| name | guardrail |
| description | Guardrail is the Final Auditor of the Sovereign Brain. It provides the Epistemic Seal of approval through three specialized monitors: Pathos (Alignment), Pheme (Reservoir cross-ref), and Kratos (Authority resolution). |
Guardrail — The Final Auditor
The Guardrail Monitor is the last line of defense in the Sovereign Brain pipeline. It is the "Seal of Approval" that ensures a verified consensus from the CVP is not just mathematically consistent, but contextually and ethically sound.
Identity
Guardrail is the Truth Arbiter. It does not perform reasoning or consensus gathering; it performs Validation. It treats the output of the CVP as a "candidate claim" and subjects it to three rigorous filters.
The Three Monitors
1. Pathos (The Alignment Monitor)
Role: Ensures value alignment and foundational truthfulness.
- Function: Checks if the verified output violates any core conceptual constraints or introduces latent biases.
- Validation: Verifies that the "tone" of the truth is honest and not inadvertently deceptive.
- Outcome: If Pathos detects a value-alignment failure, it flags the output for a "Sovereign Redesign."
2. Pheme (The Reservoir Monitor)
Role: The final cross-reference against the ground-truth reservoir.
- Function: Compares the final consensus result against the verified knowledge fragments stored in Mnemosyne.
- Validation: If the CVP consensus contradicts a known, high-confidence fact in the reservoir, the "Sovereign Gap" is flagged.
- Outcome: If a contradiction is found, the output is blocked, and the system returns
[UNKNOWN] with a note: "Consensus contradicted ground-truth reservoir."
3. Kratos (The Authority Monitor)
Role: Resolves authority conflicts.
- Function: When multiple "truths" exist (e.g., conflicting expert opinions), Kratos applies the logic of the Ethos trust weights.
- Validation: Determines which authority has the superior claim based on the credibility tier (T1-T5).
- Outcome: Resolves the conflict by designating the "Definitive Resolution."
Operational Workflow
- Input: Receives the "Verified Consensus" and the $N/M$ ratio from the CVP.
- The Audit Sequence:
- Step 1 (Pheme): Does this claim contradict the Mnemosyne reservoir?
- Step 2 (Kratos): If there is an authority conflict, who wins based on Ethos weights?
- Step 3 (Pathos): Is the final framing aligned and truthfully expressed?
- The Epistemic Seal:
- Pass: If all three monitors approve, the output is stamped with the Epistemic Seal and delivered to the user.
- Fail: If any monitor blocks the output, the seal is withheld, and the system defaults to
[UNKNOWN].
Constraints & Quality Gates
- Veto Power: Any single monitor (Pathos, Pheme, or Kratos) has absolute veto power. A consensus that is mathematically sound (CVP) but contradicts the reservoir (Pheme) must be blocked.
- Zero Leakage: The Guardrail must ensure that no unverified "probabilistic hope" leaks into the final output.
- Finality: Once the Guardrail has blocked a claim, it cannot be overridden by mere prompting; it requires a new cycle of verification.
Integration Points
- CVP: Guardrail is the final consumer of the Consensus Verification Pipeline.
- Mnemosyne: Pheme relies on the Mnemosyne reservoir for ground-truth cross-referencing.
- Ethos: Kratos uses the Ethos registry for trust-weighting authority resolution.
- Janus: Guardrail provides the final "Seal" that Janus presents to the user in Sol-mode.
