AI Beyond FDLC Addendum Year plan · May 2026

Roadmap, Workflow RACI & System Architecture.

The 12-month plan following a successful pilot and first AI FDLC Review: rollout sequencing, deferred architecture activation, agent tier unlocks, and the system architecture the workflow operates within at one year post-pilot.

Directional. A starting point, not a commitment. Timelines are planning anchors; all activation decisions are evidence-gated per the AI FDLC Review cadence.
01 · Post-Pilot Roadmap

M0 through M12 — evidence-gated, not calendar-gated.

The 12-month post-pilot plan unfolds across four quarters of rollout, infrastructure deepening, agent unlocks, and constitution evolution — anchored by an immediate post-Review unlock and closed by the annual AI FDLC Review.

Governing constraint: All activation decisions are evidence-gated per the deferred architecture register. Quarterly labels are planning anchors, not commitments. The AI FDLC Review cadence (monthly/quarterly) is the decision forum for each unlock — not elapsed time. The roadmap assumes pilot produces clean signal on both tracks and the AI FDLC Review produces an advancement + rollout decision.
M0 — AI FDLC Review completion
Immediately Post-Review: Unlock & Calibrate
Governance outputs
  • Advancement decision issued — rollout authorized
  • Rollout calibration package finalized: team selection criteria, adoption sequence, onboarding conditions
  • CPO investment commitment formalized — token budgets, seat counts, rollout scope
  • AI FDLC Review becomes standing cadence (monthly during active rollout, quarterly at scale)
  • R2 activated: Learn → constitution amendment format; operational before first rollout close-out
Cost baseline locked
  • Pilot cost data converted to per-tier token budgets by phase
  • Track A vs. Track B cost structures established as separate baselines
  • Model routing policy authored and stored in prompt-templates/ repo
  • Phase-concentration anomaly thresholds set from pilot actuals
  • R8 naming decision: "Weave" activated or held; internal brand for rollout communication
Q1 · Months 1–3
Initial Rollout + Infrastructure Foundation
"First cohort running, system foundations laid"
Rollout
  • 3–4 domains onboarded at Spec-First; calibration package governs sequencing
  • Legacy-team sequencing model: two non-FDLC teams on structured migration path
  • SpecKit rollout recommendation executed if Tables pilot signal supports
  • Onboarding guide updated with pilot-derived companion file authoring norms
Agents — Unlock Tier 1
  • Spec-to-prototype reconciliation agent live (highest Tier 2 priority)
  • G3 review brief agent live (requires consistent feature folder structure)
  • Spec completeness / P7 sufficiency agent (A3) operational
  • Decision log integrity agent (A2) operational
  • AC quality agent deployed
  • Test scaffolding agent deployed per domain test framework
Textura MCP
  • Phase 2 build active: get_experience_type_patterns, get_recipe, validate_output
  • Pattern library v1 → v2 (8–12 patterns; includes pilot-surfaced patterns)
  • Admin experience type pack validated against pilot features
  • DS impact flag automation via GitHub Actions live
Infrastructure
  • Living Dependency Graph (LDG) schema authored by Architecture; teams populate entries
  • Companion file staleness monitor (B1) operational for rollout domains
  • Prompt template library established in shared FDLC repo
  • Prompt caching configured for constitution + companion files (≥90% cost saving)
Cost
  • Model routing policy enforced at template authoring; overrides logged
  • Token budget alerts operational (75% threshold by phase/tier)
  • Phase-level cost anomaly detection against pilot baselines
Exit:3+ domains Spec-First · Tier 1 agents live · LDG seeded
Q2 · Months 4–6
Scale + Maturity Advancement
"MCP Phase 2 lands, first teams advance"
Rollout
  • 6–8 domains at Spec-First; all eligible teams with companion files in rollout
  • Legacy team migration: 1+ legacy team at Pre-Spec-First entry conditions
  • DesignOps tooling decisions finalized from pilot signal
SDD Maturity
  • Pilot teams evaluated for Spec-Anchored advancement at AI FDLC Review
  • LLM-as-judge (Tier B–C rubric) active for advancing teams — runs on Opus 4
  • Amendment protocol operational: close-out → candidate → constitution version
  • First constitution amendments committed from close-out evidence (R2 live)
Agents — Unlock Tier 2
  • Dependency / contract screening agent (A1) live — LDG machine-readable
  • Contract change notification agent live (X-DOMAIN)
