Overview
Goal
Document one real decision and one real incident to a professional bar: an RFC/ADR a peer can act on
without a meeting, and a blameless postmortem a stranger can follow -- proving your writing moves work
forward. This is a leadership ‡ design/decision capstone: no code, zero runnable files. Every
document format combined here was already taught, individually, somewhere in the Beginner,
Intermediate, or Advanced scenarios of this topic; this capstone is where they run together on one
coherent story, at a larger and more realistic scale than any single worked scenario.
The story: on 2026-04-02, a subset of Harborlight customers received duplicate shipment notifications after a Notification Worker deployment restart replayed already-processed messages. The RFC and ADR below design and decide the permanent fix; the PR implements it; the postmortem and its C4 context diagram document the incident itself, with the RFC/ADR/PR trio appearing inside the postmortem's own follow-ups as the owned fix.
%% Color Palette: Blue #0173B2, Orange #DE8F05, Teal #029E73, Purple #CC78BC, Brown #CA9161
flowchart LR
A["RFC<br/>options + trade-off + review"]:::blue
B["ADR<br/>decision + status + consequences"]:::orange
C["PR description<br/>what#47;why#47;verified#47;where-to-look"]:::teal
D["Postmortem + C4 context<br/>timeline + root cause + follow-ups"]:::purple
A --> B --> C --> D
classDef blue fill:#0173B2,stroke:#000000,color:#FFFFFF,stroke-width:2px
classDef orange fill:#DE8F05,stroke:#000000,color:#FFFFFF,stroke-width:2px
classDef teal fill:#029E73,stroke:#000000,color:#FFFFFF,stroke-width:2px
classDef purple fill:#CC78BC,stroke:#000000,color:#FFFFFF,stroke-width:2px
Diagram: the capstone's build order -- deliberate the fix in an RFC, decide it in an ADR, implement it in a reviewed PR, and close the loop with a blameless postmortem whose own follow-ups point back at the RFC/ADR/PR trio as the owned fix. (The underlying incident itself happened first, on 2026-04-02, before any of these four documents existed -- the postmortem's timeline records that; this diagram is the order the four documents were written in, not the order events occurred.)
Concepts exercised
- BLUF/reader-first structure (co-01, co-02, co-03)
- an ADR (decision + status + consequences, colocated) (co-06, co-07, co-08)
- an RFC with options + trade-off + review (co-04, co-05)
- a PR description (co-09)
- a blameless incident write-up (co-10)
- a C4-style diagram (co-11, co-12)
- precise RFC 2119 keywords + edited, register-matched prose (co-13, co-14, co-15)
- proportional, close-to-code artifacts (co-16, co-17)
Step 1: ADR and RFC
exercises co-01, co-02, co-03, co-04, co-05, co-06, co-07, co-08, co-13, co-16, co-17
The RFC states the problem (duplicate notifications after a worker restart), weighs two options (a Redis-backed idempotency cache versus a full transactional-outbox rearchitecture) with explicit pros and cons for each, names the decisive trade-off, and carries a review log showing every question raised was resolved or deferred before its "Accepted" status. The ADR distills that accepted RFC into a single, durable decision record -- context, decision, consequences, a status field, dated and colocated with the Notification Worker's own code -- without reproducing the RFC's full deliberation.
Verify: a peer reviewer can restate the decision and its rationale from the ADR and RFC alone, without a follow-up meeting -- the same reader-review rubric Worked Scenario 25 already demonstrated, applied here to the capstone's own documents.
Step 2: PR description
exercises co-09
The PR description describes the actual code change that implements the ADR's decision: what changed, why, how it was verified, and -- because the diff touches several files -- exactly where a reviewer should look first.
Verify: a reviewer knows where to start within thirty seconds of opening the PR description -- the same "where to look first" discipline Worked Scenarios 4 and 16 already demonstrated at smaller scale.
Step 3: Postmortem and C4 context diagram
exercises co-10, co-11, co-12, co-14, co-15
The postmortem records the original incident: a timestamped, actor-neutral timeline; the customer-facing impact; a systemic root cause; and owned follow-ups, one of which is exactly the RFC/ADR/PR trio above. The context diagram (C4 Level 1) shows the system boundary at the moment of the incident -- every actor and external system the postmortem's narrative names.
Verify: there is no blame language anywhere in the postmortem, every follow-up has a named owner and a concrete done-signal, and every node the context diagram shows is also named in the postmortem's or the context page's own prose, and vice versa.
Acceptance criteria
- Each document passes a reader-review rubric (skimmable, decision-first, no unexplained jargon) -- the same rubric Worked Scenario 25 walked through for ADR-0005 in the learning track.
- The postmortem is blameless (no individual named in connection with a mistake) and actionable (every follow-up has a named owner and a done-signal).
- The C4 context diagram matches the postmortem's prose exactly -- no actor or external system named in one is missing from the other.
- The RFC's Open Questions section keeps decided and undecided items visibly separate, and its "Accepted" status reflects an actual review cycle, not just a posting.
- The ADR records exactly one decision, carries a status field, and lives in the same repository path as the code it governs.
Done bar
This capstone produces the stated artifact set: an RFC and ADR a peer can act on without a meeting, a PR description that orients a reviewer in thirty seconds, and a blameless, actionable postmortem with a matching C4 context diagram -- five documents, one internally consistent story, combining every format this topic taught (BLUF and reader-first structure, co-01/co-02/co-03; the design-doc/RFC and ADR formats, co-04 through co-08; PR descriptions, co-09; blameless postmortems, co-10; C4 diagrams, co-11/co-12; precise RFC 2119 keywords and edited, register-matched prose, co-13/co-14/co-15; and proportional, close-to-code artifacts, co-16/co-17). Every technique used here traces to a primary source already cited in this topic's Accuracy notes (RFC 2119, S. Bradner, 1997, clarified by RFC 8174, 2017; Michael Nygard's 2011 ADR format; Simon Brown's C4 model); no new fact was needed to write this page or the five documents it links.
Next: Drilling →
Last updated July 15, 2026