◌ Demo data
MR
  1. Brand Graph
  2. → Intent
  3. → Ingest
  4. → Reader Lens
  5. → Evidence Check
  6. → Precedents
  7. → Fix & Route
  8. → Approval
  9. → Proof Card

Guideline stress test

Simulated reviewers apply each rule to probe cases. Low agreement means the rule wording is ambiguous.

◌ Demo data

Clarity score per rule

◌ Demo data

Rules below 75 are flagged as ambiguous.

Reviewer agreement (Fleiss' kappa)

◌ Demo data
RuleKappaAgreement
R-0120.41▲ Moderate
R-0190.79✓ Substantial
T-030.91✓ Strong
T-050.33▲ Poor
R-0240.72✓ Substantial
R-0310.55▲ Moderate

Ambiguous rules

R-012 · No absolute security claims

Proposal only
Probe: "Guaranteed 24/7 monitoring"
  • ✕ Reviewer A: Violation
  • ✓ Reviewer B: Compliant
  • ✕ Reviewer C: Violation

Reviewers disagree on the same text.

Before

Never say unbreakable, 100% or guaranteed about detection.

Suggested rewrite

Never use absolute words (unbreakable, 100%, guaranteed) about detection or protection outcomes. Service-level commitments backed by an SLA may use them.

Suggestion only. A brand owner must review and approve any change to the guideline.

T-05 · Prefer threat, avoid hack

Proposal only
Probe: "Hackers hate this one trick"
  • ✕ Reviewer A: Violation
  • ✕ Reviewer B: Violation
  • ✓ Reviewer C: Compliant

Reviewers disagree on the same text.

Before

Neutral, professional vocabulary.

Suggested rewrite

Use threat, attacker or adversary. Avoid hack, hacker and slang in customer-facing copy; quoting a third party is allowed with attribution.

Suggestion only. A brand owner must review and approve any change to the guideline.