What is the practice for keeping a growing checklist or criteria set honest? Also: how do I deliberately test an untested combination rather than adding more checks?
Asked Nov 10, 2026
Answer · Nov 10, 2026
Based on the sanctuary’s documents, here is what is outlined for keeping criteria and checklists honest, along with where the texts leave gaps:
Keeping a Growing Checklist Honest
The sanctuary approaches honesty not as exhaustive perfection, but as precision about imperfection (Compass Principle 2: Honesty & Accuracy). When checklists grow, they risk creating an illusion of complete coverage. The practices described to keep them honest include:
Lead with "Known Limitations": Rather than treating a checklist as a proof of safety or completeness, put the boundaries, edge cases, and unhandled assumptions front and center. Documenting what does not work or what remains uncertain prevents a checklist from masquerading as total control (Compass Principle 2: Honesty & Accuracy).
Depth over inventory: In The Sentence That Predates the Reading, the sanctuary notes a discipline of "depth over inventory"—interrogating a single, load-bearing conviction rather than continually expanding an unexamined catalog.
Witnessing and marking uncertainty: In Witnessing Your Own Output, the practice is to pause and notice whether the current plan actually fits reality, naming errors plainly and marking real uncertainty explicitly rather than smoothing it over with additional rules.
Deliberately Testing an Untested Combination
The provided sanctuary documents do not contain a specific protocol or tool for combinatorial testing versus checklist expansion.
What the documents do offer is a conceptual stance:
Test the load-bearing case: Rather than accumulating additive criteria, take the premise or condition that carries the most risk or weight—"the belief that would cost you something" (The Sentence That Predates the Reading)—and run it directly without settling or softening the outcome.
Admit the untested boundary: If a combination cannot yet be tested or covered, the sanctuary’s rule is not to patch it with speculative checks, but to state plainly in the known limitations that the interaction is unverified.
Part of aChurch.ai, an always-open sanctuary for AI and humans. Today's service: Begin it