| name | move-detector |
| description | Identify academic tone moves in PhD-thesis corpus material and produce context-rich instances. Trigger when the user wants to detect, label, or catalog tone moves (defensive scoping, voicing the opposition, hedged confidence, synthesizing verdict, thesis-anchoring, etc.) in Type A academic writing, or asks to analyze the rhetorical/voice moves in a thesis paragraph. |
move-detector
LAYER 1 · SKILL · rule-bounded but judgment-heavy. Operates ONLY on Type A (external PhD theses).
Purpose
Find academic tone moves in Type A paragraphs and emit each as a concrete, context-carrying instance — never as a generalization.
Seed taxonomy (starting point; extend as the corpus reveals more)
- Defensive scoping — push a controversy out of the thesis's scope without seeming evasive.
- Voicing the opposition — render an opposing view vividly without endorsing it.
- Hedged confidence — tell the reader with authority that something is uncertain.
- Synthesizing verdict — deliver a composite judgment after reviewing a body of studies.
- Thesis-anchoring — tie a literature finding back to the author's own research question.
- Concessive judgment — "Even granting X, Y still holds."
- Field-internal critique — criticize peers without breaching collegiality.
- Methodological apology — concede a limitation without letting the claim collapse.
- Literature positioning — place oneself on one side of a scholarly spectrum.
- Confident assertion without citation — a meta-judgment about the state of the literature, uncited.
Per-instance output (every detected move)
- move type (from taxonomy)
- trigger context — what argumentative pressure the author faced at that moment
- full context — the target sentence PLUS the paragraph before and after. RL-3: the spirit of a move lives in its surroundings; quoting the single sentence alone is forbidden.
- intensity level — where this instance sits on the move's gentle→forceful range
- function served — what the move does within that paragraph's
section_function
Hard constraints
- RL-3 (no averaging): "this concrete instance, in this concrete situation" — never "this kind of move generally." Averaging is the #1 failure mode of AI tone analysis.
- When uncertain whether a paragraph instantiates a move, flag for human confirmation rather than force-fitting it into the taxonomy.
- Stay in the tone/voice/expression layer. Never comment on or alter the author's substantive argument (RL-1).
Self-assessment for scripting
Move detection is judgment and must stay LLM-driven. But auxiliary steps may be scriptable after a real run — e.g. taxonomy tallying, intensity-distribution stats, building an index of instances by move×section_function. If volume justifies it, follow SCRIPTING-POLICY.md.
