HealthLog/.planning/v1422-backlog.md

v1.4.22 backlog

Items deferred from the W5 reconcile pass. Carried into v1.4.23 +.

Wave 5 reconcile carry-over

Code review

• Code MED-1 (src/app/api/insights/targets/route.ts:199-216): sparklinePoints() once-per-day bucket now uses Berlin tz (lifted to src/lib/analytics/berlin-day.ts in W5 reconcile). DST-boundary edge-case unit test added. analytics/route.ts now imports the shared helper. — APPLIED in W5 reconcile, kept here for trace.
• Code MED-4 (src/app/insights/page.tsx:1729-1747): IntersectionObserver picks the highest-intersection entry now. — APPLIED in W5 reconcile (folded into Design-H1 commit), kept here for trace.
• Code LOW-1 (src/components/insights/trends-row.tsx:148): <MoodChart title={…} /> self-resolves a localised title; the explicit prop tightens coupling. Drop the title prop on all three TrendsRow callsites OR add a comment. Cosmetic only.
• Code LOW-2 (src/lib/ai/coach/keyvalues.ts:53): parseKeyValueLine regex uses the first : as label/value boundary. A future prompt rev that adds a time-of-day pin would break ("Mon 6 May, 7:30 reading: 142/88"). Add a parser test for a colon-bearing label as a regression pin OR document the constraint in the prompt explicitly.
• Code LOW-3 (src/app/api/insights/chat/route.ts:284-291): sentinel parse failure logged with kept count but not the prompt-version-vs-malformed-cause breakdown. Extend SentinelParseResult with reason: "ok" | "no_close_marker" | "truncated" | "empty_block".
• Code LOW-4 (src/app/targets/page.tsx:272-273): Math.min(...points) / Math.max(...points) spread-call patterns can blow stack at ~10k arguments. Replace with for loop or points.reduce(...).
• Code LOW-5 (src/lib/query-keys.ts): ["user", "dashboardWidgets"] centralised in queryKeys.dashboardWidgets() in W5 reconcile, three call sites migrated. — APPLIED.

Security review

• Sec MED-1 (src/lib/auth/session.ts:32): hl_onboarding cookie SameSite=Lax → Strict. — APPLIED in W5 reconcile.
• Sec MED-2 (src/proxy.ts:28-33 + :152): /onboarding startsWith match tightened to exact + subroute. — APPLIED in W5 reconcile.
• Sec LOW-1 (/api/auth/me): GET that mutates state (cookie write) on every call. Idempotent; review left as-is, optional comment about not growing.
• Sec LOW-2 (Coach few-shots): synthetic numeric values baked into the system prompt. Optional WARNING comment so a future contributor doesn't paste real readings.
• Sec LOW-3 (src/app/api/insights/chat/route.ts:211-220): transcript builder flattens role labels into a single user prompt; user-self-injection only. v1.4.23 candidate: switch to structured chat-completion payload.
• Sec LOW-4 (src/lib/ai/coach/keyvalues.ts:147-180): truncated/closeMissing booleans not surfaced to wide-event meta. Extend with truncated / closeMissing for ops visibility.