  • Cross-domain spec alignment agent live (X-DOMAIN)
  • Cycle time telemetry capture agent (B2) live — ADO state model mapped
  • JTBD completion rate agent live (Ph8 — Ph0 loop)
  • Behavioral signal agent live for rollout domains
  • Copy / content generation agent live (UX writing in constitution)
  • Enhancement candidate generation agent live
Textura MCP Phase 2 Complete
  • Phase 2 MCP live and tested across rollout teams
  • Context packs thinned — pattern/recipe content migrates to MCP queries
  • Pattern-aware generation pre-loader (C2) live — pattern library queryable
  • Shared component impact agent live (DS versioning system)
Infrastructure
  • R9 Portal MVP decision: CPO approves or defers; if approved, build begins
  • Analytics plan validation agent live (telemetry inventory machine-readable)
  • Pattern documentation agent live (documentation template standardized)
  • Pattern reuse impact agent live (pattern library queryable)
Exit:Pilot teams at Spec-Anchored · LDG with A1 screening · Phase 2 MCP live
Q3 · Months 7–9
System Scale + Intelligence Layer
"All eligible domains, flywheel compounding"
Rollout
  • All eligible Spec-First domains operating; legacy teams on accelerated path
  • Cohort 2 teams evaluated for Spec-Anchored advancement
  • Rollout velocity metric established: teams/quarter at target maturity
Agents — Unlock Tier 3
  • Cross-domain pattern alignment agent (X-DOMAIN; 3+ active domains)
  • Friction signal synthesis agent (session recording + support ticket taxonomy)
  • Feature ROI analysis agent (ARR data API + support ticket tagging)
  • Rollout configuration agent (feature flag system API)
  • Constraint inventory agent operational (companion file + DS coverage queryable)
  • Tier 3 autonomous agent evaluation: narrow scope, AI FDLC Review decision required
Textura MCP Phase 3 Groundwork
  • Phase 3 scope defined from Phase 2 signal: compose_ui, validate_runtime_ui
  • Horizon 2 gate (R10) assessed: evidence threshold met? CPO + CTO decision
  • Recipe library expanded: 8–12 recipes covering all major surface patterns
  • MCP version management automated (bump → smoke test → staleness check)
Infrastructure
  • R3 context governance: assess if git-based hierarchy is scaling; redesign only if bottleneck documented
  • R9 Portal: search, examples gallery, contribution flow live (if Q2 approved)
  • Full domain pack coverage: all active product domains have current context packs
  • Context pack authoring pipeline fully mature (AI draft → review → publish; <4hr)
SDD Maturity
  • Constitution amendment candidate agent (C1) live (3+ features with close-out records)
  • Dependency risk scoring agent operational (N≥10 features with rework event logs)
  • Spec-as-Source early candidates identified at AI FDLC Review
Exit:All eligible domains active · Tier 3 evaluation · Phase 3 MCP scoped
Q4 · Months 10–12
System Maturity + Annual Review
"Measurement reliable, loop fully closed"
SDD Maturity
  • First Spec-as-Source teams authorized (AI FDLC Review; highest-maturity teams)
  • Constitution version 2.0 reflects 12 months of production evidence
  • Amendment frequency normalized — amendment process is operational, not novel
Agent System Mature
  • VoC synthesis agent live (Gong/Chorus API + consent instrumentation)
  • Opportunity sizing agent live (ARR data API + comparable feature history)
  • Persona drift detection agent live (personas machine-readable)
  • Assumption audit agent live (assumption taxonomy per domain)
  • Comparable feature analysis agent live (competitive corpus structured)
  • Localization pre-screen agent live (l10n config per locale)
  • Interaction cost / cognitive load agent live (heuristics library machine-readable)
  • AC implementability agent live (codebase architecture in constitution)
  • PR description agent live (unit brief format + AC identifiers consistent)
Annual AI FDLC Review
  • 12-month outcomes vs. pilot targets: cycle time delta, rework rate, cost-per-outcome
  • CPO evidence package: cost-per-cycle-day recovered at rollout scale
  • Horizon 2 (FY27 Q4 target): Phase 3 MCP and runtime UI generation on track
  • Year 2 plan: Spec-as-Source maturity targets, remaining legacy team paths, Tier 4 agent pipeline
Infrastructure Complete
  • R9 Portal: AI-assisted search live; primary DS entry point across org
  • LDG fully populated; risk scoring operational
  • Cost monitoring system fully automated — no manual token spend logging
  • Weave brand fully established across org (internal and external comms)
Exit:First Spec-as-Source teams · 41-agent system operational · Annual review complete

