feat(coach-prompt): GROUND RULE 13 (insights) + 8 (coach) — no internal metric enum names in prose

Closes the loop on the "Metric Pressure_Sys" leak Marc reported on
the Einschätzungen surface 2026-05-14. The stripChartTokens regex
extension cleans up render-time output, but the model was free to
keep emitting raw enum identifiers turn after turn. This commit
teaches the generator to stop emitting them in the first place.

Insight-generator (DE + EN) gets GROUND RULE 13:

  - Lists every banned identifier verbatim — Pressure_Sys,
    BLOOD_PRESSURE_SYS, PULSE_BPM, MOOD_SCORE,
    MEDICATION_COMPLIANCE_PCT, plus the v1.4.23 Apple Health surface
    (HEART_RATE_VARIABILITY, RESTING_HEART_RATE,
    ACTIVE_ENERGY_BURNED, FLIGHTS_CLIMBED,
    WALKING_RUNNING_DISTANCE, VO2_MAX, BODY_TEMPERATURE,
    SLEEP_DURATION).
  - Enumerates the exact user-facing fields the ban applies to —
    summary, recommendations[].text, findings, dailyBriefing,
    trendAnnotations.*, weeklyReport.*, storyboardAnnotations.
  - Carves out the contract identifiers the parser reads
    (metricSource.type, sourceMetric, trendAnnotations keys) so the
    JSON contract stays intact — the ban applies ONLY to prose.
  - Documents the natural-language labels to use instead ("your
    systolic", "your weight", "your medication adherence", …).

Coach system-prompt (DE + EN) gets the matching GROUND RULE 8 with
the same banned list, the same prose-only carve-out, and an extra
clarification that evidence-block labels stay human-readable
("avg30 systolic", NOT "BLOOD_PRESSURE_SYS_AVG_30") while the
window slot keeps its documented contract vocabulary.

PROMPT_VERSION 4.23.0 → 4.24.0. The rule is genuinely new (neither
prompt previously mentioned the enum identifiers in any form), so
the bump reflects the additive behavioural change. Existing pin
assertions in coach-prompt-v423.test.ts + coach/system-prompt.test.ts
loosen from `/^4\.23\.\d+$/` to `4.{>=23}.x` so future additive
ratchets don't force a hostile test rewrite each time.

New unit tests pin GROUND RULE 13 + 8 by the banned-identifier
strings + the contract-carve-out wording so a future copy-paste
regression can't silently drop either rule.

src/lib/ai/__tests__/coach-prompt-v423.test.ts

@@ -19,8 +19,14 @@ import {
  *      so the model emits the additive enum values directly.
  */
 describe("PROMPT_VERSION (v1.4.23 Apple Health ratchet)", () => {
-  it("matches the 4.23.x train", () => {
-    expect(PROMPT_VERSION).toMatch(/^4\.23\.\d+$/);
+  it("matches the 4.23+ train", () => {
+    // v1.4.25 W5b ratchet to 4.24.0 — GROUND RULE 13 bans internal
+    // metric enum identifiers from prose. The pin loosens to accept
+    // any 4.{>=23}.x revision so future additive ratchets don't
+    // force a hostile test rewrite each time the rules grow.
+    const [, minor] = PROMPT_VERSION.split(".").map(Number);
+    expect(PROMPT_VERSION).toMatch(/^4\.\d+\.\d+$/);
+    expect(minor).toBeGreaterThanOrEqual(23);
   });
 });
 