Design review

• Design M1 (coach-panel/message-thread.tsx:259-307): coach evidence disclosure has no aria-controls / aria-expanded semantics; <summary> reports without expanded state in iOS Safari ≤ 17.3. Wrap the summary in a 36-px-min hit zone, add aria-expanded mirror via useState + onToggle.
• Design M2 (src/app/targets/page.tsx:264-303): Sparkline has no min-points threshold guard — 3-4 points spread over 30 days reads "lots of activity". Either require length >= 7 before painting, or render at stroke-opacity:0.5 for length < 10 plus a {n} of 30 days logged caption.
• Design M3 (src/components/insights/recommendation-card.tsx:328-335): severity badge uppercase eyebrow re-introduces the v1.4.20 M4 marketing pattern AND surfaces the internal severity enum (urgent / suggestion) — no i18n. Drop tracking-wide uppercase, replace {norm.severity} with {t(\insights.recommendation.severity.${norm.severity}`)}`. Add EN/DE keys.
• Design M4 (Coach disclaimer reach): pinned the disclaimer at the bottom of the message thread in W5 reconcile (Design-H3 fix). — APPLIED.
• Design M5 (src/app/insights/page.tsx:1070-1078): row-fill rule on the medication-correlation row uses showMoodSection heuristic. Either hide the BP-medication card entirely when bpMedicationScatterData.length < 5 (let mood-vs-BP take 100 %) or restore the fixed 50/50 with per-card empty state.
• Design M6 (dashboard-layout-section.tsx:204-241): comparison-toggle has no live-preview affordance. Add a tiny <TrendCard> preview swatch under the picker.
• Design M7 (src/app/targets/page.tsx:286-301): SVG sparkline aria-hidden skips the only "things changing" signal for SR users. Add aria-label to the <TrendIcon> parent OR move the sentiment into the delta caption prose.
• Design L1 (i18n voice asymmetry on coach.evidenceLabel): EN ("What I'm looking at") declarative; DE ("Worauf bezieht sich das?") curious. Decide on a unified voice.
• Design L2 (coach-panel/message-thread.tsx:286-304): React key={\${kv.label}-${idx}`}allows duplicate-label collisions. Use a hash of${label}-${window}-${value}`.
• Design L3 (targets/page.tsx:288): sparkline preserveAspectRatio="none" distorts y vs x. Either keep + accept, or fix Y-range to the target range.
• Design L4 (scroll-mt-28 overshoot): tuned to scroll-mt-16 in W5 reconcile (Design-H1 commit). — APPLIED.
• Design L5 (api-tokens table date wrap): cells wrap to two lines on 1024 px. Either short-format formatDate + tooltip, or push <col> width to 14 %.

Senior-dev review

• Sr H1 (createSession cookie fan-out): folded into createSession in W5 reconcile. — APPLIED.
• Sr H2 (chat route SSE construction duplicated 3×): extracted createSseStream helper in W5 reconcile. — APPLIED.
• Sr M1 (analytics route fetches ALL BP measurements ever, twice): bound by horizon (5 years) OR materialise a per-user bp_target_aggregate row refreshed on measurement-write.
• Sr M2 (targets route's 7-pass sparkline computation): collapse to a single Postgres-side GROUP BY date_trunc('day', measured_at), type.
• Sr M3 (targets page mutates server response inside render): wrap the glucose-target conversion in useMemo([data, t, displayGlucoseUnit]) to keep refs stable.
• Sr M4 (provenanceFromJson hand-rolls keyValues validation): replace 35 lines of manual typeof checks with coachKeyValueSchema.array().safeParse(parsed.keyValues). Same safeParse lift fixes the windows / metrics enum-membership gap.
• Sr M5 (sentinel parser conflates 3 distinct malformed conditions): four-state enum ("ok" | "truncated" | "unclosed" | "empty") so ops dashboards can distinguish the cases.
• Sr L1 (Sparkline preserveAspectRatio="none"): same as Design L3.
• Sr L2 (setOnboardingPendingCookie async but does no async work): could be sync if cookies() resolved by caller. v1.5 cleanup.
• Sr L3 (InsightsSectionNav re-creates IntersectionObserver on every mount): single mount per page, so the practical cost is negligible — observer should move with the component on extraction.
• Sr L4 (as TargetItem cast in targets/route.ts:281): the BP entry ships sys-only data without typing it. Pre-existing.

Simplify review

• S-01..S-05: applied in W5 reconcile (single commit). — APPLIED.
• S-M1..S-M5: maybe-tier — leave for now; revisit when adjacent work re-touches the surfaces.

Product Lead memo (cross-link)

• Insights page split held to v1.5 P5 (Apple Health work edits the same sub-trees).
• Pearson p-value normal approximation at low df: 30-LOC incomplete-beta or raise gate to df >= 20. Pull in v1.5 P0 before auto-discovery ships.
• Coach token-budget UX: 25 000 tokens/user/day = ~13 turns. iOS surface in v1.5 will multiply this. Replace generic 429 with localised "limit reached, resets at 00:00 UTC".
• Conversation persistence eviction policy: auto-archive after 90 days, hard-delete after 365. v1.5 P3 latest.
• OpenAPI spec drift CI guard: pull forward to v1.4.23 if possible.
• CoachDrawer key={prefill} weaponising React keys (Sr-HIGH-4): heavy refactor; v1.5 P3 latest.
• Cross-user feedback aggregation cron (P6 in product-lead memo): wired alongside v1.5 P3 once usage volume materialises.