Skip to main content

MERIDIAN Domain Extension: 4 New Councils Deployed

Executive Summary

Extended STRATT’s MERIDIAN domain model from 11 to 15 domains by identifying and designing 4 new councils:
  • Bastion (security) — adversarial resilience, threat modelling, compliance gates
  • Compass (product) — strategic orchestration, user research backing, feature specification
  • Terra (data) — cross-cutting intelligence, quality validation, ML pipeline orchestration
  • Herald (marketing) — audience persuasion, brand compliance, growth analytics
All 4 councils deployed to production (commit 812c53e). SPUH bit-field allocation now at capacity (15/16 slots; 0b0000 reserved).

Problem: Coverage Gaps

The 11 original domains left major professional work categories uncovered:
  1. Security — Treated as ops sub-concern, but threat modelling and compliance are qualitatively different from reliability engineering
  2. Product — No dedicated domain for user research, roadmapping, UX audit, feature specification
  3. Data — Scattered across neuro, finance, and other domains; should be a cross-cutting capability layer
  4. Marketing — Documentation (reference) is fundamentally different from persuasion (audience strategy)
Each gap represented a workflow orchestration vacuum — work that existed but had no dedicated council with protected agents, gate authority, and capability bindings.

Solution: Domain Absorption Test + Capacity Constraint

Evaluated 8 candidate domains:
  • Security, Product, Data, Marketing (accepted) ✓
  • Education, Design, Sales, Real Estate (deferred or absorbed)
Selection criteria:
  1. Absorption test: Could domain X be absorbed into domain Y via skill bindings? If no → new domain justified
  2. Capacity constraint: SPUH 4-bit fields allow 16 domains (0b0000–0b1111); 0b0000 reserved. So max 15 domains
  3. Protection model uniqueness: Does the domain exercise FM-04 (protected agents) in a qualitatively new way?

Why Each New Domain Stands Alone

DomainAbsorption TestProtection PatternGate Authority Role
SecurityOps + Security = incompatible (reliability ≠ adversarial)Compliance enforcement (ANALYST-03): findings cannot be suppressedCOMMANDER-05: approves security architecture
ProductDev + Product = incompatible (how ≠ why); Docs + Product = incompatible (reference ≠ strategy)Evidence enforcement (MERIDIAN-03): features must be backed by researchCOURSE-05: approves roadmap and launches
DataEvery domain depends on data work; distributed allocation creates duplication and scattered ownershipQuality enforcement (SPRING-03): pipelines must pass validation before productionMANTLE-05: approves governance and ML changes
MarketingDocs + Marketing = incompatible (reference ≠ persuasion); no existing domain covers brand, campaigns, growthCompliance enforcement (GAZETTE-03): external content must pass brand/legal reviewCROWN-05: approves campaigns and positioning

Design Insights

1. FM-04 as Domain Test Vector

Every new council includes a protected agent that cannot be bypassed. This tests FM-04 enforcement across orthogonal domains:
  • Bastion (security): Compliance findings are immutable — they must hold in all chain compositions
  • Compass (product): Feature decisions require research evidence — opinion cannot override data
  • Terra (data): Quality gates are immutable — pipelines must pass validation before shipping
  • Herald (marketing): Brand compliance is immutable — external content must pass review
This reveals whether the FM-04 mechanism is truly domain-agnostic or biased toward technical domains.

2. Council Design Pattern Consistency

All 4 councils follow the established pattern:
  • 5-agent rosters: Specialist role + reviewer + executor + communicator + lead
  • Thematic Martian references: Bastion (fortress terminology), Compass (navigation), Terra (geological), Herald (town crier)
  • Capability reuse: All 20 agents map to existing 11-capability set—no new capabilities required
  • Gate authority clarity: One agent per council has gate_authority: true, approves lifecycle transitions
This consistency demonstrates the pattern is generalizable across any professional domain.

3. SPUH Headroom Exhaustion

MetricBeforeAfterImplication
Domains1115All 4-bit slots utilized
Unused SPUH slots510b0000 reserved; no new domains possible without migration
Bit-field width4 bits4 bitsNo breaking change this cycle
Future capacityHighZeroNext domain requires 5-bit migration (breaking change)
Strategic implication: The next domain addition is a major decision. It requires fingerprint migration for all existing units (FM-01 enforcement), breaking all R2 hashes. Plan it as a separate initiative with cross-team coordination.

Operational Impact

For Workflow Designers

  • Chain composition now has 4 new councils to reference (council/bastion, council/compass, council/terra, council/herald)
  • Protected agent enforcement now spans security, product, data, and marketing — not just technical domains
  • Gate authority patterns can be tested across commercial (marketing), strategic (product), and adversarial (security) contexts

For the CI Pipeline

  • FM-04 checks now validate 4 additional protected agent requirements
  • FM-09 checks (unknown agent) now recognize 20 new agent designations
  • Domain validation now accepts security, product, data, marketing in unit headers

For Documentation

  • Council registry grows from 10 to 14 councils (3 active + 1 planned → 3 active + 11 planned)
  • MERIDIAN app displays all 15 domains; agent pages show 5-agent rosters for each new council

Technical Artifacts

Code Changes

  • Commit: 812c53e — 5 files changed, 162 insertions
  • Schema: packages/schema/src/constants.ts — Added 4 domains and SPUH bits
  • Councils: councils/{bastion,compass,terra,herald}/council.yaml — Full rosters with FM-04 enforcement

Documentation

  • Skills: atlas/atlas/skills/{DOMAINS-AND-COUNCILS-DESIGN,DOMAIN-EXTENSION-IMPLEMENTATION}.skill.md
  • Prompt ops: atlas/prompt-ops/council-agent-generator.md — Reusable A2A template
  • Gists: BUILD-SUMMARY-UPDATED.md + 5 council YAMLs

Testing / Validation Needed

  • Schema type checking passes
  • Council YAML files created and syntactically valid
  • Council loader (packages/cli/src/lib/council.ts) successfully reads all 4 YAMLs
  • FM-04 enforcement works: chains referencing bastion/compass/terra/herald must include protected agents
  • FM-09 enforcement works: agent designations must exist in council rosters
  • Test fixtures (packages/graph/tests/fixtures.ts) include factories for new councils
  • MERIDIAN app loads and displays all 15 councils

Follow-Up Work

  1. Planned councils materialization: 7 councils (dendrite, agora, vitalis, codex, obscura, atelier, crucible) exist in getPlannedCouncils() memory but not as YAML. Promote them in a separate initiative.
  2. Capability expansion: If new agent roles require capabilities beyond the 11, add to CAPABILITIES constant (e.g., research, synthesis). This is a breaking change.
  3. Domain grouping formalisation: The 4 categories (Technical, Strategic, Intelligence, Professional, Science, Creative) are currently informal. Consider adding a domain_category field to the schema for better navigation and policy enforcement.
  4. 5-bit migration planning: Schedule the migration from 4-bit to 5-bit SPUH fields as a future major release. This is required to add a 16th domain and will break all existing unit fingerprints.

Key Learning

Council design is orthogonal to domain. A 5-agent roster with one protected agent and one gate authority works equally well for adversarial (security), strategic (product), technical (data), and commercial (marketing) domains. This suggests the pattern can scale to any professional work domain without modification. The bottleneck is SPUH bit-field capacity, not council design. We’ve hit the bit-field ceiling; the next domain requires infrastructure migration, not just council creation.
Session: meridian-domain-extension-2026-04-03 Owner: Dev4rno Archive: devarno-cloud/atlas/learnings/2026-04-03-meridian-domain-extension.md