fix(insights): align BD-Zielbereich delta math with comparison window

The tile computed `bp30 - bpAll` regardless of which baseline the
user picked under Settings → Dashboard, while rendering a caption
keyed on `comparisonBaseline` ("vs. last month" / "vs. last year").
The math therefore lied to the user. Every other trend tile uses
`tileCompareDelta()` against period-aligned `avg30LastMonth` /
`avg30LastYear` baselines.

Add `priorMonth` (now-60d…now-30d) and `priorYear` (now-395d…
now-365d) windows to `computeBpInTargetWindows()`. Ship the pcts on
the analytics envelope as `bpInTargetPctPriorMonth` /
`bpInTargetPctPriorYear`. Route the dashboard's `bpCompareDelta` to
subtract the prior window matching the user's baseline; fall back
to null for `none`.

Three new bp-in-target unit tests (priorMonth, priorYear, null
windows) and two insights-polish guards (envelope contains the
fields, page no longer subtracts `bpAll`).

Closes Code-H2.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/app/__tests__/insights-polish.test.ts

@@ -181,6 +181,40 @@ describe("v1.4.19 A3 — chart-token leak hardening", () => {
   });
 });
 
+describe("v1.4.22 W5 reconcile (Code-H2) — BD-Zielbereich tile compareDelta uses period-aligned baseline", () => {
+  /**
+   * Up to v1.4.22 W4 the BD tile computed `bp30 - bpAll`
+   * (last-30-days minus all-time) regardless of which window the user
+   * picked under Settings → Dashboard. The caption still rendered "vs.
+   * last month" / "vs. last year" via `comparison.captionLastMonth` /
+   * `comparison.captionLastYear`, so the user read a sentence whose
+   * numerator was actually a 30d-vs-all-time delta. Every other tile
+   * routes through `tileCompareDelta()` with `summary.avg30LastMonth`
+   * / `summary.avg30LastYear` for honest period-aligned math.
+   *
+   * This test pins that the dashboard now subtracts the prior-period
+   * window (priorMonth / priorYear) instead of all-time.
+   */
+  it("subtracts bpInTargetPctPriorMonth when comparisonBaseline=lastMonth", () => {
+    const src = load(DASHBOARD_PATH);
+    // The fix introduces `bpComparePrior` keyed by `compareBaseline`.
+    expect(src).toContain("bpInTargetPctPriorMonth");
+    expect(src).toContain("bpInTargetPctPriorYear");
+    // The all-time shortcut from v1.4.22 A2 must not subtract `bpAll`
+    // for the comparison delta any more.
+    expect(src).not.toMatch(/bp30\s*-\s*bpAll/);
+  });
+
+  it("ships priorMonth + priorYear pcts in the analytics envelope", () => {
+    const ROUTE_SRC = readFileSync(
+      join(ROOT, "src/app/api/analytics/route.ts"),
+      "utf8",
+    );
+    expect(ROUTE_SRC).toContain("bpInTargetPctPriorMonth");
+    expect(ROUTE_SRC).toContain("bpInTargetPctPriorYear");
+  });
+});
+
 describe("v1.4.22 A3 — comparison toggle is global Settings only", () => {
   /**
    * The comparison-overlay toggle is a global preference. Up to v1.4.21

