Skip to main content
Date: 2026-07-17 · For: honest-scholar defend skill + the “advise as a mentor” voice (meta-spec §3.7, sub-spec 1). · Status: verified-source digest; migrates to the plugin’s resources/references/. Bottom line: style typologies are well-described; matching a mentor to the author’s personality is not evidence-based (the learning-styles myth); and an AI inferring personality to pick a voice is unreliable and violates agency. The defensible design is author-selectable personas + situational/task calibration — never system-inferred personality.
One-line rule for the spec: match the voice to the task and the author’s stated choice — never to an inferred personality.

​
1. Style typologies (descriptive — solid)

  • Gatfield (2005) — structure × support 2×2: laissez-faire / pastoral / directorial / contractual. J. Higher Education Policy & Management 27(3):311–325, DOI:10.1080/13600800500283585.
  • Lee (2008) — five concepts of supervision: functional, enculturation, critical thinking, emancipation, relationship development. Best source for a reviewer-voice spectrum (critical-thinking/emancipation ≈ the defend examiner). Studies in Higher Education 33(3):267–281, DOI:10.1080/03075070802049202.
  • Kram (1985) — mentoring functions: career (challenge, sponsorship) vs psychosocial (role-modeling, acceptance). ISBN 978-0673156174. Higgins & Kram (2001) — developmental networks (a portfolio, not one dyad), AMR 26(2):264–288, DOI:10.5465/amr.2001.4378023 (verify).
  • Gurr (2001) — “rackety bridge”: dynamically align directiveness (hands-on↔hands-off) to the student’s growing competent autonomy. The developmental-matching model, doctoral-specific. HERD 20(1):81–92, DOI:10.1080/07924360120043882.

​
2. Feedback science (efficacy — solid)

  • Hattie & Timperley (2007), “The Power of Feedback,” Review of Educational Research 77(1):81–112, DOI:10.3102/003465430298487. Effective feedback answers where am I going / how am I going / where next; feedback at the self/praise level is ineffective or harmful, at the process/self-regulation level most effective. → the real axis is challenge the argument/process, not the person — not “harsh vs. nice.”

​
3. Matching to the author — grounded vs. myth (the crux)

  • Learning-styles matching is a MYTH. Pashler, McDaniel, Rohrer & Bjork (2008/09), PSPI 9(3):105–119, DOI:10.1111/j.1539-6053.2009.01038.x — the “meshing hypothesis” fails the crossover-interaction test that would validate it. Nancekivell et al. (2020), JEP 112(2):221–235, DOI:10.1037/edu0000366 (myth is pervasive + essentialized). Do not build trait-based persona auto-assignment.
  • Big Five ↔ style “fit”: weak/mixed — mostly main effects (agreeable/conscientious mentors → better relationships), no reliable matching interactions. Not actionable matching science.
  • Better-grounded alternatives:
    • Situational/developmental — calibrate directiveness to competence+stage (Gurr 2001; Hersey & Blanchard SLT, ISBN 978-0132556408 — SLT’s own evidence is mixed; use as scaffold, not mechanism).
    • Self-Determination Theory autonomy-support — the robust constant posture: support autonomy, competence, relatedness. Ryan & Deci (2000), American Psychologist 55(1):68–78, DOI:10.1037/0003-066X.55.1.68.
    • Desirable difficulties / calibrated challenge — Bjork & Bjork (2011). Keeps the defend skill’s productive-struggle bite.
  • Supervisor–student fit matters — but it’s relationship/expectation alignment, not personality matching — mismatch drives attrition/dissatisfaction (Sverdlik et al., IJDS 13:361–388).

​
4. Ethics — no AI personality inference

LLM personality inference is contested on validity and ethically fraught (consent, bias, manipulation): arXiv:2503.02082, arXiv:2405.13052. A system that silently profiles the author to pick a voice is both unreliable and an agency/consent violation. The author chooses; the system does not psychoanalyze.

​
5. Design (as settled)

A small set of author-selectable personas (derived from Lee × Gatfield, not personality theory):
  1. Sounding board — pastoral / relationship-development / high autonomy-support; early, exploratory.
  2. Critical examiner — the default examiner (critical-thinking + emancipation; calibrated difficulty).
  3. Directive editor — directorial / functional; concrete, process-level (Hattie & Timperley), deadline-driven.
  4. (opt-in) Devil’s advocate — time-boxed; explicitly challenge-to-the-argument.
Chosen by three author-controllable levers, never inferred:
  1. Self-selected (default) — the author picks. Satisfies agency; sidesteps the myth + inference ethics.
  2. Task/stage-suggested — a suggested default keyed to the visible artifact/stage (early draft → sounding board; near-submission → examiner), à la Gurr; always overridable; keyed to the work, not the person.
  3. Feedback-calibrated — adjusts to explicit author feedback (“too harsh” / “push harder”); feedback content always targets process/argument, not self.
Invariants: autonomy-support (SDT) is constant across all personas; only directiveness and challenge-intensity vary. Hidden personality profiling is explicitly forbidden. The defend skill keeps its bite (Bjork), but the author holds the difficulty dial (agency + Understanding principle).

​
Sources

Gatfield 2005 (13600800500283585) · Lee 2008 (03075070802049202) · Kram 1985 (ISBN 978-0673156174) · Higgins & Kram 2001 (amr.2001.4378023, verify) · Gurr 2001 (07924360120043882) · Hattie & Timperley 2007 (003465430298487) · Pashler et al. 2008/09 (j.1539-6053.2009.01038.x) · Nancekivell et al. 2020 (edu0000366) · Ryan & Deci 2000 (0003-066X.55.1.68) · Hersey & Blanchard (ISBN 978-0132556408) · Bjork & Bjork 2011 · Sverdlik et al. (IJDS 13:361–388) · AI personality inference (arXiv:2503.02082, arXiv:2405.13052).

​
Flags

Higgins & Kram DOI worth confirming. Pashler et al. 2008/2009 date ambiguity (one paper). Hersey–Blanchard SLT and Big-Five-matching are where popular claims outrun evidence — treated as such. Big-Five/fit sources are representative secondary literature, not single authoritative primaries.