HealthLog/.planning/phase-W2-report.md

Wave 2 — Insights surface polish (v1.4.22)

Started: 2026-05-10T20:40+02:00 Completed: 2026-05-10T21:00+02:00 Branch: develop (post-v1.4.21) Six atomic commits delivered to origin/develop.

A1 — BD-Zielbereich headline re-anchor

fix(analytics): BP target headline reads last-30d, all-time becomes sub-row

Up to v1.4.19 the headline pinned to windows.last30Days?.pct — a literal copy of the 30T sub-value, producing the 50/50/50 algorithmic pin Marc reported. v1.4.19 A1 routed the headline to allTime, which fixed the pin but introduced a new emotional problem: the headline became the slowest-moving aggregate possible, punishing recent improvement.

v1.4.22 A1 re-anchors the route's bpInTargetPct to windows.last30Days?.pct and adds a new bpInTargetPctAllTime field so the long-arc number stays accessible as a sub-value. The computeBpInTargetWindows() helper is unchanged — only the route's headline pick changed.

Files touched: src/app/api/analytics/route.ts, tests/integration/bp-in-target.test.ts (two new integration cases pinning the headline source + all-three-windows envelope shape).

A2 — BD-Kachel feature parity

feat(insights): BP tile gains trend arrow + comparison overlay parity

Up to v1.4.21 the BD-Zielbereich tile was the only one in the strip without a trend arrow, 7-day-trend chip, or comparison overlay caption. Synthesised a slope from the difference between the 7-day and 30-day in-target shares: 7d > 30d means recent improvement (up-good ⇒ green arrow). The 7-day-trend chip uses the same delta.

Extended <TrendCard> with optional avgAllTime / avgAllTimeLabelKey / avgAllTimeColorClass props so the all-time aggregate stays visible next to 7d: and 30d: after the A1 re-anchor. Other tiles leave the field undefined so the third sub-row never renders for them.

Files touched: src/components/charts/trend-card.tsx, src/components/charts/__tests__/trend-card-all-time.test.tsx (new), src/app/page.tsx, messages/{en,de}.json (new charts.avgAllTimeShort key).

A3 — Comparison-Toggle → global Settings

feat(settings): comparison-overlay becomes a global dashboard preference

Per feedback_settings_no_split.md, removed the on-surface <CompareToggle /> from /insights. The canonical picker already lives in Settings → Dashboard (since v1.4.16 phase B8), and the User.dashboardWidgetsJson.comparisonBaseline field already persists the value. Every chart on /insights still reads the resolved baseline so flipping the Settings toggle propagates atomically.

Files touched: src/app/insights/page.tsx (removed the meta-slot <CompareToggle /> from <DailyBriefing> and the matching import), src/app/__tests__/insights-polish.test.ts (new A3 describe-block asserting the toggle is gone but compareBaseline plumbing remains).

A4 — Insights layout normalisation

fix(insights): row-fill rule + equal-height trends + hide-on-insufficient-data

Three related layout polish wins:

• Row-fill rule: BP-medication grid, weight-correlation grid, and medications-per-day grid all hard-coded xl:grid-cols-2 / sm:grid-cols-2, leaving half-rows of empty space when one of the two slots wasn't populated. All three now switch to a single full-width column when fewer than 2 cards render.
• Trends row equal heights: each <TrendsRow> card is a flex column with min-h-[300px] so a multi-line AI annotation under one chart doesn't push that card taller than its neighbours.
• Hide insufficient correlations: <CorrelationRow> drops cards whose status !== "ok" instead of painting a placeholder. Zero ok cards hides the row entirely (no header, no disclaimer); 1 card spans 100 % width; 2-3 cards split 50/50 on >=md.

Files touched: src/components/insights/correlation-row.tsx, src/components/insights/trends-row.tsx, src/components/insights/__tests__/correlation-row.test.tsx, src/app/insights/page.tsx.

A5 — "Muster" rename + section tabs lift

feat(insights): rename "Muster" + lift section tabs above the hero

• DE rename decision: "Muster" → "Zusammenhänge". The W1b research suggested "Trends" or "Zusammenhänge"; picked the latter because the row directly above the correlation row already uses "Trends" for the trend-annotation strip. "Zusammenhänge" carries the relational meaning a correlation surface needs without borrowing terminology from its neighbour.
• EN parallel: "Patterns" → "Relationships" for the same reason.
• Tab lift: <InsightsSectionNav> moved from below the advisor card to the top of the page. Sticky scroll-anchored behaviour preserved (incremental — a full SPA-tab rewrite is out of scope for v1.4.22 polish). The user lands on Allgemein by default and sees the metric-tab strip before scrolling.

Files touched: messages/{en,de}.json, src/app/insights/page.tsx, src/components/insights/__tests__/correlation-row.test.tsx, src/components/insights/__tests__/correlation-card.test.tsx.

A6 — Token-leak fix + DE component labels

fix(insights): strip metric tokens in recommendation prose + DE component labels

W1a §3 traced the production "metric:WEIGHT" leak to src/components/insights/recommendation-card.tsx:336: every other AI-prose surface ran through stripChartTokens(), the recommendation text didn't. Two-line fix.

The bonus DE-locale defect: componentMood: "Mood" (English in the German file) plus the related componentBp: "BD" / componentCompliance: "Medis" abbreviations. Marc reported "Mut" — voice-to-text rendering of "Mood". All four labels now read natively German: "Blutdruck", "Gewicht", "Stimmung", "Einnahmetreue".

i18n integrity test added to pin the four DE component labels so a copy-paste regression can't reintroduce the leak.

Files touched: src/components/insights/recommendation-card.tsx, src/components/insights/__tests__/recommendation-card.test.tsx, messages/de.json, src/lib/__tests__/i18n-locale-integrity.test.ts.

Test delta

Layer	Before	After
Unit (vitest)	2036 tests / 238 files	2049 tests / 239 files
Integration	(covered separately)	+2 cases in bp-in-target.test.ts

13 new unit cases + 2 new integration cases. All green.

Decisions logged

• DE rename: "Zusammenhänge" over "Trends" — relational meaning, doesn't collide with the Trends-row title directly above.
• Section tabs: kept the scroll-anchored sticky strip and lifted it above the hero. A full SPA-tab rewrite is deferred (v1.5+).
• All-time sub-value: rendered as a third <span> inside the existing flex gap-3 sub-row so on mobile it wraps naturally below 7d: and 30d: instead of crowding into a single line.
• Synthetic slope: BP-tile uses slope = (pct7 - pct30) / 30 as the slope-per-day signal. Confidence pinned to 1 because the source is a deterministic difference, not a regression fit.