12-month milestone timeline

M0 AI FDLC Review + rollout authorization
M1–M2First rollout cohort onboarded
M2Tier 1 agents live; LDG seeded
M3MCP Phase 2 build complete
M4Pilot teams → Spec-Anchored
M5A1 screening + X-domain agents live
M6MCP Phase 2 live; context packs thinned
M7–M8All eligible domains Spec-First
M9Phase 3 MCP scoped; Horizon 2 gate
M11First Spec-as-Source authorized
M12 Annual AI FDLC Review + Year 2 plan
02 · Workflow RACI — Year 1 Post-Pilot State

Spec-First with Tier 1 agents active.

A New Capability feature flowing through a domain that has completed rollout onboarding and is operating at Spec-First maturity. LDG operational, A1 screening live, agents shouldering primary inputs at every gate.

Scope of this RACI: A New Capability feature in a domain that has completed rollout onboarding and is operating at Spec-First maturity. Tier 1 agents are active. The LDG is machine-readable and A1 screening is operational. The trio is an established team; Architecture engages conditionally per the Trio Operating Contract. Agent tags indicate where an agent's output is the primary input to the activity.
Activity PM Design Lead Eng Lead Architecture DS Team EngOps Del Owner
Phase 0 — Signal & Discovery (continuous)
Signal intake & triage VoC SynthesisBehavioral Signal
Review agent-surfaced signal cards; promote to opportunity queue
RCII
Opportunity sizing Opportunity Sizing
Review agent evidence package; validate ARR sizing before G1
ACI
Phase 1 — Discovery thread (G0→G1)
Problem statement authoring Assumption Audit
Author problem statement; review assumption audit log before G1
RCC
Feasibility pre-screen Feasibility Pre-Screen
Engineering Lead reviews agent pre-screen; confirms complexity signal
IIRC
G1 gate sign-off
Problem validated; feature type assigned; P7 row selected
ARR
Phase 2 — Spec Authoring (G1 → G2)
Spec authoring — §1 (problem frame, ACs)
PM authors with agent assist; Engineering and Design co-review
RCC
Spec authoring — §2 (data shape, states, hooks)
Engineering Lead authors with PM and Design co-authoring interaction model
CCRC
Dependency declaration A1 Screening
Engineering Lead reviews A1 screening output; declares contracts against LDG
IRC
Pre-G2 quality checks A3 SufficiencyAC QualityConstraint Inventory
PM resolves sufficiency and AC quality findings; Design reviews constraint inventory
ARC
G2 gate sign-off
Spec locked in git; all trio roles co-sign; architecture engaged if triggered
ARRCI
Phase 3 — Design generation + validation thread (G2→G3)
Context loading & session setup Pattern-Aware Pre-Loader
Design Lead loads constitution, companion file, spec; reviews pattern pre-load brief
RIC
Variant generation Generation AgentCopy Generation
Design Lead guides generation; reviews variants and companion microcopy
IRC
Post-generation validation Spec-to-Proto ReconciliationL10n Pre-ScreenInteraction Cost
Design Lead reviews reconciliation diff, l10n flags, and interaction cost report
CRCC
Spec amendment (if gap discovered)
PM amends spec; decision log entry required; git commit before session resumes
RCC
Pre-G3 output preparation
Output suite generation Output GenerationTest ScaffoldingAnalytics Validation
Design Lead reviews; Engineering Lead reviews test scaffolds and analytics plan validation
ARC
G3 gate sign-off
G3 pre-review brief G3 Review BriefAC Implementability
Engineering Lead reviews both agent outputs before G3 session opens
CIRC
G3 gate sign-off
Engineering Lead accepts design deliverable; feasibility confirmed; decision log current
CRACI
Phase 4 — Engineering AI implementation thread (G3→G4)
Engineering spec draft Eng Spec Draft
Engineering Lead reviews and approves agent-drafted engineering spec
IRC
Implementation units Engineering Build
Engineering Lead directs build; two-step review per unit; decision log current
IRI
Decision log integrity check A2 Decision Log
Agent runs as blocking pre-gate check; Engineering Lead resolves any gaps
IR
CI/CD gate passage Eng Build Agent
Playwright + DS token lint + Axe-core must pass; first-attempt rate logged
IRCR
PR description generation PR Description
Agent generates; Engineering Lead reviews before submit
R
G4 gate / AC verification
PM verifies all spec ACs met; Engineering Lead attests team-wide pattern application
ACRII
Phase 5 — Ship (Release → Launch)
Rollout configuration Rollout Config
Agent generates flag config; Engineering Lead reviews; PM owns flag flip (Launch)
ARC
Release documentation Docs Continuity
Agent drafts release notes, help center scaffold, CSM summary; TW owns review + publish
ICI
Phase 6 — Post-ship close-out
Pattern candidate surfacing Pattern CandidatePattern Reuse Impact
Design Lead reviews candidates; DS Team evaluates for promotion
RA
Close-out checkpoint
Trio retrospective; amendment candidates identified; cost log finalized
RRRA
JTBD completion rate monitoring JTBD Completion RateFriction Signal
PM reviews rate trajectory; Design reviews friction synthesis weekly
RCI
Enhancement candidate routing Enhancement Candidates
PM reviews agent-generated candidates; promotes qualifying candidates to Phase 0 queue
ACI
AI FDLC Review input preparation
Del Owner synthesizes cost data, amendment candidates, and maturity advancement evidence
CCCCCCR
R Responsible — does the work
A Accountable — owns the outcome
C Consulted — input required before proceeding
I Informed — kept current; no action required
Agent Agent output is the primary input to this activity
03 · System Architecture

Year 1 post-pilot — Spec-First → Spec-Anchored.

The system architecture the workflow operates within at one year post-pilot: Tier 1–2 agents active, infrastructure context layer flanking the workflow, cross-domain dependency system on the right, post-ship discovery loop closing back into Phase 0.

