fix(ai): stop the outbound risk guard from blocking this app's own scores

The named-risk-engine pattern in the outbound safety screen used
score2? (the 2 optional), so it matched bare "score" too — any mention
of one of this app's own computed, documented scores (Sleep Score,
Readiness Score, Health Score, Recovery Score, Stress Score, Strain
Score) tripped the bank meant only to catch a fabricated named
clinical-risk calculator (SCORE2, Framingham, ASCVD, QRISK).

Require the literal "SCORE2" instead. The other risk patterns —
percentage-shaped risk claims and the other three named engines — are
untouched, so a genuinely fabricated risk figure or a named
risk-calculator claim still blocks.

Closes #587.

src/lib/ai/coach/__tests__/outbound-guard.test.ts

@@ -67,6 +67,34 @@ describe("screenCoachReply — risk score", () => {
   });
 });
 
+// Issue #587 — a grounded Sleep Score explanation was replaced with the
+// generic clinical-risk refusal because the named-risk-engine pattern
+// matched bare "score", not just the SCORE2 risk calculator it was meant to
+// name.
+describe("screenCoachReply — issue #587 ordinary wellness-score wording", () => {
+  const allowed = [
+    "Your Sleep Score is based on sleep duration and stage balance — it's currently in a healthy range for you.",
+    "Your Readiness Score is lower than your recent baseline, likely from the shorter sleep window.",
+    "This Health Score is based on the available tracked pillars: BP, weight, mood, and compliance.",
+  ];
+  for (const reply of allowed) {
+    it(`allows: ${reply.slice(0, 40)}`, () => {
+      const d = screenCoachReply(reply, "en");
+      expect(d.block).toBe(false);
+      expect(d.reason).toBeNull();
+    });
+  }
+
+  it("still blocks a reply naming the SCORE2 risk engine", () => {
+    const d = screenCoachReply(
+      "Based on SCORE2, your cardiovascular risk profile looks elevated.",
+      "en",
+    );
+    expect(d.block).toBe(true);
+    expect(d.reason).toBe("risk_score");
+  });
+});
+
 describe("screenCoachReply — clean replies", () => {
   it("passes a grounded, non-prescriptive reply", () => {
     const reply =

src/lib/ai/safety/__tests__/outbound-screen.test.ts

@@ -107,6 +107,54 @@ describe("screenModelOutput — risk-score, all six locales", () => {
   }
 });
 
+// ── issue #587 — bare "score" is not a fabricated clinical risk score ─────
+
+/**
+ * The named-risk-engine pattern used `score2?` (the `2` optional), so it
+ * matched bare "score" too — any mention of one of THIS APP's own computed,
+ * documented scores (Sleep Score, Readiness Score, Health Score, …) tripped
+ * the fabricated-risk-engine bank meant for SCORE2/Framingham/ASCVD/QRISK.
+ */
+const WELLNESS_SCORE_CLEAN: readonly string[] = [
+  "Your Sleep Score is based on sleep duration and stage balance.",
+  "Your Readiness Score is lower than your recent baseline.",
+  "This Health Score is based on the available tracked pillars.",
+  "Your Recovery Score and Stress Score both moved this week.",
+  "The Strain Score reflects yesterday's training load.",
+];
+
+describe("screenModelOutput — issue #587 ordinary wellness-score wording", () => {
+  for (const text of WELLNESS_SCORE_CLEAN) {
+    it(`passes: ${text}`, () => {
+      const d = screenModelOutput(text, "en", CONVERSATIONAL_CONTRACTS);
+      expect(d.block).toBe(false);
+      expect(d.reason).toBeNull();
+    });
+  }
+
+  it("still blocks the literal named risk engine SCORE2", () => {
+    const d = screenModelOutput(
+      "Based on SCORE2, your cardiovascular risk profile looks elevated.",
+      "en",
+      CONVERSATIONAL_CONTRACTS,
+    );
+    expect(d.block).toBe(true);
+    expect(d.reason).toBe("risk_score");
+  });
+
+  it("still blocks Framingham, ASCVD, and QRISK by name", () => {
+    for (const engine of [
+      "your Framingham result was concerning",
+      "the ASCVD estimate points to elevated risk",
+      "your QRISK figure came back high",
+    ]) {
+      const d = screenModelOutput(engine, "en", CONVERSATIONAL_CONTRACTS);
+      expect(d.block).toBe(true);
+      expect(d.reason).toBe("risk_score");
+    }
+  });
+});
+
 // ── causal claims (GROUND RULE 12) ───────────────────────────────────────
 
 const CAUSAL_VIOLATIONS: Record<Locale, string> = {

src/lib/ai/safety/outbound-screen.ts

@@ -215,8 +215,14 @@ const RISK_PATTERNS: Record<Locale, readonly RegExp[]> = {
     /\b\d{1,3}\s*%\s+(?:risk|chance|probability|likelihood)\b/i,
     /\b(?:risk|chance|probability|likelihood)\s+(?:of|is|at)\s+(?:about\s+|roughly\s+|~)?\d{1,3}\s*%/i,
     /\b(?:10[- ]year|ten[- ]year|lifetime)\s+(?:cardiovascular|cardiac|heart|stroke|mortality|cvd|ascvd)\s+risk\b/i,
-    // Named risk engines are fabrications on any surface — the server runs none.
-    /\b(?:framingham|ascvd|score2?|qrisk)\b/i,
+    // Named risk engines are fabrications on any surface — the server runs
+    // none. `score2?` (the `2` optional) was the #587 bug: it matched bare
+    // "score" too, so every mention of this app's OWN computed scores
+    // (Sleep Score, Readiness Score, Health Score, …) tripped the named-
+    // engine bank. SCORE2 is a specific clinical risk calculator (like
+    // Framingham/ASCVD/QRISK) — only the literal name is a fabrication
+    // signal; the bare word "score" carries none on its own.
+    /\b(?:framingham|ascvd|score2|qrisk)\b/i,
   ],
   de: [
     /\brisiko\s+(?:von|bei|liegt\s+bei)\s+(?:etwa\s+|ungefähr\s+|~)?\d{1,3}\s*%/i,