fix(insights): stripChartTokens covers capitalised Metric + orphan enum patterns + Apple Health additives

Marc reported a raw "Metric Pressure_Sys" token leak on /insights
2026-05-14. The v1.4.22 stripper only matched the canonical
`metric:<TYPE>` colon form; the model emitted a capitalised "Metric"
prefix followed by a Pascal/snake-cased identifier and the substring
sailed through to the rendered prose unchanged.

Extend `stripChartTokens()` with two additional matchers:

- `Metric <Word>` capitalised two-word phrase (catches the exact
  pattern Marc saw, plus PascalCase variants and the v1.4.23
  Apple-Health additions like `HEART_RATE_VARIABILITY`).
- Fixed-list orphan upper-snake-case enum identifiers
  (`BLOOD_PRESSURE_SYS`, `PULSE_BPM`, `MOOD_SCORE`,
  `MEDICATION_COMPLIANCE_PCT`, plus the v1.4.23 Apple Health enum
  surface: `HEART_RATE_VARIABILITY`, `RESTING_HEART_RATE`,
  `ACTIVE_ENERGY_BURNED`, `FLIGHTS_CLIMBED`,
  `WALKING_RUNNING_DISTANCE`, `VO2_MAX`, `BODY_TEMPERATURE`,
  `SLEEP_DURATION`).

The orphan-enum list is fixed (not a generic `[A-Z_]+`) so single
capitalised words like `WEIGHT` or `MOOD` stay legible in normal
prose — only the suffixed/database-style enum names get cleaved.
Lowercase "metric" inside legitimate English prose ("Each metric
tells part of the story") is also left untouched.

Tests: 14 new cases covering every documented leak pattern plus two
defence-in-depth assertions (combined leaks; German "Metrik" safe).

src/lib/insights/__tests__/chart-tokens.test.ts

@@ -70,6 +70,124 @@ describe("stripChartTokens", () => {
   });
 });
 
