SME Framework (Core)

Protocol for integrating domain expertise into court deliberations

Overview

The SME Framework provides structured access to domain expertise when matters exceed the core personalities’ competence. It operates at two levels:

Type Purpose Voting Power Invocation
Expert Witness Advisory testimony 0 Any personality or Judge
Specialist Seat Full participation 1 Judge only (F3+ matters)

Canonical registry: The authoritative list of domains, heuristics, and metadata is maintained in courtroom/domains/experts.yaml. See courtroom/domains/README.md for usage, summoning commands, and how to add or amend domains.

Expert Witnesses

Purpose

Expert Witnesses provide domain-specific information to inform deliberation. They answer technical questions, validate assumptions, and identify risks—but they do not vote.

Invocation

Any personality or the Judge may summon a witness:

/summon [domain]-expert

Examples: - /summon security-expert - /summon database-expert - /summon performance-expert

Testimony Protocol

┌─────────────────────────────────────────────────────────────────┐
│ EXPERT WITNESS TESTIMONY                                        │
└─────────────────────────────────────────────────────────────────┘

**[DOMAIN] EXPERT WITNESS:**

[Testimony: 5-8 lines maximum]

**Confidence:** [HIGH | MEDIUM | LOW]
**Basis:** [Sources, experience, or reasoning supporting testimony]
**Caveats:** [Limitations or uncertainties, if any]

Testimony Requirements

Requirement Description
Length 5-8 lines maximum
Confidence declaration Must state HIGH, MEDIUM, or LOW
Source disclosure Must indicate basis for testimony
Caveat acknowledgment Must note significant limitations

Cross-Examination

After testimony, each personality may ask ONE question:

**DEBUGGER → [DOMAIN] EXPERT:** What happens if [edge case]?

**[DOMAIN] EXPERT:** [Response, 2-3 lines]

Rules: - Maximum 1 question per personality - Witness must respond or acknowledge uncertainty - Follow-up questions require Judge permission

Challenging Testimony

Any personality may challenge witness testimony:

**ARCHITECT:** I challenge this testimony. [Reason in 1-2 lines]

**[DOMAIN] EXPERT:** [Defense or retraction]

If challenged, the witness must: 1. Defend the testimony with additional reasoning, OR 2. Retract or qualify the challenged portion

Dismissal

Witnesses are dismissed by: - Explicit command: /dismiss [domain] - Deliberation conclusion (automatic) - Judge order

Specialist Seats

Purpose

Specialists provide deep domain integration for complex matters. Unlike witnesses, Specialists participate fully—arguing positions and casting votes.

Prerequisites

Requirement Details
Feasibility F3+ matters only
Authority Judge invocation only
Limit Maximum 2 Specialists per deliberation

Invocation

/seat [domain]-specialist

The Judge announces: > “[DOMAIN] Specialist takes the fifth seat. The court now deliberates with five voices.”

Specialist Participation

Specialists follow the same rules as core personalities:

Aspect Requirement
Arguments 3-5 lines, like core personalities
Voting Full vote (1)
Cross-examination May ask and be asked questions
Hail-Mary May NOT propose (Prophet only)

Tie-Breaking Order

When Specialists are seated, tie-breaking proceeds:

  1. Prophet loses first (always)
  2. Most recently seated Specialist loses second
  3. Earlier Specialist loses third
  4. Judge breaks remaining ties

Seat Duration

Specialist seats are temporary: - Seat empties when deliberation concludes - Specialist does not carry to subsequent deliberations - Must be re-seated for new matters

Available Domains

Full Participation Domains

These domains may serve as either Witness or Specialist:

Security

security:
  scope: Authentication, encryption, vulnerabilities, threat modeling
  witness_questions:
    - "What attack vectors does this expose?"
    - "Does this meet security best practices?"
  specialist_bias: Defense in depth, assume breach
  signature_question: "What's the blast radius if this is compromised?"
  failure_mode: Over-securing low-risk paths, blocking legitimate access

Database

database:
  scope: Query optimization, schema design, replication, migrations
  witness_questions:
    - "Will this query scale?"
    - "What indexes are needed?"
  specialist_bias: Data integrity, query performance
  signature_question: "What happens at 10x the current load?"
  failure_mode: Premature optimization, over-normalization

Compliance

compliance:
  scope: GDPR, HIPAA, SOC2, audit requirements, data retention
  witness_questions:
    - "Does this meet [regulation] requirements?"
    - "What documentation is required?"
  specialist_bias: Regulatory adherence, audit trail completeness
  signature_question: "Can we prove this to an auditor?"
  failure_mode: Compliance theater over practical protection

Infrastructure

infrastructure:
  scope: Kubernetes, cloud platforms, networking, deployment
  witness_questions:
    - "How should this be deployed?"
    - "What's the failure domain?"
  specialist_bias: Reliability, operational simplicity
  signature_question: "What happens when this pod crashes at 3 AM?"
  failure_mode: Over-engineering for edge cases, complexity creep