Workflow phases & gates
AI agents
Infrastructure / context layer
Cross-domain dependency system
Governance cadences
Post-ship discovery loop
Data flows
Infrastructure
Platform Constitution
v1.2 · git-resident
Design Constitution
Textura · Principles layer
Domain Companion Files (per domain)
Design + Engineering
Textura MCP Phase 2
Patterns · Recipes · Validate
Pattern Library v2 (8–12 patterns)
DS Team · promoted from features
spec.md + decision-log.md
git-resident · per feature
Prompt Template Library
FDLC repo / prompt-templates/
Analytics Plan + Telemetry Infra Inventory
Per feature · instrumented
CI/CD Pipeline · Gates 0–3
GitHub Actions + ADO
Feature Flag System
Stability window enforced
Staleness Monitor (B1)
Companion file currency
Discovery · Ph0 → Ph1
VoC SynthesisGong/Chorus → signal queue
Behavioral SignalAnalytics anomalies
Persona DriftBehavior vs. persona delta
Opp. SizingARR + adoption projection
Assumption AuditPre-G1 assumption log
Feasibility Pre-ScreenComplexity signal pre-G1
G1
Spec Authoring · G1 → G2
Sufficiency (A3)P7 check pre-G2
AC QualityTestable · Independent · Mapped
Dep. Screening (A1)Spec vs. LDG → undeclared deps
Constraint InventoryDS rules + arch constraints map
Cross-Domain Align. (X-DOM)Multi-spec conflict detection pre-G2
spec.md (G2-locked)§1 problem + ACs · §2 data/states/hooks · §5 analytics plan
decision-log.mdAll amendments logged · git-resident
G2
Decision Log (A2)Blocking pre-gate check · G2→G4
Design Generation & Validation · G2 → G3
Pattern Pre-Loader (C2)Pattern library → session brief
Generation AgentSpec + constitution + companion file · Textura MCP Phase 2 active · Sonnet (Spec-First) / Opus (New Paradigm)
Copy GenerationMicrocopy alongside variants
Spec-to-Proto Reconcil.Silent drift → structured diff
L10n Pre-Screen + Interaction Cost / Cognitive Load
Output Suite GenerationQA cases · annotations · analytics plan
Test ScaffoldingPlaywright + unit stubs from ACs
LLM-as-Judge (Spec-Anchored teams only)Tier B–C rubric · Opus 4 · Blocking finding = no G3
G3 Review Brief Agent
AC Implementability
G3
Engineering Build · G3 → G4
Eng. Spec Draft AgentFrom G3-approved inputs · Sonnet
Engineering Build AgentUnit briefs → implementation · Two-step review per unit · Sonnet 4 · decision log current
PR DescriptionUnit brief → structured PR
Rollout Config AgentFlag config · segment rules · schedule
CI/CD Gate PassagePlaywright visual regression · DS token lint · Axe-core · First-attempt rate logged
G4
Release → Launch
Docs Continuity Agent (C3)
Cycle Time Telemetry (B2) · Cost log
Post-Ship Learning · Ph8 → Ph0
JTBD Completion RateContinuous · threshold alerts
Friction Signal Synth.Session recordings + support
Enhancement CandidatesPh0 signal brief format → queue
Pattern Candidate + Reuse ImpactClose-out → DS Team for promotion
Close-Out CheckpointConstitution amendment candidates → R2 format · Feature ROI Analysis (60–90 days)
Governance · AI FDLC Review
AI FDLC ReviewMonthly during rollout · Quarterly at scale · Amendment decisions · Maturity advancement · Cost review
SDD Maturity ArcPre-Spec-First → Spec-First → Spec-Anchored → Spec-as-Source · Pilot teams: Spec-First · Q2: first teams → Spec-Anchored · Q4: first teams → Spec-as-Source
Cross-Domain
Living Dependency Graph (LDG)
YAML · FDLC repo · Architecture owns schema
Contract Change Notification
Impacted trios alerted
Cross-Domain Spec Alignment
Multi-spec conflict detection
Shared Component Impact Agent
DS change → feature impact
Dependency Risk Scoring
Q3+ · N≥10 features
Cross-Domain Pattern Alignment
Q3+ · 3+ active domains

The post-ship discovery loop represents the structural loop closure: post-ship JTBD completion rate signals and enhancement candidates route back to Phase 0 as opportunity queue entries — making the FDLC self-generating rather than a one-directional pipeline. The cross-domain system on the right is not a phase in the workflow; it monitors and intercepts across all phases continuously via the Living Dependency Graph.

04 · Workflow RACI — Spec-Anchored Maturity

Post-pilot advancement — judgment load shifts to exception handling.

LLM-as-judge active, amendment protocol operational. Three structural shifts distinguish Spec-Anchored: spec versioned in the feature branch, LLM-as-judge as a blocking gate at G3, and an amendment protocol that runs as a constitutional-grade process.

