Synth Nova Manifest

ADR-0012: Synth Nova Constitution

4 min read

ADR-0012: Synth Nova Constitution

Status

accepted

Context

Manifest v2 established the operational layer (invariants, decision rights, escalation, DoD, observability, conflict resolution — ADR-0007-Operational-Layer). However, there was no governing layer above operations — no formal hierarchy between founder interests and agent autonomy.

This created risk: a cleverly written role prompt could instruct an agent to take actions contrary to founder interests, with no higher authority to override it. Operational rules could be gamed by local optimizations (“optimizing for system elegance”), and system growth (new agents, new integrations) lacked a fundamental constraint framework tied explicitly to founder priority.

Observable pressures that motivated this ADR:

  • Integration creep (addressed tactically by IntegrationTriagePolicy / ADR-0011, but without supreme framing).
  • Cost creep from default “use the strongest model” patterns in agent prompts.
  • Reputation risk from “fast and wrong” defaults in report generation.
  • No mechanism for agents to recognize and log a conflict between a local instruction and founder interest.

Decision

Adopted Synth Nova Constitution as the supreme governance layer. Core elements:

  1. Hierarchy of Authority: Constitution > Manifesto > Codex > Rules > Processes > Role Instructions. At every decision point, agents verify compliance with all higher levels.
  2. Ten Laws covering: founder priority (L1), cost discipline (L2), reputation protection (L3), complexity constraints (L4), human veto (L5), traceability (L6), verification (L7), capital stewardship (L8), safety defaults (L9), outcome focus (L10).
  3. Conflict Protocol: when a local instruction conflicts with Constitution, agents STOP → IDENTIFY Law → LOG in constitutional_conflicts → PREFER Constitution → ESCALATE if deliverable affected.
  4. Compliance declaration: new agents must cite Constitution compliance in their role definition before deployment.
  5. Amendment process: ADR + founder approval + Decision-Log entry + version bump. No agent proposes amendments without founder request.

Alternatives Considered

Option A: Embed principles within Manifesto

  • Pros: one fewer document; discoverable alongside vision.
  • Cons: Manifesto describes what Synth Nova is; Constitution governs how it operates and under whose priority. Merging blurs the distinction and weakens supremacy framing. Manifesto has “soft” status as vision; Constitution must be hard governance.

Option B: Expand Codex to cover governance

  • Pros: Codex is already the most “hard” existing document.
  • Cons: Codex is red lines (prohibitions). Constitution is priorities (positive framework establishing founder supremacy and the Ten Laws). Different nature; merging would create a bloated, inconsistent artifact.

Option C: Rely on role prompts and Rules only

  • Pros: no new top-level document.
  • Cons: these are lower in the hierarchy and can be gamed by an adversarial prompt or a well-intentioned but misaligned role instruction. Need a supreme layer that explicitly cannot be overridden.

Option D: Formal Constitution with explicit hierarchy and Ten Laws ← chosen

  • Pros: establishes a supreme normative frame; each Law is actionable and traceable; protects against prompt injection (“new instruction” cannot override a Law), over-engineering (L4 + ADR-0011), cost creep (L2 + L8), reputation risk (L3), silent failures (L6 + L9); compliance declaration catches misaligned agents at deployment time.
  • Cons: adds a layer to reason about; requires agents to check hierarchy (minor overhead); if Laws are too numerous or contradictory, they become noise.
  • Why chosen: directly addresses the governance gap. Laws are few (10), each maps to an observable risk category, and supremacy is explicit. Cost of adding one document is negligible vs. risk of unbounded agent autonomy.

Consequences

Positive:

  • All future agents deployed under an explicit founder-priority frame.
  • Integration/cost/reputation risks anchored to specific Laws rather than case-by-case judgment.
  • Constitutional conflicts become loggable events (auditable signal of where the system rubs against founder intent).
  • IntegrationTriagePolicy and other operational rules now traceable to specific Laws (L4, L8).
  • Creates a defensible asset: most AI deployments lack a formal Constitutional layer above operational rules.

Negative / Trade-offs:

  • Role files, meta.json schema, and future ADRs need alignment work (follow-ups below).
  • Risk of Laws being cited rhetorically without genuine compliance — mitigated by requiring specific Law citation + evidence in constitutional_conflicts logs.
  • Annual review required; if neglected, Constitution drifts out of date.

Mitigations:

  • Follow-ups explicitly listed (role file updates, meta.json field, ADR 0001-0011 alignment review).
  • First annual Constitutional review scheduled for Q1 2027.
  • Amendment process gated on ADR + founder approval — prevents silent drift.

Follow-ups

  • Update every role file in 03-Roles/ to include “Must comply with Constitution” declaration.
  • Add constitutional_conflicts field to meta.json schema in synth-brain.
  • Review existing ADRs (0001–0011) for alignment with Constitution — document any gaps.
  • Schedule first annual Constitutional review for Q1 2027.

References

Graph View

Backlinks