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

​
Context

An AI research-workflow tool could plausibly try to do research and make scientific decisions. The author insisted the opposite: the skills assist, keep accounts, advise, and discuss; the human makes every material decision and is accountable. Grounded by a dedicated research pass.

​
Decision drivers

  • Research accountability is non-delegable and attaches only to named humans.
  • Automation bias / over-reliance is a documented failure mode (esp. LLM citation fabrication).
  • The author wants growth and ownership, not push-button papers.

​
Considered options

  1. Assistant/advisor; every material decision is a human’s, recorded with a named sign-off.
  2. Autonomous/semi-autonomous agent that proposes verdicts/decisions.

​
Decision

Option 1, elevated to the plugin’s highest guiding principle.

​
Consequences

  • Material decisions (findings verdict, publish decision, thesis defensibility) carry a named human sign-off; the backend stamps evidence but never adjudicates it.
  • Drafts are proposals; the author authors. No unattended end-to-end generation.

​
Rejected alternatives

  • Autonomous adjudication — violates ICMJE/COPE authorship-accountability norms, EU AI Act Art. 14 human-oversight, and invites automation bias.
meta-spec ⚑ guiding principle + §2.1; digest agency-principle.md.