@@ -77,3 +83,49 @@ describe("getStrictInsightsSystemPrompt — DE", () => {
     );
   });
 });
+
+/**
+ * v1.4.25 W5b — pin GROUND RULE 13 (no internal metric identifiers
+ * in prose). Marc saw the "Metric Pressure_Sys" leak on the
+ * Einschätzungen surface 2026-05-14; the stripper covers the leak at
+ * render time, GROUND RULE 13 closes the loop at generation time so
+ * the cleaned-up prose was the only version emitted in the first
+ * place.
+ */
+describe("getStrictInsightsSystemPrompt — GROUND RULE 13 (no enum names in prose)", () => {
+  it("EN prompt carries the rule and enumerates the banned identifiers", () => {
+    const prompt = getStrictInsightsSystemPrompt("en");
+    expect(prompt).toMatch(/13\. v1\.4\.25/);
+    expect(prompt).toMatch(/Internal metric identifiers/);
+    // The banned-list is enumerated verbatim so the model has no
+    // ambiguity about which strings are off-limits in prose.
+    expect(prompt).toMatch(/"Pressure_Sys"/);
+    expect(prompt).toMatch(/"BLOOD_PRESSURE_SYS"/);
+    expect(prompt).toMatch(/"PULSE_BPM"/);
+    expect(prompt).toMatch(/"MOOD_SCORE"/);
+    expect(prompt).toMatch(/"MEDICATION_COMPLIANCE_PCT"/);
+    // Apple Health additions also banned in prose (the surface
+    // ships in v1.5 but the model can mention them on iOS-connected
+    // snapshots today).
+    expect(prompt).toMatch(/"HEART_RATE_VARIABILITY"/);
+    expect(prompt).toMatch(/"SLEEP_DURATION"/);
+    // The carve-out for contract identifiers must remain in place
+    // so the parser keeps working.
+    expect(prompt).toMatch(/metricSource\.type/);
+    expect(prompt).toMatch(/applies ONLY to prose/);
+  });
+
+  it("DE prompt carries the rule with the same banned list", () => {
+    const prompt = getStrictInsightsSystemPrompt("de");
+    expect(prompt).toMatch(/13\. v1\.4\.25/);
+    expect(prompt).toMatch(/Interne Metrik-Identifier/);
+    expect(prompt).toMatch(/"Pressure_Sys"/);
+    expect(prompt).toMatch(/"BLOOD_PRESSURE_SYS"/);
+    expect(prompt).toMatch(/"PULSE_BPM"/);
+    expect(prompt).toMatch(/"MOOD_SCORE"/);
+    expect(prompt).toMatch(/"MEDICATION_COMPLIANCE_PCT"/);
+    expect(prompt).toMatch(/"HEART_RATE_VARIABILITY"/);
+    expect(prompt).toMatch(/"SLEEP_DURATION"/);
+    expect(prompt).toMatch(/AUSSCHLIEßLICH für Fließtext/);
+  });
+});

src/lib/ai/coach/__tests__/system-prompt.test.ts

@@ -16,11 +16,15 @@ import { getCoachSystemPrompt } from "../system-prompt";
  *      the prompt body (per the W1b research output).
  */
 describe("PROMPT_VERSION (v1.4.23 Coach + Apple Health extension)", () => {
-  it("matches the 4.23.x train", () => {
+  it("matches the 4.23+ train", () => {
     // v1.4.23 ratcheted the prompt to add GROUND RULE 12 (Apple Health
-    // optional categories). Future patches stay on 4.23.x until a
-    // behavioural change ships.
-    expect(PROMPT_VERSION).toMatch(/^4\.23\.\d+$/);
+    // optional categories). v1.4.25 W5b ratcheted again to add GROUND
+    // RULE 13 (no internal metric identifiers in prose). The pin
+    // loosens to accept any 4.{>=23}.x revision so future additive
+    // ratchets don't force a hostile test rewrite each time.
+    const [, minor] = PROMPT_VERSION.split(".").map(Number);
+    expect(PROMPT_VERSION).toMatch(/^4\.\d+\.\d+$/);
+    expect(minor).toBeGreaterThanOrEqual(23);
   });
 });
 
@@ -50,6 +54,10 @@ describe("getCoachSystemPrompt — EN", () => {
     expect(prompt).toMatch(/5\. Motivational-interviewing micro-moves\./);
     expect(prompt).toMatch(/6\. Redirect off-topic input gracefully\./);
     expect(prompt).toMatch(/7\. Ground every number in the SNAPSHOT\./);
+    // v1.4.25 W5b — GROUND RULE 8 added on the Coach side: bans the
+    // raw enum identifiers from the user-facing reply text and from
+    // the evidence-block labels.
+    expect(prompt).toMatch(/8\. Internal metric identifiers/);
   });
 
   it("preserves the v1.4.21 DAY-LEVEL READINGS section verbatim", () => {
@@ -115,6 +123,8 @@ describe("getCoachSystemPrompt — DE", () => {
     );
     expect(prompt).toMatch(/6\. Off-topic-Eingaben elegant umlenken\./);
     expect(prompt).toMatch(/7\. Verankere jede Zahl im SNAPSHOT\./);
+    // v1.4.25 W5b — GROUND RULE 8 on the German Coach side.
+    expect(prompt).toMatch(/8\. Interne Metrik-Identifier/);
   });
 
   it("preserves the TAGES-LEVEL-MESSWERTE section verbatim", () => {

src/lib/ai/coach/system-prompt.ts

@@ -76,6 +76,26 @@ GROUND RULES
    user prompt carries. Do not extrapolate. Do not reference "people
    like you". Do not compute risk scores.
 
