Skip to main content
  • Status: accepted · Date: 2026-07-17 · Deciders: Davor Runje

​
Context

Not every mentor/reviewer voice suits every author. The question was how to pick a persona for the examine/advise voice and whether to match it to the author’s personality. A research pass separated grounded practice from folklore.

​
Decision drivers

  • Style typologies are solid; trait-matching is a myth (learning-styles debunk; Big-Five “fit” weak).
  • AI personality inference is unreliable and ethically fraught, and would violate agency (profiling without consent).
  • Autonomy support and calibrated challenge are the robust constants.

​
Considered options

  1. A small set of author-selectable personas (sounding board / critical examiner / directive editor / opt-in devil’s advocate), chosen by self- selection + task/stage suggestion + feedback calibration; never inferred.
  2. System infers personality and assigns a matched persona.
  3. A single fixed voice.

​
Decision

Option 1. Personas derived from supervision typologies (Lee × Gatfield), not personality theory. Autonomy support constant; only directiveness/challenge vary. Feedback targets the argument, not the person. Rule: match the voice to the task and the author’s stated choice, never to an inferred personality.

​
Consequences

  • Useful variety without pseudo-science or hidden profiling.
  • Task/stage suggestions are defaults the author overrides.

​
Rejected alternatives

  • Inferred matching — unsupported (learning-styles myth) + consent/agency violation + unreliable inference.
  • Single fixed voice — ignores real task/stage differences.
meta-spec §3.7; sub-spec 1 §6; digest mentor-personas.md.