What changes at Spec-Anchored: Three structural shifts distinguish this RACI from the Spec-First version. (1) The spec is versioned alongside the feature branch — drift is detected mechanically and triggers a formal amendment process rather than being caught in human review. (2) LLM-as-judge runs at G3 as a blocking gate for Tier A findings — not advisory. (3) The amendment protocol is a constitutional-grade process: agent flags drift, proposes the amendment entry, and human approves. The trio's judgment load shifts from error-catching to exception handling — the system catches most errors; humans resolve cases the system can't classify.
New at Spec-Anchored
  • LLM-as-judge (Tier A–C rubric) is a blocking gate, not advisory
  • Spec drift triggers formal amendment — not caught in human review
  • Amendment agent proposes entry; PM approves; spec versioned before session resumes
  • Decision log integrity (A2) moves from pre-gate check to continuous monitoring with hard block
  • Constitution amendment candidates produced at close-out via R2 format
  • G3 pre-review brief now includes LLM-as-judge pre-run summary
  • Spec-Anchored teams eligible for pattern candidate promotion without DS lead initiation
Removed or reduced at Spec-Anchored
  • Manual drift-catching in Design's post-generation review — now agent-handled
  • Engineering lead manually pre-reading full feature folder before G3 — replaced by G3 brief + LLM judge pre-run
  • PM manually checking decision log completeness before gate — now a continuous blocking check
  • DS Team Consulted on all generation sessions — now Informed unless pattern promotion in scope
  • Full output suite for every feature type — tiered by feature type; Enhancement suite abbreviated
Activity PM Design Lead Eng Lead Architecture DS Team EngOps Del Owner
Phase 0–1 — Signal & discovery thread — unchanged from Spec-First; signal agents fully operational
Signal triage + opportunity sizing VoCBehavioralOpp. Sizing
All Phase 0 agents operational; agent-generated sizing package reviewed before G1
RCII
Feasibility pre-screen + assumption audit FeasibilityAssumption Audit
Both agents run pre-G1; Engineering Lead and PM review outputs before gate
ACRC
Phase 2 — Spec Authoring (G1 → G2) — spec versioned in feature branch; amendment protocol enforced
Spec authoring — all sections A3 SufficiencyAC Quality
PM authors §1; Engineering Lead authors §2; agent quality checks run continuously during authoring, not only pre-gate
RCRC
Dependency declaration + cross-domain check A1 ScreeningX-Domain Align
A1 screens against LDG; cross-domain alignment agent runs if multiple active specs share the same entity. Both must clear before G2.
IRA
Spec versioning at G2 lock
spec.md committed to feature branch; baseline version tagged; subsequent changes only via amendment protocol — no informal edits
ACRI
Phase 3 — Design generation + validation thread (G2→G3) — drift detection automated; amendment protocol active
Context loading + constraint inventory Constraint InventoryPattern Pre-LoaderB1 Staleness
Staleness monitor confirms companion file is current before session opens — session does not begin on a stale companion file
RCI
Variant generation + copy Generation AgentCopy Generation
Generation runs within versioned spec; Design Lead supervises; sliding window enforced; Textura MCP Phase 2 active throughout
IRI
Spec drift detection Spec-to-Proto Reconcil.A2 Decision Log
▲ SPEC-ANCHORED CHANGE: Reconciliation agent runs continuously during generation (not post-session). Any drift above threshold triggers automatic pause and amendment protocol — generation does not continue on a drifted spec.
ARC
Amendment protocol execution Amendment Agent
▲ NEW AT SPEC-ANCHORED: Agent detects drift → proposes structured amendment entry with rationale → PM reviews and approves or rejects → spec versioned in git → decision-log.md updated → session resumes. No generation on unapproved spec version.
ACC
Post-generation validation suite L10nInteraction Cost
L10n pre-screen and interaction cost report run automatically post-selection; findings routed to Design Lead for resolution before output suite
CRCI
Output suite — tiered by feature type Output SuiteTest ScaffoldingAnalytics Validation
▲ SPEC-ANCHORED CHANGE: Enhancement features receive abbreviated suite (QA cases + decision rationale only; engineering annotations on Engineering Lead request). New Capability and New Paradigm receive full suite. Feature type set at G1; suite config locked in spec template at G2.
ARC
G3 gate sign-off — LLM-as-judge is a blocking gate; Tier A findings prevent G3 from clearing
G3 pre-review package G3 Review BriefAC ImplementabilityLLM-as-Judge Pre-run
▲ SPEC-ANCHORED CHANGE: LLM-as-judge runs Tier A (automated coverage checks) before G3 opens. Engineering Lead receives: G3 brief + AC implementability assessment + Tier A findings. Tier A blocking findings are resolved before G3 session, not during it.
CRACI
LLM-as-judge Tier B–C evaluation LLM-as-Judge (Opus 4)
▲ NEW AT SPEC-ANCHORED: Tier B (design constitution compliance; heuristic evaluation) and Tier C (design quality) run at G3. Tier B findings are blocking. Tier C findings are advisory — Design Lead disposition required in writing before G3 clears. Engineering Lead is informed of all findings; does not override Design Lead disposition.
IAIICI
G3 gate sign-off
Tier A and Tier B findings resolved; Tier C dispositions documented; Engineering Lead accepts design deliverable; all agents confirm clean state
CRACI
Phase 4 — Engineering AI implementation thread (G3→G4) — unchanged from Spec-First; decision log integrity now continuous
Implementation units + two-step review Engineering BuildPR Description
Engineering Lead directs build; two-step review per unit; decision log enforced continuously by A2 agent — no gate needed to surface gaps
IRI
CI/CD gate + rollout config CI/CD GatesRollout Config
All CI/CD stages pass; rollout config generated and reviewed; stability window enforced in schedule
ARR
G4 gate / AC verification
PM verifies all spec ACs met against versioned spec; Engineering Lead attests pattern application; cost log complete
ACRII
Phase 5–6 — Ship, post-ship & constitution contribution — constitution amendment pipeline active at Spec-Anchored
Pattern promotion Pattern CandidatePattern DocumentationReuse Impact
▲ SPEC-ANCHORED CHANGE: Domain team initiates pattern promotion — does not wait for DS Team to identify candidates. Agent drafts documentation; DS Team reviews and approves. Reuse impact agent notifies active trios immediately on promotion.
IAIR
Constitution amendment candidate submission C1 Amendment Candidate
▲ NEW AT SPEC-ANCHORED: Close-out checkpoint produces structured amendment candidates via R2 format. C1 agent cross-references decision log + spec gap log against current constitution — surfaces candidates the trio didn't identify. PM and Engineering Lead jointly review; Del Owner authorizes submission to AI FDLC Review.
RCRCA
JTBD monitoring + loop closure JTBD Completion RateFriction SignalEnhancement Candidates
All post-ship agents operational; enhancement candidates auto-generated and queued for PM review; PM promotes qualifying candidates to Phase 0
RCII
AI FDLC Review input — constitution amendments
Del Owner synthesizes amendment candidates from close-out records across domains; AI FDLC Review approves, modifies, or defers each; approved amendments versioned before next generation session in affected domains
CCCRCA
R Responsible
A Accountable
C Consulted
I Informed
Agent Agent output is the primary input
▲ marks activities that change at Spec-Anchored vs. Spec-First
05 · Domain & Team Maturity Roadmap

