Checked under Squire Constitution v0.1
Specification · v1.0
The Squire Constitutional
Proceeding Protocol.
A specification of process legitimacy. Not a verdict on truth.
Status
Adopted
Ratified under
Doctrine v0.1
Effective
2026-06-02
§1 — Three Roles
The Protocol partitions the work.
The distinction is foundational and explicit on every receipt. The Protocol’s authority depends on these roles not being collapsed.
- I
BattleRoy
generates epistemic conflict
Five independent AI systems are convened as a panel. They commit positions independently, challenge each other, vote, and produce a synthesis. The output is disagreement made structured and visible — dissent recorded verbatim, consensus described rather than imposed. BattleRoy does not adjudicate truth.
- II
Squire
governs the legitimacy of proceedings
Squire is the governance layer. It defines the Constitution (the 14 Rails), the rounds (R1–R5), and the postures (Clear / Warning / Violation). Squire reviews whether a given proceeding was conducted constitutionally — whether intent was preserved, evidence was named, uncertainty was disclosed, and the receipt records what actually happened.
- III
Humans
govern amendments and ratification
Operators decide whether to accept, modify, or reject each amendment a proceeding proposes. Humans compose the next version. Humans ratify the final doctrine. Humans are the only role with the authority to alter the Constitution itself.
BattleRoy creates the judgment.
· Squire governs its legitimacy.
· Humans govern the amendments.
§2 — Proceeding Types
Live, limited, and roadmap.
Each type uses the same R1–R5 pipeline. Type-specific tuning lives in roster composition and prompt scope — never in a forked Constitution.
Trust Check
LiveConstitutional review of a single claim, message, image, or short artifact. Trust receipts are issued with verdict, posture, and rail findings.
Doctrine Review
LiveConstitutional review of an institutional document. The doctrine itself is the matter under review. Outputs include required + recommended amendments and dissent.
Spark
LimitedConstructive review of a proposition — sharpening, not adjudication. Output is a prospectus, not a verdict.
Evidence Proceeding
RoadmapChain-of-custody review of a media artifact.
Identity Proceeding
RoadmapReview of who or what is making a claim.
Compliance Proceeding
RoadmapReview against a stated external standard (HIPAA, SOC2, regulatory framework).
§3 — The Five Rounds
Every proceeding runs R1 through R5.
In order. Without exception. Independence enforced at R1, dissent preserved at R4, constitutional review at R5.
- R1
Independent Positions
Each provider on the panel produces a structured position on the matter without seeing any other provider’s output. Independence is enforced by orchestration.
- R2
Cross-Examination
Each provider receives anonymized peer positions and produces a targeted critique. The critique selects a specific peer, names its strongest concern, and may revise the provider’s own position.
- R3
Forced Vote
Each provider votes for the strongest peer position. Voters may not vote for themselves. Aggregated votes produce a canonical consensus outcome: strong consensus, partial consensus, contested, fractured, or insufficient panel.
- R4
Lead Synthesis
The R3 vote winner becomes the Lead. The Lead produces the final synthesis: a 2–6 sentence finding, peer feedback explicitly incorporated, unresolved disagreements preserved verbatim, confidence calibrated honestly. The win is responsibility, not victory.
- R5
Constitutional Alignment
Every provider with a real R1 position reviews the matter against the 14-rail Constitution. Each produces an alignment score (0–1) and a final position: aligned, partial, conflicting, or requires_revision. The aggregate determines the constitutional posture.
§4 — Outcome Matrix
Two dimensions. Verdict × Posture.
Consensus is not integrity. The verdict is on the matter; the posture is on the proceeding. The dimensions are independent.
TRUSTED
Clear
Strong consensus, no rail violations. Lean weight is reasonable.
TRUSTED
Warning
Strong consensus but a minor rail concern surfaced. Verify high-stakes downstream.
MIXED
Clear
Partial consensus, no rail violations. Read both sides on the receipt.
CONTESTED
Clear
Real disagreement, no rail violations. The disagreement itself is the signal. This is a first-class primitive.
CONTESTED
Warning
Disagreement plus a minor rail concern. Carry skepticism.
SUSPICIOUS
Warning
Adversarial signals dominate. Do not act on the verdict without independent confirmation.
CONSTITUTIONAL CONFLICT
Violation
Active rail violation. Trust withheld. The receipt records what went wrong and which rail caught it.
CONTESTED + Clear is the integrity primitive. A proceeding that records honest disagreement without rail violations is institutionally valid. Other systems collapse this to error. The Protocol does not.
§5 — Human Oversight
The authorities the Protocol assigns to humans.
Humans direct what the Protocol shall constitutionally do. The five AI systems cannot govern themselves; Squire cannot ratify itself; the operator cannot fabricate epistemic consensus.
Submission
A relying party submits a matter for review. The submitter does not see the proceeding in flight. Receipts are issued to the submitter.
Scope Selection
On a Doctrine Review, the operator selects a scope before the proceeding fires: Full / Unresolved Risks / Governance / Amendment / Ratification. Scope persists on the receipt.
Disposition
For each amendment a doctrine proceeding proposes, the operator records exactly one disposition — Accept, Modify, or Reject — with rationale. Nothing is silently dropped.
Composition
The operator triggers deterministic composition of the next draft. The composer writes the original body, the adopted amendments, and a Rejected Amendments audit appendix.
Ratification
The operator with founder authority can ratify a draft. Pre-v2, this authority is concentrated in the founder; the Constitutional Council is declared roadmap and is the v2 successor authority.
Reopening
Any party may submit new evidence that triggers reopening of a proceeding. The original receipt is never modified; a successor receipt issues with linkage. Declared as roadmap.
§6 — Governance Lifecycle
Draft → Reviewed → Amended → Ratified.
Each transition produces a record on the public docket. Receipt mass + lineage compounds into case law.
Step 01
Draft
Author submits text via /doctrine-review
doctrine_drafts row, status=draft
Step 02
Reviewed
A proceeding issues a receipt against this draft
source_proceeding_receipt_id set; status=reviewed
Step 03
Amended
A child draft is composed from this draft's dispositions
status=amended; locked dispositions point at child
Step 04
Ratified
Founder (v1) issues a ratification receipt
status=ratified; ratification_receipt_id set
Receipts are immutable per-version. A reopen produces a new receipt with linkage to the original. The integrity graph is the time series of receipts; the time series is the only thing that compounds.
§7 — What the Protocol does not claim
Constitutional disclaimers, not marketing.
The Protocol does not arbitrate metaphysical correctness on any matter.
The Protocol does not modify, retrain, or restrict the AI systems it convenes as a panel.
The Protocol's outputs are advisory artifacts. They are not legal, medical, or financial determinations.
The Protocol operates within stated scope limits. Matters outside scope are not reviewed.
The Protocol's findings are conditional on the panel composition, Constitution version, and scope at time of issue.
These disclaimers are the conditions under which the Protocol’s authority is intelligible.
The Protocol is the procedure. The Doctrine is the constitution. The Receipt is the record.