Skip to main content

FAANG Incremental Delivery

Ship Fast, Learn Fast, Optimize Fast

xOps follows FAANG incremental delivery principles: working software every 2 weeks, evidence-based quality gates, autonomous PDCA cycles targeting ≥99.5% validated accuracy, and DORA metrics for continuous improvement.


Delivery Cadence

CadenceActivitiesDeliverable (Git-Tracked)
DailyCode commits with tests, agent coordinationPassing CI pipeline + coordination logs
Sprint (2 weeks)Working software increment, HITL review, retrospectiveTagged release + PDCA validation report
Release (per phase)Production deployment, customer validation, metrics baselineTerraform state + 4-way cross-validation evidence

DORA Metrics: Measuring Delivery Excellence

xOps tracks all 4 DORA metrics as git-tracked deliverables:

DORA MetricBC1 TargetHow MeasuredGit Artifact
Deployment Frequency1/sprint (2 weeks)git tag count per sprintRelease tags: v0.1.0-phase1 through v1.0.0
Lead Time for Changes<2 daysFirst commit → deploymentgit log --first-parent delta
Change Failure Rate<5%Failed deploys / total deploysCI workflow conclusion status in .github/workflows/
Mean Time to Restore<1 hourIncident open → resolutionIncident log in docs/docs/operations/incidents.md

Why DORA for xOps

DORA metrics are the industry standard (Google, FAANG, DORA State of DevOps Report) for measuring software delivery performance. For ANZ FSI enterprise buyers, DORA provides:

  • Benchmarking: Compare xOps delivery against industry elite performers
  • Predictability: Sprint-over-sprint trend lines for capacity planning
  • Trust: Evidence-based claims about delivery velocity, not opinions

Evidence Architecture: Git-Tracked vs Agent Workspace

Critical Distinction

tmp/ is ephemeral agent workspace — not permanent evidence. Git-tracked deliverables are the real evidence.

Evidence TypeLocationLifecyclePurpose
Git-tracked deliverablesRepository (src/, docs/, terraform/, .github/)Permanent, versioned, auditableProduction evidence — what was shipped
Agent workspace (tmp/)tmp/<project>/Ephemeral, session-scoped, gitignoredAgent coordination scratch space — how decisions were made
CI/CD artifactsGitHub Actions, Terraform CloudRetention-policy governedBuild evidence — how it was validated
DORA metricsRelease tags, CI logs, incident docsDerived from git historyDelivery performance — how well we're shipping

What Goes Where

ArtifactGoes InWhy
Terraform modules, source codesrc/, modules/ (git)Permanent, versioned, reviewable
Test results, coverage reportsCI artifacts + docs/ summaryReproducible from git
Architecture Decision Recordsdocs/docs/adrs/ (git)Permanent architectural knowledge
PDCA validation reportsdocs/ or release notes (git)Sprint-level quality evidence
Agent coordination logstmp/ (gitignored)Ephemeral — informs decisions, not the decision itself
Cross-validation raw datatmp/ during session, summary in docs/Raw data is transient; conclusions are permanent
FinOps FOCUS reportsdocs/ or S3 (git-tracked reference)Business deliverable, must survive sessions

PDCA Cycle: How Quality is Measured

Every deliverable goes through autonomous Plan-Do-Check-Act cycles before HITL review:

PLAN → DO → CHECK → ACT
↑ |
└────────────────────┘
(max 3 cycles before HITL escalation)
PhaseWhoWhatGit Deliverable
PLANPO + CAScore requirements, design architectureADR + INVEST stories committed to docs/
DOSpecialist agentsImplement, test, collect evidenceCode + tests committed to feature branch
CHECK4-way cross-validationValidate across boto3, MCP, Runbooks, ConsoleValidation summary in PR description
ACTAll agentsFix gaps, re-score, iterateFix commits on same branch

Targets:

  • Cross-validated accuracy: ≥99.5%
  • Agent consensus: ≥95% across PO, CA, MEE, IE
  • Max autonomous cycles: 3 (then HITL escalation)

Enterprise Coordination Protocol

Every feature follows the ADLC coordination chain — no exceptions:

HITL Manager Directive

1. product-owner (BLOCKING) → INVEST stories, acceptance criteria

2. cloud-architect (BLOCKING) → Architecture ADR, deployment target

3. Specialists (PARALLEL) → IE + QA + SCE + OE + FDE

4. Evidence collection → Git-tracked deliverables + tmp/ workspace

5. PDCA cycles until ≥99.5%

6. HITL review → Merge to main → Release tag

Anti-patterns prevented: 25 documented patterns blocked by 5 governance hooks. See PR/FAQ Q19 for the full hook reference.


HITL Gates per Phase

PhaseHITL RoleTime InvestmentGate CriteriaGit Artifact
PLAN (Wk 1-2)Give directive15 minAgent scores ≥90%ADR + stories in docs/
BUILD (Wk 3-4)Review code30 minLinting PASSED, tests GREENFeature branch merged
TEST (Wk 5-6)Approve evidence15 min4-way cross-val ≤0.5%Test report in CI
DEPLOY (Wk 7-8)SNS approve5 mincheckov 0 FAILED, infracost ≤+5%terraform apply state
MONITOR (Wk 9-10)Review SLOs15 minBaselines establishedSLO definitions in docs/

Total HITL time per sprint: ~80 minutes. Agents handle the remaining ~80 hours of execution.


xOps BC1 Delivery Timeline with DORA Tracking

PhaseSprintDeliverablePDCA TargetDORA: Deploy Freq
1Wk 1-2docker-compose + CLAUDE.md + Playwright≥90%v0.1.0-phase1
2Wk 3-4CloudOps-Runbooks MCP + Open WebUI pipeline≥95%v0.2.0-phase2
3Wk 5-6FinOps FOCUS 1.2+ + 4-way cross-validation≥95%v0.3.0-phase3
4Wk 7-8Terraform M3 + CI gates (checkov+trivy+infracost)≥99%v0.4.0-phase4
5Wk 9-10AWS deploy + 4-layer evidence + PDCA closure≥99.5%v1.0.0

Evidence

  • PDCA validation log: PR/FAQ Section 8
  • Cross-validation pattern: .claude/skills/finops/cross-validation.md
  • Anti-pattern list: .claude/rules/adlc-governance.md (25 patterns)
  • Coordination: tmp/adlc-framework/coordination-logs/product-owner-2026-03-11-docs-expansion.json