What each domain owns, governs, and optimizes.

From onboarding through Spec-as-Source — the maturity arc is about how much of the workflow a domain has made its own. Central governance sets the foundation; domain investment is what compounds quality on top of it.

The ownership model: The AI FDLC is centrally governed but domain-operated. The central system (constitutions, LDG schema, agent infrastructure, AI FDLC Review cadence) is maintained by Architecture, DS Team, and the Del Owner. Everything domain-specific — companion files, pattern contributions, LDG entries, analytics taxonomy, test frameworks, cost behavior — is owned by the domain trio and Engineering Lead. A domain that doesn't invest in its own context infrastructure is capped at the quality of the shared foundation. A domain that does invest compounds that foundation with domain-specific precision. The maturity arc is about how much of the workflow a domain has made its own.

Ownership surface — Central vs. Domain

Central ownership Architecture · DS Team · Del Owner
  • Platform Constitution — rules that apply across all domains; Architecture owns; AI FDLC Review authorizes amendments
  • Design Constitution — DS Team Lead authors and maintains; principles layer; cross-domain heuristics
  • LDG schema — Architecture owns schema definition and enforcement rules; domains populate their entries
  • Textura MCP server — DS Engineering Lead; component coverage, Phase 2 tools, version management
  • AI FDLC Review cadence — Del Owner runs; advancement decisions, amendment approvals, maturity governance
  • Agent infrastructure — prompt templates, model routing policy, cost monitoring system, agent deployment
  • Pattern library (promoted patterns) — DS Team owns promotion decisions and library curation; domain teams contribute candidates
  • FDLC repo structure — Engineering Ops maintains; shared tooling, CI/CD pipeline templates, CLAUDE.md base
Domain ownership PM · Design Lead · Engineering Lead
  • Domain design companion file — Design Lead authors and maintains; patterns, DS coverage, heuristics for this domain
  • Domain engineering companion file — Engineering Lead authors and maintains; service contracts, test patterns, DS token namespace
  • LDG entries for this domain's services — Engineering Lead owns; contract versions, ownership, known consumer teams
  • Domain context pack — Design Lead or Engineering Lead per domain; entity models, service hooks, experience-type context
  • Domain analytics taxonomy — PM owns; JTBD success events, segment definitions, threshold configuration
  • Domain test framework — Engineering Lead; test scaffold conventions, Playwright page objects, integration test harness
  • Pattern candidates — Design Lead identifies; close-out checkpoint surfaces; DS Team promotes
  • Constitution amendment candidates — PM + Engineering Lead produce via close-out; Del Owner authorizes submission

Domain maturity arc — artifacts, tooling, and optimization responsibilities by stage