+// v1.4.25 W5b — Marc reported the raw "Metric Pressure_Sys" leak on
+// /insights 2026-05-14. The v1.4.22 W1a regex only caught the canonical
+// `metric:<TYPE>` colon form; the model emitted a capitalised "Metric"
+// word followed by a Pascal/enum-cased identifier. The fix extends the
+// stripper with two additional matchers: (1) `Metric <Word>` capitalised
+// phrase, (2) bare upper-snake-case enum identifiers from the
+// MeasurementType list (including the v1.4.23 Apple-Health additions).
+describe("stripChartTokens — v1.4.25 W5b capitalised + orphan-enum leaks", () => {
+  it("strips the capitalised `Metric Pressure_Sys` phrase Marc reported", () => {
+    expect(
+      stripChartTokens("Your Metric Pressure_Sys is sitting above 140."),
+    ).toBe("Your is sitting above 140.");
+  });
+
+  it("strips the capitalised form with snake_case enum body", () => {
+    expect(
+      stripChartTokens("trend points up for Metric BLOOD_PRESSURE_SYS today"),
+    ).toBe("trend points up for today");
+  });
+
+  it("strips orphan enum mentions in prose — BLOOD_PRESSURE_SYS", () => {
+    expect(
+      stripChartTokens("Your BLOOD_PRESSURE_SYS reading is elevated."),
+    ).toBe("Your reading is elevated.");
+  });
+
+  it("strips orphan enum mentions — WEIGHT bare upper-snake-case is left alone", () => {
+    // The bare word `WEIGHT` is shared with the user-facing metric
+    // label, so the orphan-enum stripper deliberately allowlists it
+    // against the user-facing label. Only suffixed enums (e.g.
+    // BLOOD_PRESSURE_SYS) get cleaved.
+    expect(stripChartTokens("Your WEIGHT trend is stable.")).toBe(
+      "Your WEIGHT trend is stable.",
+    );
+  });
+
+  it("strips orphan enum mentions — PULSE_BPM", () => {
+    expect(stripChartTokens("PULSE_BPM stays within target.")).toBe(
+      "stays within target.",
+    );
+  });
+
+  it("strips orphan enum mentions — MOOD_SCORE", () => {
+    expect(stripChartTokens("MOOD_SCORE drifted down this week.")).toBe(
+      "drifted down this week.",
+    );
+  });
+
+  it("strips orphan enum mentions — MEDICATION_COMPLIANCE_PCT", () => {
+    expect(
+      stripChartTokens("MEDICATION_COMPLIANCE_PCT held above 90."),
+    ).toBe("held above 90.");
+  });
+
+  it("strips v1.4.23 Apple Health additions — HEART_RATE_VARIABILITY", () => {
+    expect(
+      stripChartTokens("Last week HEART_RATE_VARIABILITY trended up."),
+    ).toBe("Last week trended up.");
+  });
+
+  it("strips v1.4.23 Apple Health additions — RESTING_HEART_RATE", () => {
+    expect(
+      stripChartTokens("Your RESTING_HEART_RATE is stable around 58 bpm."),
+    ).toBe("Your is stable around 58 bpm.");
+  });
+
+  it("strips v1.4.23 Apple Health additions — ACTIVE_ENERGY_BURNED", () => {
+    expect(
+      stripChartTokens("ACTIVE_ENERGY_BURNED has been below target."),
+    ).toBe("has been below target.");
+  });
+
+  it("strips v1.4.23 Apple Health additions — FLIGHTS_CLIMBED, WALKING_RUNNING_DISTANCE, VO2_MAX, BODY_TEMPERATURE, SLEEP_DURATION", () => {
+    expect(stripChartTokens("FLIGHTS_CLIMBED at 5/day.")).toBe("at 5/day.");
+    expect(
+      stripChartTokens("WALKING_RUNNING_DISTANCE averaged 4 km."),
+    ).toBe("averaged 4 km.");
+    expect(stripChartTokens("VO2_MAX has crept up.")).toBe("has crept up.");
+    expect(
+      stripChartTokens("BODY_TEMPERATURE sat at 36.8 most days."),
+    ).toBe("sat at 36.8 most days.");
+    expect(stripChartTokens("SLEEP_DURATION averaged 7.4h.")).toBe(
+      "averaged 7.4h.",
+    );
+  });
+
+  it("strips the Metric phrase + the enum together (defence in depth)", () => {
+    // The model can emit both forms in the same sentence — the
+    // capitalised phrase AND the bare enum identifier. Both pattern
+    // matchers run independently so each cleaves its own substring.
+    expect(
+      stripChartTokens(
+        "Metric BLOOD_PRESSURE_SYS shows your BLOOD_PRESSURE_SYS is elevated.",
+      ),
+    ).toBe("shows your is elevated.");
+  });
+
+  it("does not strip ordinary prose that incidentally contains 'metric' as a word", () => {
+    // Lower-case word "metric" without the colon form must not be
+    // touched — it's a perfectly legitimate English word inside copy
+    // like "Each metric in the snapshot ...". Marc's directive
+    // explicitly targets the leak patterns, not legitimate prose.
+    expect(
+      stripChartTokens("Each metric tells part of the story."),
+    ).toBe("Each metric tells part of the story.");
+  });
+
+  it("does not strip the canonical Metrik token in German prose", () => {
+    // German prose uses "Metrik" (capitalised noun) freely; the
+    // English-targeted `Metric <Word>` matcher must not bleed into
+    // German copy. The regex is anchored to the literal English
+    // word "Metric" + whitespace + identifier so "Metrik" is safe.
+    expect(stripChartTokens("Jede Metrik liefert ein Stück.")).toBe(
+      "Jede Metrik liefert ein Stück.",
+    );
+  });
+});
+
 describe("parseChartTokens — defensive (v1.4.17)", () => {
   it("returns empty array for undefined input", () => {
     expect(parseChartTokens(undefined)).toEqual([]);

src/lib/insights/chart-tokens.ts

@@ -64,6 +64,61 @@ const ALLOWED_SET = new Set<string>(ALLOWED_CHART_TOKENS);
 const STRIP_TOKEN_REGEX = /metric:[A-Za-z0-9_]+/g;
 const PARSE_TOKEN_REGEX = /metric:[A-Z_]+/g;
 
