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

​
Context

The dual of agency (ADR-0003): if the skills won’t do the research, the author must actually understand their work and methods to a mentor/reviewer standard — otherwise they can’t make good material decisions. Two examination objects emerged: the author’s own work, and the methodology the skills embody (anti-cargo-cult).

​
Decision drivers

  • Illusion of explanatory depth: self-assessed understanding is unreliable until probed; examination both reveals and (via retrieval/self-explanation) builds it.
  • Following rigor rituals blindly is cargo-cult science.
  • Goal is growth, not producing work the author can’t defend.

​
Considered options

  1. Second top-level principle + a cross-cutting defend tutor-examiner that verifies and teaches understanding.
  2. No explicit understanding check (rely on the author).
  3. A grader that scores understanding.

​
Decision

Option 1. The defend skill probes → teaches (source-grounded) → re-probes; self-invoked and fired as an automatic guardrail at material-decision checkpoints.

​
Consequences

  • Adds a cross-cutting skill and a guardrail at the three material decisions.
  • Must not assert an answer key for novel claims (see ADR-0015).

​
Rejected alternatives

  • No check — leaves the illusion of explanatory depth unaddressed.
  • Scoring — Goodhart-prone; would grade substance the AI can’t authoritatively judge.
meta-spec §2.2, §3.7; digest understanding-and-defense.md.