Cross-Cutting — In Practice
Some practices apply across every stage. Governance wraps Direction, Issues, Work Items, Delivery, and Operation. It does not sequence after them — it is active continuously.
Relationship Map
Governance (metastage)
│
├── Principles — 8 core principles shape every stage
├── Policies — 13 policies with clear enforcement
├── Metrics — DORA 4, flow metrics, quality metrics
├── Process Improvement — retro actions drive improvement
└── Friction Budget — process costs capped at 20%
Process Improvement (from retro actions) adjusts how every stage operates over time.
Principles Applied Per Stage
| Principle | In Direction | In Issues | In Work Items | In Delivery | In Operation |
|---|
| Ship small, learn fast | One Mission at a time | One Issue at a time | One Aggregate per Use Case | Smallest valuable batch per cycle | Deploy incrementally; canary before rollout |
| Feedback is ground truth | Production data changes Missions | User reports generate Issues | Data beats opinion in prioritisation | DORA metrics, demo feedback, production signals | Monitoring and incidents guide everything |
| Process has cost | Do not explore if domain is understood | Do not debate type when path is clear | Do not write Use Cases for trivial work | Skip TDD for config changes; skip gate for low-risk | Remove alerts that never fire |
| Unknowns kill delivery | Study unknowns during exploration | Study when root cause is unknown | Study before committing to a Feature | Study if acceptance criteria are unclear | Investigate root cause before treating symptoms |
| Complexity contained | Bounded contexts via Context Mapping | — | One Aggregate per Use Case | Application layer coordinates between Aggregates | Runbooks contain operational complexity |
| Teams make predictable errors | — | — | Peer review catches spec gaps | Automated CI gates catch integration errors | Runbooks prevent repeat incidents |
| Judgment over rules | Skip Event Storming if domain is shared | Skip formal triage if type is obvious | Skip Use Case if Story is clear | Skip any technique that does not fit | Skip any practice that does not fit team size |
| Prepare to succeed, plan to fail | Do enough discovery, then stop | — | Impact-rank; do not estimate by default | Release plans exist only for failure scenarios | Rollback plans exist only for failure |
Policy Table
| Policy | Applies To | What It Means | Enforced By |
|---|
| No Broken Code on Main | Delivery (Build) | Every commit must pass CI. Red build is top priority. Revert if not fixed in minutes. | CI pipeline (automated) |
| Feature Flag Off by Default | Delivery (Release) | Every user-facing change deploys behind a flag, off by default. | Release Owner |
| No Gatekeeper | All stages | Any team member can talk to a stakeholder directly. No ticket-forwarding. | Team culture |
| Face-to-face first | All stages | Co-located: conversation beats tickets. Remote: sync call beats threaded decision. | Team culture |
| One Aggregate per Use Case | Work Items | Do not cross aggregate boundaries in one Use Case. Split the Story. | Code Review |
| No Future bucket | Work Items (Roadmap) | Roadmap has Now and Next only. Ideas live in Issues. | Product Owner |
| No estimates by default | Work Items | Impact-rank instead. Estimate only for external commitments. | Whole team |
| Tests during Build | Delivery (Build) | Tests written during Build, not after. They define “done.” | QA Engineer |
| No manual-only testing | Delivery (Verify) | Automation is default. Manual is exception for complex UI only. | QA Engineer |
| Alert on symptoms | Operation (Monitoring) | If an alert does not require action, remove it. | SRE |
| One dashboard per audience | Operation (Monitoring) | One for stakeholders, one for the team. Do not mix. | SRE |
| Error budget decides | Operation (SLOs) | Full → ship features. Low → fix reliability. | Product Owner + SRE |
| Production Readiness mandatory | Operation | New service must pass checklist before traffic. | SRE, Security Engineer |
| Friction Budget <20% | All stages | Process activities capped at 20%. Cut one item if exceeded. | Whole team |
| Judgment over rules | All stages | No rule applies in all contexts. Team decides when to bend or break. | Whole team |
Process Improvement
Retro → 1-3 items → Hypothesis → Owner → Timeline → Measure
→ Next retro: keep / adjust / discard
| Element | How It Works |
|---|
| Source | Retrospective action items |
| Hypothesis | “If we do X, then Y will improve” |
| Owner | One person per item |
| Timeline | One cycle or by a specific date |
| Measurement | Track Y before and after |
| Review | Next retro — keep, adjust, or discard |
Common improvements by stage:
| Stage | Typical Items |
|---|
| Direction | “Add domain exploration before Features” / “Shorten mission review cadence” |
| Issues | “Define SEV levels explicitly” / “Add automated alert → Issue capture” |
| Work Items | “Write acceptance criteria before building” / “Stop estimating” |
| Delivery | “Reduce CI build time” / “Add SAST gate” / “Timebox retro” |
| Operation | “Write missing runbook for X” / “Reduce toil by automating Y” / “Tune noisy alert” |
DORA Metrics
| Metric | What It Measures | Where | Target Signal |
|---|
| Deployment frequency | How often the team deploys | Release | Increasing over time |
| Lead time | Time from commit to production | Build → Release | Decreasing over time |
| Change failure rate | % of deployments that cause failure | Release | Low and stable |
| Time to restore | Time to recover from failure | Incident Response | Decreasing over time |
Skip when: Pre-PMF, team of 2-3, no production users.