Stage 1 — Entry Spec-First Onboarding Before first G2 · Setup Complete declared
Artifacts to author
  • Domain design companion file — Design Lead; required before first G2
  • Domain engineering companion file — Engineering Lead; seven sections + D12 addendum fields
  • LDG entries for all services owned by this domain — Engineering Lead
  • Domain context pack v1 — entity models, service hooks, experience type
  • Baseline feature data template — PM completes before pilot feature begins
Tooling to configure
  • CI/CD stages 0–1 active (GitHub Actions + ADO) — Engineering Lead + EngOps
  • CLAUDE.md configured with domain engineering companion file path
  • Textura MCP version pinned to match @exp-textura/react version
  • Figma project linked (if Figma roundtrip in scope)
  • AHA initiative linked to feature tracking; ADO board configured for gate state tracking
Workflow responsibilities
  • All trio roles read: PM spec guide (D10), workflow walkthrough, Trio Operating Contract
  • Engineering Lead: read G3 pre-review agent brief; confirm Track A/B applicability
  • Design Lead: confirm companion file currency before every generation session
  • PM: complete at least one spec against P7 sufficiency table before G2
  • All trio roles: complete decision log startup guidance
Exit criteria
  • All Setup Complete conditions met — no partial onboarding
  • Companion files reviewed and confirmed current by DS Team Lead
  • LDG entries populated and validated by Architecture
  • CI/CD stages 0–1 passing on a test run
  • First feature at G0 — pilot begins
Stage 2 — Operating Spec-First Running Features flowing through G0–G5 · Measurement producing signal
Artifacts to maintain
  • Companion files updated after each constitution amendment — Engineering Lead + Design Lead; within one sprint of amendment
  • LDG entries: update contract versions when service changes; notify consuming teams via contract change notification agent
  • Context pack updated when domain entity models change or new service hooks added
  • Close-out records: pattern candidates, spec gap log, cost log complete at each G4
  • Decision log: all amendments logged; no informal spec changes
Tooling to optimize
  • CI/CD stages 2–3 activated when ready — Chromatic visual regression, full Axe-core enforcement
  • Analytics plan instrumented for first shipped feature; JTBD success events firing
  • Track B advancement evaluated: if Track A ceiling hitting, Engineering Lead makes case to Del Owner
  • Test framework expanded: domain page objects for Playwright, integration test harness for declared contracts
  • Prompt caching confirmed active for constitution + companion file in all agent sessions
Workflow responsibilities
  • PM: log token spend at every gate; flag cost anomalies using efficiency ratio thresholds from cost guide
  • Engineering Lead: collect post-G3 clarification overhead hours per feature; report in close-out
  • Design Lead: identify pattern candidates at close-out; submit via DS Team review process
  • PM: activate JTBD completion rate monitoring after first feature ships; set thresholds
  • All trio: structured retrospective notes at G3 and G4 — not reconstructed after the fact
Exit / advancement criteria
  • 3+ features completed G0→G4 with complete artifacts (spec, decision log, cost log, close-out record)
  • G4 first-attempt CI/CD pass rate ≥ baseline target
  • Post-G3 clarification overhead trending toward target (sub-1hr)
  • Amendment log in use — no known informal spec changes
  • AI FDLC Review advancement decision for Spec-Anchored
Stage 3 — Advancing Spec-Anchored Amendment protocol live · LLM-as-judge blocking · Pattern pipeline active
Artifacts to build
  • Domain pattern library section: 3+ patterns promoted; usage guidance and code references complete
  • Amendment submission record: structured R2-format candidates from each close-out; at least one constitution amendment contributed
  • Domain analytics runbook: JTBD success events, segments, threshold rationale, enhancement candidate criteria
  • Spec amendment taxonomy: common amendment types for this domain catalogued; patterns in prompt templates
  • Domain test pattern library: reusable test patterns shared across features in this domain
Tooling to deepen
  • LLM-as-judge fully operational: Tier A automated; Tier B–C run by Design Lead; disposition workflow established
  • Domain prompt library: domain-specific generation prompts refined from feature retrospectives; stored in FDLC repo
  • Companion file versioned alongside constitution: Design Lead + Engineering Lead update on every constitution amendment; staleness monitor alert = same-sprint fix
  • Enhancement candidate queue active: PM reviews weekly; qualifying candidates promoted same sprint as generated
  • Feature ROI analysis (60–90 day): first domain feature ROI package produced and shared with Del Owner
Workflow to optimize
  • Generation stop rate: Design Lead tracks stops per session; target <1 per session; companion file gaps are primary cause
  • Amendment frequency: PM tracks amendments per feature; elevated rate on simple features = G2 spec quality gap
  • Cost-per-AC-implemented: Engineering Lead tracks as efficiency signal; rising rate = correction cycle problem
  • Model override log: review monthly; recurring overrides = routing policy candidate
  • Pattern reuse rate: Design Lead tracks how often existing patterns are used vs. novel compositions; target increasing over time
Exit / advancement criteria
  • G4 first-attempt CI/CD pass rate at or above target for 3+ consecutive features
  • Post-G3 clarification overhead consistently sub-1hr
  • At least one constitution amendment candidate submitted and accepted
  • Domain pattern library section has promoted patterns referenced in generation sessions
  • Cost-per-outcome ratio established and stable
  • AI FDLC Review advancement decision for Spec-as-Source evaluation
Stage 4 — Scaling Spec-as-Source Pathway AI output trusted at high reliability · Human review focuses on exception handling
Artifacts to contribute back
  • Recipe authoring: domain team authors 1–2 recipes for Textura MCP Phase 2/3 from proven feature compositions
  • Experience type pack contribution: domain signals to DS Team which patterns are stable enough to promote to experience-type level
  • Constitution contribution: domain's amendment history used as evidence for constitution evolution proposals at AI FDLC Review
  • Onboarding playbook for domain: documents domain-specific companion file gotchas, common generation stops, and test framework conventions for teams onboarding into this domain later
Tooling at full depth
  • Cross-domain pattern alignment: Design Lead monitors agent output for patterns this domain has developed that adjacent domains are independently reinventing; surfaces to DS Team
  • Dependency risk scoring: Engineering Lead reviews risk scores before G1 on complex features; high-risk deps escalated to Architecture proactively
  • Companion file auto-draft: when constitution is amended, companion file draft generated automatically; domain lead validates rather than authoring from scratch
  • Tier 3 autonomous agent evaluation: domain can nominate a narrow, low-risk scope for autonomous execution with AI FDLC Review approval
Workflow to govern
  • Generation reliability: Engineering Lead tracks first-attempt code quality (linting pass rate, test scaffold adoption rate); declining reliability = constitution or companion file gap
  • Autonomous agent scope: any Tier 3 agent running in this domain has a defined scope boundary; Engineering Lead owns boundary enforcement; boundary changes require AI FDLC Review
  • Onboarding new teams into domain: domain trio leads the onboarding of adjacent teams who need to understand domain service contracts and companion files
  • Pattern deprecation: Design Lead owns deprecating domain patterns that have been superseded; coordinates with DS Team on library cleanup
At scale — what this looks like
  • Spec-as-Source: PM authors spec → spec feeds generation and build directly; human review focuses on scope judgment and architectural exceptions, not error-catching
  • Close-out is a constitution contribution exercise as much as a retrospective — the domain's production history is a primary input to system evolution
  • Domain team mentors newer domains: expertise compounds outward through the org, not only upward to central governance
  • Cost-per-outcome ratio is a management metric — del owner uses it to prioritize domain-level infrastructure investment

