Backed by Y CombinatorTwo documents and two sheets that classify every protocol change the moment it happens, before the affected work makes the registry's own amendment option disappear.
Free download · No account needed
What you get
Pack of 4 files: 2 documents, 2 spreadsheets
PROSPERO Protocol Registration TemplateAmendment Register
| Not yet | Already happened |
|---|---|
| Amendment — submit a registry revision now, before that work starts | Deviation — nothing to submit; log it and disclose it in the manuscript |
One fact sorts every change. A registry's own amendment mechanism updates the plan going forward; it cannot reach back and change work already done.
A prospective registry captures one snapshot: the review question and the planned methods, filed before screening starts. PROSPERO's own interface offers two ways to update that snapshot afterward, a minor amendment for administrative fields and a major revision for the review methods, and both only work forward: they update something not yet done. Neither can touch work already finished under the old plan. PRISMA 2020's own checklist routes that kind of change somewhere else entirely, item 24c, a manuscript paragraph, not a registry field.
The pack ships filled in for a review of structured discharge-planning interventions and 30-day readmission. Five protocol changes get logged across its timeline, sorted by one test: had the affected work already happened when the change was decided? Three had not: a database access change, a risk-of-bias tool swap, an added outcome, each submitted as a registry amendment first. Two had already happened: a primary-outcome definition narrowed after 9 of 14 studies were extracted, and a subgroup dropped after extraction finished on all 14. Neither had anywhere to go but the log.
Most PROSPERO guidance treats registration as one event: fill in the fields, submit, done. A few mention keeping a change log as good practice, without saying which changes even have a registry mechanism to use. This pack classifies at the moment of the decision, then routes each one correctly. It registers the plan; it does not build the search behind it, which the search strategy pack does, screen records, which the screening pack does, or appraise study quality, which the critical appraisal pack does.
Protocol Item Checklist4 of 17 required fields, chosen to show which ones changed after registration.
| Field | Current value | Changed since registration |
|---|---|---|
| Searches | MEDLINE, Embase and CENTRAL | Yes — 2025-11-10 |
| Primary outcome | Unplanned readmission, 30 days | Yes — 2026-05-08 |
| Risk of bias tool | Cochrane RoB 2 | Yes — 2026-01-15 |
| Types of study | Randomized controlled trials | No |
Every changed field points at the Amendment Register row that explains it, so the checklist's current value and the log never drift apart.
Amendment RegisterAll 5 logged changes across the review's timeline.
| Date | Field | Classification | Registry action |
|---|---|---|---|
| 2025-11-10 | Searches | Amendment | Minor amendment |
| 2026-01-15 | Risk of bias tool | Amendment | Major revision |
| 2026-02-20 | Secondary outcomes | Amendment | Minor amendment |
| 2026-05-08 | Primary outcome | Deviation | None — already extracted |
| 2026-06-12 | Subgroup analysis | Deviation | None — already extracted |
3 amendments went through the registry before the affected work started. 2 deviations never had a registry mechanism to use, because the extraction they affected was already finished when each was found.
Every field written in the order a registry form asks for it, ready to paste, always showing the current text with a dated note on anything changed since the original submission.
One row per required field with its current value, status and the date last confirmed, so a field silently left blank cannot pass for one deliberately marked not relevant.
The seven fields every logged change needs, and why the manuscript's disclosure paragraph is a query against this log rather than a document drafted separately at submission.
One dated row per change, classified by whether the affected work had already happened, with the reason, who decided it, and where it gets reported.
The standing space rule every prompt reads first, which is the one test that sorts every change into a registry submission or a log entry with nowhere else to go.
The review question and planned methods, plus title, timescale, funding and conflicts of interest, drafted straight into registry-ready fields.
Whatever you report, big or small, gets classified immediately: had the affected work already happened when you decided this? That single fact sorts it.
Not yet happened gets drafted as a minor amendment or major revision for your registry's own interface. Already happened gets a dated log entry, because there is nothing to submit.
Every logged row, amendment or deviation, becomes one sentence in date order, so the manuscript's own protocol-and-registration section is a transcription, not a reconstruction.
Yes. Download the whole pack as Word documents and CSV sheets with no signup and no credit card. Edit with AI is a separate, optional path that has the agent draft the registry fields and log changes with you. The template library holds the rest of the packs.
The two write-ups download as Word files and the two sheets as CSV, zipped together. Every format opens natively in Word, Pages, Google Docs, Excel, Numbers or Sheets, so nothing needs converting before your team can use it. Open the same pack in River instead, and it becomes a live Doc and Sheet you edit directly.
The form captures one snapshot. This tracks what happens for the months afterward, and it applies the one test that decides where each change goes: had the affected work already happened when you decided it? Not yet is a registry amendment; already happened is a deviation with nowhere to go but a log and, later, the manuscript.
PROSPERO is the worked example because it is the largest registry for health-related systematic reviews and the one most readers of this page use. The same fields, the same amendment-versus-deviation test and the same disclosure paragraph apply to any registry, since the underlying reporting requirement comes from PRISMA 2020, not from one platform's form.
Timing, not size. A change decided before the affected work starts is an amendment, submitted to the registry while there is still time. A change discovered only after that work is already done is a deviation; the registry's own revision mechanism has nothing forward-looking left to update, so it goes in the log and the manuscript instead.
No. It drafts the revision note and the updated field text in the registry's own style, and you paste that into your registry account and submit it yourself, since that step requires your own login and your co-authors' approval. This space keeps the record of what was drafted, submitted and when.
Registration still works. Log any changes already made before today as accurately as you can reconstruct them, dated to when they actually happened rather than to today. That way the record does not read as though the review started clean and everything changed at once on registration day.
Written by
Co-founder & CEO, River
Chandler has spent the last six years working in AI, building tools for millions of users and Fortune 100 companies. In 2025, he co-founded River to build a new kind of operating system for AI agents.
Send your review question and planned methods. Every field gets registry-ready text, and every change from here gets classified the moment you report it.
Edit with AI