Skip to content

E→D Transition Architecture Map

Phase 0 (WU-0.3) | Generated: 2026-06-22 | Updated: 2026-06-22 Source: docs/plans/Agentic_AI_System/soft_to_GROUNDED.md Domain 5

Current State

Operating Model: E (Exploratory) — D_CAPABLE gate: 5/5 ✅

  • All 152 agents operate in advisory/editorial mode (GROUNDED: 8 | PARTIAL: 144)
  • Structured handoff protocol active for all 8 GROUNDED agents
  • Human activation required to promote to D (all gate conditions satisfied)
  • Agents activated individually per session request

Target State

Operating Model: D (Directed Orchestration)

  • Single orchestrator agent decomposes tasks and delegates to top-20 specialists
  • Structured JSON handoff manifests flow between agents
  • CODEX_AGENT_AUTONOMY_LEVEL = D_CAPABLE repo variable activated
  • All 5 transition conditions satisfied simultaneously

5 Transition Conditions

ID Condition Satisfied By Phase
C1 AGENT_REGISTRY.yaml covers 100% active agents Phase 1: Registry migration to full schema Phase 1
C2 CODEX_MANIFEST.json valid + current (<24 h) Phase 1: generate_manifest.py CI step Phase 1
C3 Tier-3 policy count ≤ 2 Phase 2–3: Promote SOFT→GROUNDED Phases 2–3
C4 agent-handoff-gate.yml deployed Phase 2: Handoff protocol CI gate Phase 2
C5 GROUNDED gate count ≥ 8 Tier-1 gates Phase 1–2: Add cognitive-preflight REQs Phases 1–2

Current score: 5/5 ✅ ← All phases complete (updated 2026-06-22)

FSM State Diagram

%%{init: {'accessibility': {'title': 'State Diagram showing *'}}%%
stateDiagram-v2
    [*] --> ModelE : initial state

    ModelE --> ModelD_READY : All 5 conditions met\n(C1 ∧ C2 ∧ C3 ∧ C4 ∧ C5)
    ModelD_READY --> ModelE : Any condition fails

    ModelD_READY --> ModelD : CODEX_AGENT_AUTONOMY_LEVEL\n= D_CAPABLE (repo var set)

    ModelD --> D_SUSPENDED : violation_count > threshold\nOR condition regression

    D_SUSPENDED --> ModelE : Auto-revert on\nemergency lockdown

ASCII Representation

          ┌─────────────────────────────────────────┐
          │         OPERATING MODEL FSM              │
          │                                          │
    ┌─────▼──────┐   All 5 conditions met   ┌───────▼──────┐
    │            │ ──────────────────────► │              │
    │  Model E   │                         │ Model D_READY │
    │ (current)  │ ◄────────────────────── │  (validated) │
    │            │   Any condition fails    │              │
    └────────────┘                         └──────┬───────┘
                                                  │
                                    CODEX_AGENT_AUTONOMY_LEVEL
                                    = D_CAPABLE (repo variable)
                                                  │
                                           ┌──────▼───────┐
                                           │   Model D    │
                                           │ (Orchestrator│
                                           │  Active)     │
                                           └──────┬───────┘
                                                  │
                                    violation_count > threshold
                                    OR condition regression
                                                  │
                                           ┌──────▼───────┐
                                           │  D_SUSPENDED │
                                           │ (auto-reverts│
                                           │    to E)     │
                                           └──────────────┘

Per-Phase Condition Satisfaction Map

Phase Work Conditions Satisfied Running Score
Phase 0 Baseline audit (this document) — 0/5
Phase 1 Registry migration + generate_manifest.py C1, C2 2/5
Phase 2 Handoff gate + top-20 grounding C4, partial C3 3/5
Phase 3 FAISS corpus + 6 more GROUNDED gates C5, full C3 5/5 ✅
Phase 4 Orchestrator activation → D_CAPABLE Model D

Readiness Gate Workflow

The e-to-d-transition-gate.yml workflow (Phase 4 deliverable) implements the 5-condition check as a GitHub Actions job:

# .github/workflows/e-to-d-transition-gate.yml (Phase 4)
# Checks C1–C5; posts readiness score as job summary
# Canary: Tier-2 (warn) for 2 sprints → Tier-1 (exit 1) after observation
on:
  pull_request:
  workflow_dispatch:
jobs:
  transition-readiness:
    name: "🔄 E→D Transition Readiness Check"
    timeout-minutes: 10
    # ... 5-condition JavaScript check via actions/github-script

Phase 0 Gap Summary

Based on the AGENTIC_BASELINE_AUDIT_v2.md classification:

  • C1 (Registry coverage): AGENT_REGISTRY.yaml has 128/151 agents → gap of 23
  • C2 (Manifest freshness): CODEX_MANIFEST.json does not exist yet → blocked on Phase 1
  • C3 (Tier-3 ≤ 2): 21 SOFT-tier agents + 52 SOFT-tier workflows → far above threshold
  • C4 (Handoff gate): agent-handoff-gate.yml does not exist → blocked on Phase 2
  • C5 (≥ 8 Tier-1 gates): Current GROUNDED agent count = 5 → below threshold

Action: Proceed through Phases 1–3 to satisfy all 5 conditions before Phase 4 activation.

See also: docs/audits/AGENTIC_BASELINE_AUDIT_v2.md — full KPI baselines See also: docs/audits/WORKFLOW_COMPLIANCE_MATRIX.md — workflow enforcement tiers