Friction Budget Per Stage
| Stage | Process Activities (count towards 20%) | Delivery Activities |
|---|
| Direction | Workshops, review meetings | Productive discovery |
| Issues | Triage meetings, classification debates | Direct resolution |
| Work Items | Use Case writing, specification meetings | Building, coding |
| Delivery | Gate reviews, demo prep, retro facilitation | Coding, testing, shipping |
| Operation | Post-mortems, cost reviews, handover | Monitoring, incident response, automation |
Cut first when over budget: Meetings with no agenda, gates that never fail, reports no one reads, approvals always approved.
Agentic Roles
| Role | Active In | What They Do |
|---|
| Agent Orchestrator | All stages | Composes specifications, runs execution in agentic mode |
| Software Engineer | Specify, Build, Verify, Review, Operate | Writes code, runs TDD, pairs, integrates, deploys |
| QA Engineer | Specify, Verify, Review | Defines acceptance criteria, runs acceptance tests |
| Product Owner | Discover, Specify, Review | Ranks backlog, defines Features, release decisions |
| Release Owner | Review, Ship | Deployment pipeline, Go/No-Go, release coordination |
| Site Reliability Engineer | Operate | Monitors, handles incidents, on-call, SLOs |
| Platform Engineer | Build, Ship | CI/CD pipeline, infrastructure, deployment |
| Engineering Lead | Discover, Governance | Quality standards, architecture decisions |
| Security Engineer / Champion | All stages | Security gates, vulnerability management |
| FinOps Engineer | Operate | Cost management, budget tracking |
| Incident Manager | Operate | Incident coordination, post-mortem facilitation |
| Data Analyst | Discover, Operate | Production data signals for Direction |
| Documenter | All stages | ADRs, runbooks, institutional knowledge |
Procedure Table
Artifact Table
| Artifact | Produced By | Consumed By | Format |
|---|
| DORA Metrics | CI/CD pipeline + incident tracking | Iteration Review | Dashboard per cycle |
| Flow Metrics | Work tracking system | Iteration Review | Throughput, cycle time, WIP |
| Definition of Done | Team agreement | Acceptance Testing, Review | Shared checklist |
| Tech Debt Register | Software Engineer | Iteration Planning | Known debt with size and impact |
| Friction Budget Estimate | Whole team at retro | Process Improvement | Estimated % — not precise |
| ADR | Software Engineer | Architecture evolution | Context, options, trade-offs, decision |
Skip Conditions
| Activity | Skip When |
|---|
| DORA measurement | Pre-PMF, team <5, no production users |
| Definition of Done checklist | Team of 2-3 knows implicitly what done means |
| Tech Debt Register | No known debt; exploration mode |
| Friction Budget estimation | Team discusses process costs naturally |
| Formal Process Improvement tracking | Action items followed up informally |
| Agentic role assignment | Team is not operating in agentic mode |
| ADR writing | Decision is routine, reversible, or has no alternatives |
Use what fits. Skip what does not. If a policy slows the work more than it helps, suspend it.
Comments
Comments on each page are welcomed and are visible here and under the discussions tab of the github repository.
View discussions on GitHub ↗