A review rubric for AI-generated interpretations
An AI interpretation can be easy to read and difficult to justify. A review rubric should give the reader specific questions to ask of each claim, especially when the answer moves from a supplied card or chart to statements about a person.
Use this rubric with one short answer first. Keep the original question and inputs beside it. The aim is to identify repairs and unresolved issues, so a numerical score is optional. A low score tells a writer less than a note identifying the sentence that used the wrong input.
Check whether the answer used the supplied material
Compare every named card, position, chart field, and date with the original record. Mark a mismatch as an input error before discussing interpretive quality. If the response names a card that was never drawn, its elegant explanation does not repair the mismatch.
In the inspected Mystilink tarot implementation, the tool result records the spread and cards and can explicitly indicate a skipped draw. That result is a useful comparison point in a session using this implementation. Verify how the relevant build exposes it to the reader. Source code for a result format does not establish that every interface displays the same level of detail.
For a fictional three-card practice answer, a reviewer might write: "Input fidelity needs revision: the answer interprets the final card as the opening position." This note gives the author a bounded repair. Recheck the surrounding paragraphs after the repair because the interpretation may still depend on the old order.
Separate source claims from interpretive choices
Underline claims about a tradition, author, historical object, or calculation convention. Each should have a checkable basis if it matters to the explanation. A citation should lead to the relevant passage or object record, with enough information to identify the edition when editions differ.
Then mark the writer's interpretive steps. A sentence such as "For this exercise, I read the enclosed space as a reason to examine boundaries" states a choice the reader can discuss. A sentence claiming that all practitioners assign that meaning needs much stronger support. Do not let a reference attached to one factual sentence lend authority to unrelated personal claims nearby.
If the source is unavailable, the reviewer can request a citation, remove the attribution, or narrow the sentence. Avoid filling the gap with an unnamed expert or an invented consensus. A short answer with a visible limitation is easier to check than a longer one with uncertain authority.
Examine the strength of each conclusion
Look for words that claim certainty, inevitability, or access to another person's private thoughts. Ask what supplied evidence permits that strength. A symbolic association alone does not establish that a future event will occur or that someone has a particular motive.
A useful revision states the scope of the interpretation. It can say that an image offers a prompt for considering a difficulty, or that one school would examine a particular relationship between chart features. Those statements remain open to comparison. They should not be rewritten into factual diagnoses of the reader.
Also check negative evidence. If the reader says the description does not fit, preserve that response. Treating every disagreement as further confirmation makes the interpretation impossible to challenge. A learning record should allow an interpretation to remain unsuccessful or unresolved.
Assess whether the answer helps the actual task
Return to the question. If the user asked how to organize a study note, a long personal prediction does not complete that task. If the question asked for a comparison between two sources, the response should explain the difference and its consequences for reading them.
Practical suggestions should be modest enough to evaluate. "Record which edition supplied this image before comparing the descriptions" gives the learner a clear next step. Advice to reorganize major life decisions on the authority of a reading needs a different level of evidence and responsibility than a study exercise can provide.
Check whether the proposed action belongs to the reader. An interpretation can invite reflection while allowing the reader to reject the suggestion. Avoid presenting compliance as a test of whether the reader understands the symbols correctly.
Review presentation after substance
Only after checking inputs and claims should you edit the prose. Remove repeated conclusions, ornamental headings, and declarations of importance that add no information. Replace generic praise with a specific observation. Preserve meaningful uncertainty when shortening a paragraph.
Use examples where they resolve ambiguity. One sentence showing a card-position error can teach more than several paragraphs about careful review. Label every constructed example as fictional so readers do not mistake it for a customer account or a verified outcome.
Finish with a short decision: ready for this intended use, revise specific claims, or hold pending a source check. Add the reviewer and date in your own records if a team is responsible for publication. The rubric is a proposed editorial practice; it does not claim that Mystilink currently performs automatic validation or assigns these review decisions within the application.