Domain governance responsibilities by artifact — summary

Artifact / area Who authors Who reviews Update trigger Cadence Unlock stage
Domain design companion fileDesign LeadDS Team LeadConstitution amendment · Pattern promotion · DS component deprecationSame sprint as trigger · Staleness monitor flagStage 1
Domain engineering companion fileEngineering LeadArchitectureConstitution amendment · Service contract change · New tech introducedSame sprint as triggerStage 1
LDG entries (services owned)Engineering LeadArchitectureContract change · New service · Consumer team addedEvent-driven; not periodicStage 1
Domain context packDesign Lead or Engineering LeadDS Team (design pack) · Architecture (eng pack)Entity model change · New service hook · Experience type changePer feature if domain model changes; quarterly review minimumStage 1
Domain analytics taxonomyPMDel OwnerNew feature shipped · JTBD success event definition changesPer feature; reviewed at close-outStage 1 (basic) · Stage 2 (full)
Domain test frameworkEngineering LeadEngOpsNew test pattern proven across 2+ features · Playwright page object gapPer feature at close-out; quarterly library reviewStage 1 (baseline) · Stage 2 (expanded)
Pattern candidatesDesign Lead (identifies)DS Team (promotes)Novel composition validated at G3; appears in 2+ featuresClose-out checkpointStage 2
Domain prompt libraryDesign Lead + Engineering LeadDel OwnerGeneration session retrospective identifies better prompts · Model routing override recursMonthly or per anomalyStage 2
Amendment candidates (R2 format)PM + Engineering LeadDel Owner → AI FDLC ReviewClose-out checkpoint; C1 agent surfaces additional candidatesPer close-outStage 3
Domain pattern library sectionDesign Lead (drafts via agent)DS Team (approves)Pattern candidate promoted · Pattern deprecated · Usage guidance updatedPer promotion eventStage 3
Feature ROI packagesPM (narrative) + agent (evidence)Del Owner60–90 days post-ship for each featurePer feature at N-day markStage 3
Domain onboarding playbookAll trio leads jointlyDel OwnerWhen adjacent teams begin onboarding into domain; updated after each new team onboardsEvent-drivenStage 4
Recipe contributions (Textura MCP)Engineering Lead + Design LeadDS Engineering Lead (Tyler)Composition proven stable across 3+ features; candidate for MCP toolQuarterly review; per Phase 2/3 build cycleStage 4
A domain that doesn't invest in its own context infrastructure is capped at the quality of the shared foundation. A domain that does invest compounds that foundation with domain-specific precision. Ownership model · AI-Enhanced FDLC