Ask for the strongest supported objection
An adversarial review can help expose an unstated assumption, a missing qualification, or a gap between a proposition and its evidence. Asking a model to act as a hostile judge is less useful than specifying exactly what it should inspect. The goal is a list of testable objections, not a performance of hostility or an estimate of who will win.
The workflow here is an original teaching exercise. It uses the general evaluation principle of defining what a reviewer should assess and inspecting the result against external criteria. A model-based reviewer also needs checking. [source]
Give the reviewer a bounded record
Use a draft and the underlying sources. Identify the issue the draft addresses and separate assumptions from established facts. If you provide only your preferred summary, the reviewer may simply reason within your framing. A confident criticism built on a missing document is no more reliable than a confident supporting argument built on one.
For the fictional Alder exercise, suppose a draft says the supplier promised delivery by 1 April. The source instead discusses possible delivery in the first week of April, subject to inspection. An effective review should identify those differences directly.
Review the supplied argument against the supplied record.
Return up to five material objections, strongest first.
For each: identify the draft passage, objection, supporting source
and location, possible response, and unresolved evidence question.
Distinguish an actual contradiction from a hypothetical alternative.
Do not invent authorities, facts, or concessions to make an objection.
If a point survives review, say so without manufacturing a criticism.Keep criticism connected to evidence
| Review finding | Useful follow-up |
|---|---|
| A proposal is described as an agreement | Compare the exact source wording and any acceptance |
| An argument depends on an unseen attachment | Locate the attachment or preserve the assumption |
| The draft ignores a conflicting account | Identify both passages and the unresolved difference |
| A reviewer alleges a missing fact | Ask whether it is absent from the record or merely absent from the draft |
That last distinction matters. The draft may properly omit an irrelevant fact. Conversely, a fact that would be helpful cannot be assumed simply because a reviewer says the argument needs it.
Resolve each objection before redrafting
Create a disposition beside each criticism: accept, reject with reason, investigate, or preserve as uncertainty. Revise only the points that warrant a change. Otherwise an articulate reviewer can pull a sound draft toward a weaker argument merely by sounding assured.
A second model or separate session can provide another perspective, but agreement between reviewers is not independent corroboration of a fact. They may receive the same incomplete sources or make the same inference. Locate the original evidence and verify any legal authority separately before relying on it.
Try it: the reviewer invents a concession
A reviewer says, “The supplier conceded the date was binding.” You cannot find that concession in the supplied documents. Should the draft anticipate it as a fact?
Check your understanding
No. Require the source and location. If none exists, reject the criticism as unsupported or recast it as a clearly labelled hypothetical. Do not allow the review stage to contaminate the factual record. A successful exercise produces better questions and justified revisions; it does not certify the argument, predict the court's decision, or replace professional judgment.