Teaching gapSession 4, paragraph drill
Counterargument named but not answered8460.9%Teaching gapSession 7, objection handling
Quotation dropped in without interpretation7957.2%Teaching gapSession 5, evidence clinic
Citation in the text, missing from the reference list7453.6%Teaching gapA submission checklist, because no teaching fixes a checking failure
Conclusion opens a new argument6849.3%Watch, close to the thresholdSession 9, closing moves
Evidence supports a weaker claim than the one made6144.2%Individual feedbackFeedback bank only
TOTAL6 above the 50% threshold4 sessions gain content, 1 goes to a checklist

The threshold is stated rather than assumed, and anything within ten points of it is flagged rather than rounded away. The referencing row is the useful exception: high frequency, and no amount of teaching fixes it.

Marking Time Tracker

Four stages, measured rather than estimated, so it is visible which one the bank can touch and which three it cannot.

StageMin per scriptCohort hoursShare of marking timeAfter the bankCohort hours after
Reading the script4.29.6633.3%4.29.66
Writing comments6.114.0348.4%3.989.15
Scoring against the rubric1.43.2211.1%1.43.22
Admin, files and mark recording0.92.077.1%0.92.07
TOTAL12.628.98100%10.4824.10

Assumptions, written into the sheet: 26 seconds per repeated comment and 35 per bespoke one before the bank, 11 seconds to select and personalise after. Saving 4.86 hours a round against a 4.0 hour build, which is 16.8 per cent of total marking time.

What is in the pack

01

Feedback Bank

Twenty entries, clustered out of a real cohort's comments rather than written from imagination. Each carries the rubric criterion, the error described in what the script does rather than what it lacks, a fix specific enough to act on, and the next step for the next piece of work. The 27-type tail is deliberately left unbanked, because a comment used twice does not repay being written well.

02

Common Error Register

The same clustering with a different job. Any error appearing on more than half the cohort's scripts is a teaching gap rather than a marking issue, and the register names the class session that gains the content. One high-frequency error deliberately does not go to teaching: a citation present in the text and missing from the reference list is a checking failure, and it goes to a submission checklist.

03

Marking Time Tracker

Reading, comment writing, rubric scoring and admin measured separately, because only one of the four has slack in it. Both seconds-per-comment assumptions are written into the sheet, so the 4.86 hour saving and the 4.0 hour build cost are arguable numbers rather than a promise.

04

How the Feedback Bank Is Built

The mechanism doc and the arithmetic. Why the bank is built backwards, why it is the top twenty types rather than the top five or all 47, and why the eleven seconds a banked comment costs is eleven rather than two. Personalisation is counted into the model, so a bank used without it is worth less than nothing.

05

Marking Procedure

The order the work happens in. Three scripts marked against the rubric before anything is recorded, marking in one pass rather than two, every comment named to a criterion, and a hard cap of three sentence-level style notes per script. Marking every instance of a recurring style error costs an hour a cohort and teaches nothing the first three did not.

06

Moderation Guidance

A deliberate sample rather than a random one: the first five marked, the last five, everything within two points of a boundary, and both extremes. Then four checks, including whether the comments alone predict the band. Order drift gets scripts remarked rather than adjusted by a constant, because a constant moves the correct scripts too.

How it works

  1. 1

    Send a previous cohort's marked work

    Annotated PDFs, an LMS comment export, a rubric tool's log or Word files with tracked comments, plus the rubric they were written against. Volume matters more than recency, and one full cohort is enough.

  2. 2

    Cluster the comments by the error they name

    Not by wording. Two comments are one type when the same fix answers both. The output is four numbers: total comments, distinct types, the share the top twenty cover, and the share of scripts your single most repeated comment appeared on.

  3. 3

    Bank the top twenty and price the change

    Each written once with criterion, error, fix and next step. Then the time model against your own measured minutes, quoting the reduction as a share of total marking time rather than of comment-writing time.

  4. 4

    Route the recurring errors out of marking

    Anything above half the cohort goes to a class session or a submission checklist with a named owner. The rubric pack is where a criterion that turned out to be hard to apply gets rewritten.

Frequently asked questions

Is this template free?

Yes. Download the three documents and three sheets as Word and CSV files with nothing to sign up for. Opening it as a working space, where the clustering runs against your own marked scripts and the time model recomputes, needs an account. The rest sit in the template library.

What format are the downloaded files?

Three Word documents and three CSV sheets in one zip. The sheets open in Excel, Numbers or Google Sheets and arrive carrying the worked cohort, so the clustering shape and the four-stage time model are visible before you replace them with your own numbers.

Does a feedback bank make feedback worse?

It does if a banked comment goes out unchanged, which is why that is a rule in the space rather than advice. The banked text carries the criterion, the error and the fix; the marker adds one sentence naming what this script actually did. Students compare feedback within a day.

How much time does this actually save?

On the worked cohort, 4.86 hours a round against a 4.0 hour build cost, taking 138 scripts from 29.0 hours to 24.1. That is 16.8 per cent of total marking time. Quoting the comment-writing figure instead would have looked like 34.6, which is the number to distrust.

Why does an error on half the scripts leave the bank?

Because writing it 118 times is the most expensive way to address it available. The Department for Education's marking review argued that marking should be meaningful, manageable and motivating, and a cohort-wide error fails the first test as individual feedback.

We have no record of how long marking takes. Can we still start?

Yes, but time the first twenty scripts of the next round, because every claim in this space rests on that number. Without it, the bank is still worth building from the clustering alone, and the saving is a guess rather than a measurement.

How is this different from an item analysis?

An item analysis works on objective questions and asks which items discriminate. This works on written comments and asks which you keep retyping. For exams, item analysis review is the right tool, and the question bank pack rewrites the items. Whether the marking lands consistently once more than one person applies the rubric is a separate check, in grade analysis and moderation.

Find the comment you wrote a hundred times

Send a previous cohort's marked scripts with the comments still on them. The first pass clusters every comment by the error it names and tells you what share of your marking was retyping.

Mine a real cohort