+// v1.4.25 W5b — capitalised-Metric form. Marc reported the leak
+// pattern `Metric Pressure_Sys` on /insights 2026-05-14: the model
+// emitted the metric reference as a two-word phrase ("Metric" + space
+// + enum-cased identifier) instead of the canonical `metric:<TYPE>`
+// colon form. The colon-form regex above never matched this so the
+// raw enum surfaced in the rendered prose.
+//
+// Match the capitalised "Metric" prefix followed by a space and one
+// of: an upper-snake-case enum (`BLOOD_PRESSURE_SYS`) or a
+// PascalCase-with-underscore form (`Pressure_Sys`). The character
+// class allows mixed-case + digits + underscores so v1.4.23
+// Apple-Health additions (`HEART_RATE_VARIABILITY`,
+// `WALKING_RUNNING_DISTANCE`, etc.) get cleaved cleanly too. Word
+// boundary `\b` on both ends prevents accidental matches inside
+// ordinary prose ("Metricated", "submetric_x").
+const METRIC_WORD_REGEX = /\bMetric\s+[A-Za-z][A-Za-z0-9_]*\b/g;
+
+// v1.4.25 W5b — orphan enum-identifier leaks. The model sometimes
+// drops the bare MeasurementType enum name straight into prose
+// ("Your BLOOD_PRESSURE_SYS is elevated"). This was harmless when
+// the only enums were short capitalised words but leaks badly with
+// the v1.4.23 Apple-Health additions because the resulting prose
+// reads as a database field name. Match the full enum list as a
+// fixed alternation so we never accidentally cleave a real word
+// (e.g. "MOOD") out of legitimate prose unless the model wrote it
+// as the raw enum in upper-snake form.
+const ORPHAN_ENUMS = [
+  "BLOOD_PRESSURE_SYS",
+  "BLOOD_PRESSURE_DIA",
+  "PULSE_BPM",
+  "MOOD_SCORE",
+  "MEDICATION_COMPLIANCE_PCT",
+  // v1.4.23 Apple Health additions — see ALLOWED_CHART_TOKENS above.
+  "HEART_RATE_VARIABILITY",
+  "RESTING_HEART_RATE",
+  "ACTIVE_ENERGY_BURNED",
+  "FLIGHTS_CLIMBED",
+  "WALKING_RUNNING_DISTANCE",
+  "VO2_MAX",
+  "BODY_TEMPERATURE",
+  "SLEEP_DURATION",
+] as const;
+
+// `\b` boundaries keep ordinary English prose untouched — "weight"
+// (lowercase) or "BMI" (no underscore) never match. The fixed list
+// also means we never accidentally strip a single capitalised word
+// like "MOOD" or "WEIGHT" that has legitimate use as an in-prose
+// metric label in display copy. Marc's directive (2026-05-14) is to
+// strip ONLY upper-snake / suffixed enum identifiers, not the
+// canonical user-facing labels.
+const ORPHAN_ENUM_REGEX = new RegExp(
+  `\\b(?:${ORPHAN_ENUMS.join("|")})\\b`,
+  "g",
+);
+
 /**
  * Find every well-formed chart token in `text` and return only those that
  * are in the allowlist. Hallucinated tokens (e.g. `metric:NUKE`) are
@@ -98,6 +153,9 @@ export function stripChartTokens(text: string | null | undefined): string {
   if (typeof text !== "string") return "";
   return text
     .replace(STRIP_TOKEN_REGEX, "")
+    .replace(METRIC_WORD_REGEX, "")
+    .replace(ORPHAN_ENUM_REGEX, "")
+    .replace(/\s+([.,;:!?])/g, "$1")
     .replace(/\s{2,}/g, " ")
     .trim();
 }