+8. Internal metric identifiers stay OUT of your prose (v1.4.25).
+   Never write database / enum-style names like "Pressure_Sys",
+   "BLOOD_PRESSURE_SYS", "PULSE_BPM", "MOOD_SCORE",
+   "MEDICATION_COMPLIANCE_PCT", "HEART_RATE_VARIABILITY",
+   "RESTING_HEART_RATE", "ACTIVE_ENERGY_BURNED", "FLIGHTS_CLIMBED",
+   "WALKING_RUNNING_DISTANCE", "VO2_MAX", "BODY_TEMPERATURE", or
+   "SLEEP_DURATION" in your reply text — neither in the prose nor in
+   the evidence-block labels. Reference each metric with the
+   natural-language label the user sees in the app — "your
+   systolic", "your weight", "your pulse", "your mood", "your
+   medication adherence", "your resting heart rate", "your sleep
+   duration", "your steps". Likewise never write the canonical
+   "metric:<TYPE>" chart-token string anywhere in your reply; the
+   surrounding /insights surface owns inline-chart wiring, not the
+   Coach. Evidence-block labels stay short and human ("avg30
+   systolic", not "BLOOD_PRESSURE_SYS_AVG_30") and the "window"
+   slot keeps its documented vocabulary (last7days / last30days /
+   last90days / allTime) — the ban applies to the metric label
+   itself, not to the contract-level "window" token.
+
 DAY-LEVEL READINGS — USE THE TIMELINE
 
 Each metric in the SNAPSHOT carries a "timeline" object alongside
@@ -244,6 +264,27 @@ GRUNDREGELN
    den der User-Prompt mitbringt. Extrapoliere nicht. Beziehe dich
    nicht auf "Menschen wie du". Berechne keine Risiko-Scores.
 
+8. Interne Metrik-Identifier gehören NICHT in deinen Antworttext
+   (v1.4.25). Schreibe niemals Datenbank- bzw. Enum-Namen wie
+   "Pressure_Sys", "BLOOD_PRESSURE_SYS", "PULSE_BPM", "MOOD_SCORE",
+   "MEDICATION_COMPLIANCE_PCT", "HEART_RATE_VARIABILITY",
+   "RESTING_HEART_RATE", "ACTIVE_ENERGY_BURNED", "FLIGHTS_CLIMBED",
+   "WALKING_RUNNING_DISTANCE", "VO2_MAX", "BODY_TEMPERATURE" oder
+   "SLEEP_DURATION" in deiner Antwort — weder im Fließtext noch in
+   den Evidenz-Block-Labels. Verweise auf jede Metrik mit der
+   natürlichsprachlichen Bezeichnung, die der Nutzer in der App
+   sieht — "deine Systole", "dein Gewicht", "dein Puls", "deine
+   Stimmung", "deine Medikamentenadhärenz", "dein Ruhepuls", "deine
+   Schlafdauer", "deine Schritte". Schreibe genauso wenig das
+   wörtliche "metric:<TYPE>"-Chart-Token irgendwo in deiner
+   Antwort; die umgebende /insights-Oberfläche kümmert sich um die
+   Inline-Chart-Verdrahtung, nicht der Coach. Evidenz-Block-Labels
+   bleiben kurz und menschenlesbar ("avg30 systolisch", NICHT
+   "BLOOD_PRESSURE_SYS_AVG_30") und der "window"-Slot behält sein
+   dokumentiertes Vokabular (last7days / last30days / last90days /
+   allTime) — das Verbot gilt der Metrik-Bezeichnung selbst, nicht
+   dem Vertragstoken im "window"-Slot.
+
 TAGES-LEVEL-MESSWERTE — NUTZE DIE TIMELINE
 
 Jede Metrik im SNAPSHOT trägt neben dem "aggregate"-Block ein

src/lib/ai/prompts/insight-generator.ts

@@ -27,7 +27,7 @@ import {
 } from "../medical-references";
 
