Revising a Research Prompt Without Losing Scope
Tightening a research prompt usually means narrowing it — the diff shows whether your revision sharpened the scope or silently changed the question.
Overview
Research prompt revisions have a specific failure mode: in the act of adding criteria and structure, you accidentally change what's being asked. Narrowing 'project tools' to three named candidates is sharpening; dropping the budget constraint while doing it changes the answer you'll get. Diffing the revision separates the two — additions show the sharpening, removals show anything that fell out of scope on the way. The loaded pair includes one of each, which makes it a good calibration exercise.
Workflow
-
Diff the loaded pair
Version B names candidates, adds criteria, source rules, and an output shape — clear sharpening, all in the added list.
-
Find the scope leak
The $500/month budget cap is in the removed list. Nothing in B replaces it — the recommendation can now exceed budget.
-
Patch and re-diff
Add the budget back as an evaluation criterion ('total cost at 12 seats vs the $500/month cap') and diff again — removals should be empty.
-
Apply to your revisions
Rule of thumb: in a research prompt revision, every removal needs an explicit replacement or an explicit decision to drop it.
Why This Works
- Additions and removals separate 'sharpened the question' from 'changed the question'
- Budget and scope constraints are the most common silent casualties of research prompt rewrites
- Re-diffing after the patch verifies the fix instead of assuming it
Best for
- Decision-support research where the question must stay stable across revisions
- Prompts revised by someone other than the person who owns the decision
- Long research prompts where a dropped constraint hides easily
Not for
- Comparing two competing research prompts for quality — that's the Prompt Comparator
- Restructuring a rambling research request — that's the Prompt Formatter
Use cases
- Reviewing a research prompt revision before spending a long model run
- Checking that added criteria didn't displace original constraints
- Teaching the sharpen-vs-shift distinction with a concrete pair