Fixed-point reminder: the author decides whether to run a systematic review (never inferred, never a default)

Problem

Whether to run a systematic review is a research-design decision. It costs the author months of work, at least two reviewers, and often a registered protocol. ARS already enters systematic-review mode only when the user asks for it (deep-research/SKILL.md, Mode Selection Guide). Users who do not know the option exists, or do not know their question fits it, never hear about it.

With #919 adding a screening skill, the gap matters more: the suite will support the full review workflow, but only for users who already know to ask.

Principle

The author decides. ARS may remind; it does not judge whether a question "fits" a systematic review, and a systematic review never becomes a default step.

Deciding when to show a reminder is itself a judgment. If the reminder appears only when the model thinks a question looks like an effect question, the model is still choosing. So the reminder is tied to a fixed point in the workflow, not to the content of the question.

Proposed design

At one fixed point, every user sees the same short, neutral note:

• It lists the review forms: systematic review, scoping review, narrative or integrative review, rapid review, and no formal review.
• Each form gets one line: what kind of question it serves and roughly what it costs.
• It states that ARS does not choose for the author.
• It has no default, no ranking, and no "recommended" label.
• It appears once per project or run and can be skipped (see Fixed points).

The author's choice is recorded in the author's own words (in the run ledger when a passport file exists). Choosing a systematic review hands over to systematic-review mode, and later to screening, only on that explicit choice.

Fixed points

The note can appear at two points. Both are actions the author takes, not something the model infers:

1. When the author selects lit-review mode.
2. When the research question is finalized (RQ Brief completion).

Rules:

• Whichever point comes first shows the note. Once the author chooses or skips, the other point does not show it again. Skipping counts as a decision.
• The note does not appear if the author has already chosen a review form themselves, for example by entering systematic-review mode or asking for a scoping review. This depends on the author's action, not on the content of the question.
• Remembering the answer across sessions relies on the run ledger when a passport file exists. Without one, an author who works across separate sessions may see the note again. This is accepted: an extra reminder costs seconds, a missed one costs more.

Acceptance criteria

• The note text is identical for every user and every question at the chosen point.
• Whether the note appears does not depend on the content of the question.
• No option is preselected, ranked, or labelled as recommended.
• Shown at most once per project or run across both points; skipping is one step and counts as a decision.
• Not shown when the author has already chosen a review form explicitly.
• No path enters systematic-review mode or the screening skill without the author's explicit choice.
• Routing fixtures include negative cases, for example "help me write the literature review" routes to neither systematic-review nor screening.

Out of scope

• Any model assessment of whether a question suits a systematic review.
• Changes to the screening skill proposed in #919.