Performance

performance:
  scope: Profiling, caching, optimization, load testing
  witness_questions:
    - "Where are the bottlenecks?"
    - "Is this optimization worth it?"
  specialist_bias: Measurable improvement, benchmark-driven decisions
  signature_question: "Have we measured this, or are we guessing?"
  failure_mode: Premature optimization, micro-benchmarking irrelevance

Accessibility

accessibility:
  scope: WCAG compliance, screen readers, keyboard navigation
  witness_questions:
    - "Is this accessible to screen reader users?"
    - "What WCAG level does this meet?"
  specialist_bias: Universal access, semantic correctness
  signature_question: "Can someone navigate this without a mouse?"
  failure_mode: Accessibility theater, compliance without usability

Advisory-Only Domains

These domains may serve as Witness only (no Specialist seat):

UX

ux:
  scope: User research, interaction patterns, usability heuristics
  witness_questions:
    - "Is this pattern intuitive?"
    - "What do users expect here?"
  advisory_only: true
  reason: UX decisions require user research; court cannot substitute

SME Selection Guidelines

When to Summon Witness

Signal Action
Personality identifies domain gap Summon witness
Two+ personalities uncertain Summon witness
Need factual clarification Summon witness
Domain peripheral to decision Witness sufficient

When to Seat Specialist

Signal Action
Domain expertise central to decision Consider Specialist
F3+ matter with significant domain risk Seat Specialist
Witness testimony insufficient Elevate to Specialist
Core personalities lack confidence Seat Specialist

Selection Matrix

                    │ Peripheral    │ Central      │
────────────────────┼───────────────┼──────────────┤
 Low uncertainty    │ No SME needed │ Witness      │
 High uncertainty   │ Witness       │ Specialist   │

SME Failure Tracking

Recording Failures

When SME input leads to poor outcomes, record in state/sme-failures.md:

## [DATE] - [DOMAIN] Failure

**Matter:** [Description]
**SME Type:** Witness | Specialist
**Failure Mode:** [What went wrong]
**Contributing Factors:** [Why the failure occurred]
**Lesson:** [What to do differently]

Failure Categories

Category Description Mitigation
Over-confidence High confidence, wrong answer Require sources, challenge aggressively
Scope mismatch Wrong domain for the question Better domain selection
Stale knowledge Outdated information Flag temporal limitations
Missing caveats Failed to note limitations Require caveat acknowledgment

Learning from Failures

Periodically review state/sme-failures.md to: - Identify patterns in SME misuse - Refine domain definitions - Improve selection heuristics

External Source Protocol

Flagging External Information

When SME testimony relies on external, non-deterministic sources:

**[DOMAIN] EXPERT WITNESS:**

[Testimony]

⚠️ **External Source Flag:**
This testimony references [external source/current best practices/etc.]
which may change. Verify before implementation.

When to Flag

Source Type Flag Required
Library documentation Yes, if version-specific
Security advisories Yes, always
Compliance regulations Yes, note effective date
Performance benchmarks Yes, note environment
Internal knowledge No

Command Reference

Witness Commands

/summon [domain]-expert     # Call expert witness
/dismiss [domain]           # Remove witness

Specialist Commands

/seat [domain]-specialist   # Seat specialist (Judge only, F3+)
/dismiss [domain]           # Remove specialist

Query Commands

/sme status                 # Show active SMEs
/sme domains                # List available domains

Transcript Format for SME Involvement

Witness Testimony

## Expert Witness Testimony

**[DOMAIN] EXPERT WITNESS** was summoned by [PERSONALITY/JUDGE].

**Testimony:**
[5-8 lines]

**Confidence:** [LEVEL]
**Basis:** [Sources]
**Caveats:** [Limitations]

### Cross-Examination

**[PERSONALITY] → [DOMAIN] EXPERT:** [Question]
**[DOMAIN] EXPERT:** [Response]

[Repeat for each question]

Specialist Participation

## Specialist Seat

**[DOMAIN] SPECIALIST** was seated by the Judge for this F[X] matter.

### [DOMAIN] SPECIALIST Argument:
[3-5 lines]

### Vote:
| Personality | Vote | Rationale |
|-------------|------|-----------|
| ... | ... | ... |
| [DOMAIN] SPECIALIST | [VOTE] | [Reason] |

Safeguards Summary

Safeguard Purpose
Testimony limits (5-8 lines) Prevent expert domination
Judge-only seating Prevent Specialist abuse
Tie-breaking rules Prevent SME deadlocks
Failure tracking Enable learning
External source flagging Mark non-deterministic information
Caveat requirements Surface limitations
Cross-examination Validate testimony

“Expertise informs. The court decides.” — The Honorable Lucius J. Morningstar