ADR-0038: continuous owner feedback loop and a standing design authority
- Status: Accepted (owner sign-off is the merge of the PR that carries this ADR, 2026-08-05)
- Date: 2026-08-05
- Amends: ADR-0026 (the acceptance practice in step 4 and the review-role roster in step 3; the batch mechanics, decision gates, and bookkeeping close-out stand unchanged)
Context
ADR-0026 step 4 assumes the owner reviews behavior per epic batch — effectively a UAT walkthrough at each acceptance. Practice after roughly three sprints: that walkthrough happened once, partially. This is not a discipline failure to be exhorted away; it is the same structural fact ADR-0026 and ADR-0036 already recognized — one owner, whose attention is the scarcest resource in the project. A process step that requires a scheduled block of owner time per batch will keep not happening.
Meanwhile the opposite channel works without being designed: day-to-day use of
the live instance produces a steady stream of concrete observations (loading
affordances, visual drift, ambiguous warnings, missing view options). Until
2026-08-05 these had no defined intake — the owner’s choice was to interrupt
with unstructured notes or to sit on them until a review session that rarely
comes. The first triaged dump
(_bmad-output/planning-artifacts/feedback-triage-2026-08-05.md) showed the
mode works: of the observations reported, several were already tracked
(#560, #568, #572) — pipeline dedup is exactly the work an agent should be
doing, not the owner.
The same dump confirmed a second structural gap. The project has a design spec from one early session (DESIGN.md + EXPERIENCE.md, 2026-06-12, tracked in #356) but no standing role that holds a consistent design language while features ship. The owner’s expectation of the UX role — that it critiques aesthetics, consistency, and UX simplicity on an ongoing basis — was never wired into the workflow. The result is visible drift: text-only tabs next to an icon-based menu, bare text-field date pickers, explanatory prose dumped under charts in violation of the project’s own UX-DR11 tooltip rule, inconsistent loading placeholders. Every one of these passed review, because no review role was looking.
Decision
- Feedback intake is continuous, and the agent carries the structuring
cost. The owner drops observations at any time, in any form, unsorted.
The PM agent triages each dump into a dated
planning-artifacts/feedback-triage-YYYY-MM-DD.md: cluster, check the pipeline for existing coverage, route to epics/ADRs/decision gates, and propose next steps. The owner reviews the triage — minutes, not hours — and confirms or corrects routing. Only then are thin issues created per the issue-tracking convention. Design-affecting items go through a design pass before implementation stories are cut. - Owner acceptance stays at the merge, unchanged. The behavior review against the reviewer briefing (ADR-0026 step 4) remains the acceptance act. What this ADR removes is the fiction of an additional per-epic owner UAT session; systematic screen-by-screen evaluation happens when the owner chooses to do it, and its findings enter through the same feedback intake.
- The UX designer role becomes the standing design authority. It owns a living design-language spec (starting by refreshing and promoting the 2026-06-12 DESIGN.md + EXPERIENCE.md), and design decisions route through it the way architecture decisions route through ADRs.
- The agentic review closing act gains a design-critic role. Alongside the correctness hunter, edge-case hunter, and UAT persona (ADR-0026 step 3), every batch with user-visible surface is reviewed against the design-language spec for visual consistency, aesthetic quality, and UX simplicity. Design drift becomes a review finding on the branch, not an owner discovery months later.
Consequences
- No observation waits for a review slot, and no owner session is spent on transcription or deduplication.
- The owner’s recurring costs shrink to two small, well-defined acts: confirming a triage and accepting a batch.
- Design consistency gets an enforcement point with teeth (a review role and a spec to hold work against) instead of an aspiration.
- On acceptance of this ADR, AGENTS.md’s Epic-Batch Workflow section is updated in the same PR: step 3’s role roster gains the design critic, and step 4 drops the implied per-epic owner walkthrough.
- Risk: triage documents accumulate as paperwork. Mitigation: they stay
short, live in
planning-artifacts/, and items the owner does not confirm within a reasonable time are dropped rather than carried forward — the live instance will resurface anything that still hurts.