src/app/api/analytics/route.ts

@@ -70,6 +70,16 @@ export const GET = apiHandler(async () => {
   let bpInTargetPct7d: number | null = null;
   let bpInTargetPct30d: number | null = null;
   let bpInTargetPctAllTime: number | null = null;
+  /**
+   * v1.4.22 W5 reconcile (Code-H2) — period-aligned prior-window
+   * pcts so the BD-Zielbereich tile's comparison-overlay caption
+   * stops mismatching its math with its label. The tile's
+   * `compareDelta` is `last30Days - priorMonth` (or `… - priorYear`)
+   * matching the user's `comparisonBaseline` selection, never
+   * `last30Days - allTime` (the v1.4.22 A2 shortcut).
+   */
+  let bpInTargetPctPriorMonth: number | null = null;
+  let bpInTargetPctPriorYear: number | null = null;
   const bpTargets = getBpTargets(user.dateOfBirth);
   if (bpTargets) {
     const now = new Date();
@@ -99,6 +109,8 @@ export const GET = apiHandler(async () => {
     bpInTargetPct7d = windows.last7Days?.pct ?? null;
     bpInTargetPct30d = windows.last30Days?.pct ?? null;
     bpInTargetPctAllTime = windows.allTime?.pct ?? null;
+    bpInTargetPctPriorMonth = windows.priorMonth?.pct ?? null;
+    bpInTargetPctPriorYear = windows.priorYear?.pct ?? null;
   }
 
   // Per-context glucose summaries (canonical mg/dL).
@@ -153,6 +165,8 @@ export const GET = apiHandler(async () => {
     bpInTargetPct7d,
     bpInTargetPct30d,
     bpInTargetPctAllTime,
+    bpInTargetPctPriorMonth,
+    bpInTargetPctPriorYear,
     glucoseByContext,
     correlations,
     healthScore,

src/app/page.tsx

@@ -103,6 +103,15 @@ interface AnalyticsData {
    * on the BD-Zielbereich tile (alongside `7d` and `30d`).
    */
   bpInTargetPctAllTime?: number | null;
+  /**
+   * v1.4.22 W5 reconcile (Code-H2) — period-aligned prior-window
+   * pcts. The BD-Zielbereich tile's comparison-overlay caption picks
+   * `priorMonth` for `comparisonBaseline === "lastMonth"` and
+   * `priorYear` for `lastYear` so the rendered "Δ X% vs. last month"
+   * stays honest. Null when the prior window has no paired readings.
+   */
+  bpInTargetPctPriorMonth?: number | null;
+  bpInTargetPctPriorYear?: number | null;
   glucoseByContext?: Record<string, DataSummaryType>;
 }
 
@@ -817,6 +826,8 @@ export default function DashboardPage() {
           const bp7 = data?.bpInTargetPct7d ?? null;
           const bp30 = data?.bpInTargetPct30d ?? null;
           const bpAll = data?.bpInTargetPctAllTime ?? null;
+          const bpPriorMonth = data?.bpInTargetPctPriorMonth ?? null;
+          const bpPriorYear = data?.bpInTargetPctPriorYear ?? null;
           const bpTrendDelta =
             bp7 !== null && bp30 !== null ? bp7 - bp30 : null;
           const bpSlope30: import("@/lib/analytics/trends").TrendSlope | null =
@@ -835,10 +846,26 @@ export default function DashboardPage() {
                   // mis-interpret a `0` as "very low confidence".
                   confidence: 1,
                 };
+          // v1.4.22 W5 reconcile (Code-H2) — match the comparison
+          // window the user picked. `lastMonth` baseline ⇒ compare
+          // last30 against the now-60d…now-30d window (priorMonth);
+          // `lastYear` baseline ⇒ compare against the matching window
+          // shifted back 365 days. The previous shortcut subtracted
+          // `bpAll` regardless of baseline, which produced a number
+          // whose magnitude was honest but whose label ("vs. last
+          // month") lied to the user.
+          const bpComparePrior =
+            compareBaseline === "lastMonth"
+              ? bpPriorMonth
+              : compareBaseline === "lastYear"
+                ? bpPriorYear
+                : null;
           const bpCompareDelta =
-            compareBaseline === "none" || bp30 === null || bpAll === null
+            compareBaseline === "none" ||
+            bp30 === null ||
+            bpComparePrior === null
               ? null
-              : Math.round((bp30 - bpAll) * 10) / 10;
+              : Math.round((bp30 - bpComparePrior) * 10) / 10;
           trendCards.push({
             id: "bpInTarget",
             order: widgetOrder("bpInTarget"),