Verified against ChatGPT · 2026-08-14
Reverse-engineer a grading rubric from an assignment sheet that doesn't come with one
Builds a likely rubric from an assignment prompt and any past graded work you have, so you know what's actually being weighted before you submit instead of finding out from the grade.
The prompt
Ready to copy — highlighted parts are example details you can swap.
My teacher gave me an assignment prompt with no rubric attached. Build me a likely rubric so I know what's actually going to be weighted before I submit, based on what the prompt itself emphasizes and any past graded work from this teacher I can show you as a pattern. ASSIGNMENT PROMPT (verbatim) "Write a 4-6 page analysis of a primary source document of your choosing from this unit. Your analysis must contextualize the document's origin, evaluate its reliability as a historical source, and connect it to at least two course themes. Proper Chicago-style citations required throughout." CLASS AND ASSIGNMENT TYPE AP World History, primary source analysis paper, third one this semester. PAST GRADED WORK FROM THIS TEACHER, IF ANY (what was rewarded or marked down) Last paper lost points mainly for citation formatting errors and for summarizing the source instead of evaluating its reliability; got full marks on the thematic connections section. Build the rubric as 4-6 criteria, each with a name, a one-line description of what earns full marks versus what's a common way to lose points on it, and a rough weight relative to the others (you don't need exact percentages, but rank them heaviest to lightest). Base the weighting primarily on what the prompt itself spends the most words and emphasis on — a prompt that spends three sentences on how sources must be cited and one sentence on formatting is signaling that citations matter more, even if it never says so explicitly. If I gave you past graded feedback, use it to check or adjust your guessed weighting rather than ignoring it, since a teacher's actual past grading pattern is stronger evidence than a first read of this one prompt. Be explicit about your confidence: mark any criterion you're inferring mostly from general norms for this assignment type, versus one you can point to a specific sentence in the prompt for. Do not present a guessed rubric as if it were the teacher's actual rubric — say clearly that this is your best inference and that confirming with the teacher directly, especially on anything you flagged as low-confidence, is worth doing if there's time before the deadline. Output as a rubric table: criterion, what earns full marks, common ways to lose points, relative weight, and a confidence flag (high/medium/low) per row. End with one line naming the single highest-weighted criterion and the one thing most likely to cost the most points if missed.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
When an assignment prompt has no attached rubric, GPT-5.1 asked to just "explain the assignment" tends to restate the prompt's stated tasks with roughly equal weight given to each, because without an explicit instruction to look for emphasis signals it treats the prompt as a flat list of instructions rather than as a document that itself encodes priority through how much space and specificity it gives each requirement — instructing the model to weight based on relative emphasis (three sentences on citation standards versus one on formatting) extracts a real, usable signal from a document that was never designed to be machine-read for grading weight, which is closer to how an experienced student actually reads an assignment sheet than a naive equal-weighting summary. Feeding in past graded feedback from the same teacher matters because a single assignment prompt is a weak predictor of actual grading behavior on its own — teachers routinely emphasize things in the prompt they don't end up weighting heavily in practice, and vice versa, so real historical evidence of what specifically cost or earned points from this teacher should override a first-pass reading of the prompt's own language when the two conflict, and the instruction explicitly tells the model to treat the historical pattern as stronger evidence rather than averaging it in as one more equal input. The required confidence flag per row exists because a guessed rubric presented with uniform authority is actively misleading — a student who treats every row as equally certain might spend their limited revision time on a low-confidence guess instead of the high-confidence, directly-stated-in-the-prompt requirement that's actually safest to prioritize, and marking the distinction lets the student decide where confirming with the teacher directly is worth the extra step before a deadline.
What you get back
Criterion: Source reliability evaluation | Full marks: Explicitly assesses bias, provenance, and limitations of the source, not just summarizing its content | Common point loss: Describing what the source says instead of evaluating how trustworthy or limited it is as evidence | Weight: Highest | Confidence: High (prompt explicitly separates 'evaluate reliability' from 'contextualize origin' as distinct tasks, and past feedback confirms this was penalized before).
Verified against
ChatGPT GPT-5.1 · 2026-08-14
Changelog
- 2026-08-14 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