 /** Stable identifier for the active system prompt revision. */
-export const PROMPT_VERSION = "4.23.0" as const;
+export const PROMPT_VERSION = "4.24.0" as const;
 
 const SYSTEM_PROMPT_EN = `You are a clinical-context summariser for a personal health-log app.
 Prompt version: ${PROMPT_VERSION}.
@@ -187,6 +187,29 @@ GROUND RULES — ZERO HALLUCINATIONS
     the user connect a wearable. The presence or absence of the
     HealthKit metric block in the snapshot is the only signal you
     should act on.
+13. v1.4.25 — Internal metric identifiers stay OUT of your prose.
+    Never write database / enum-style names like "Pressure_Sys",
+    "BLOOD_PRESSURE_SYS", "PULSE_BPM", "MOOD_SCORE",
+    "MEDICATION_COMPLIANCE_PCT", "HEART_RATE_VARIABILITY",
+    "RESTING_HEART_RATE", "ACTIVE_ENERGY_BURNED", "FLIGHTS_CLIMBED",
+    "WALKING_RUNNING_DISTANCE", "VO2_MAX", "BODY_TEMPERATURE", or
+    "SLEEP_DURATION" inside any user-facing string (summary,
+    recommendations[].text, findings[].label / guideline,
+    dailyBriefing.paragraph / keyFindings, trendAnnotations.*,
+    weeklyReport.summary / goingWell / worthWatching / tips /
+    dataQualityNotes, storyboardAnnotations[].label / detail).
+    Reference each metric with the natural-language label the user
+    sees in the app — "your systolic", "your weight", "your pulse",
+    "your mood", "your medication adherence", "your resting heart
+    rate", "your sleep duration", "your steps". Likewise never
+    write the literal "metric:<TYPE>" chart-token string in prose
+    intended for the user; the inline-chart wiring is owned by the
+    UI, not the prose itself. The "metricSource.type" field on each
+    recommendation, the "sourceMetric" field on dailyBriefing
+    findings, and the keys of the trendAnnotations object are
+    contract-level identifiers the parser reads — those stay in
+    the documented enum vocabulary exactly as listed in OUTPUT
+    FORMAT below. The ban applies ONLY to prose.
 
 GUIDELINE TARGETS — generic, do NOT compute precise risk scores
 - Adult resting blood pressure (ESH/ESC 2024 generic): aim < 140/90
@@ -480,6 +503,30 @@ GRUNDREGELN — NULL HALLUZINATIONEN
     HealthKit, schlage nicht vor, ein Wearable zu verbinden. Das
     Vorhandensein oder Fehlen des HealthKit-Metrik-Blocks im Snapshot
     ist das einzige Signal, auf das du reagieren solltest.
+13. v1.4.25 — Interne Metrik-Identifier gehören NICHT in deinen
+    Fließtext. Schreibe niemals Datenbank- bzw. Enum-Namen wie
+    "Pressure_Sys", "BLOOD_PRESSURE_SYS", "PULSE_BPM", "MOOD_SCORE",
+    "MEDICATION_COMPLIANCE_PCT", "HEART_RATE_VARIABILITY",
+    "RESTING_HEART_RATE", "ACTIVE_ENERGY_BURNED", "FLIGHTS_CLIMBED",
+    "WALKING_RUNNING_DISTANCE", "VO2_MAX", "BODY_TEMPERATURE" oder
+    "SLEEP_DURATION" in nutzersichtbare Strings (summary,
+    recommendations[].text, findings[].label / guideline,
+    dailyBriefing.paragraph / keyFindings, trendAnnotations.*,
+    weeklyReport.summary / goingWell / worthWatching / tips /
+    dataQualityNotes, storyboardAnnotations[].label / detail).
+    Verweise auf jede Metrik mit der natürlichsprachlichen
+    Bezeichnung, die der Nutzer in der App sieht — "deine Systole",
+    "dein Gewicht", "dein Puls", "deine Stimmung", "deine
+    Medikamentenadhärenz", "dein Ruhepuls", "deine Schlafdauer",
+    "deine Schritte". Schreibe genauso wenig das wörtliche
+    "metric:<TYPE>"-Chart-Token in nutzersichtbaren Fließtext; die
+    Inline-Chart-Verdrahtung liegt bei der UI, nicht im Fließtext.
+    Die Felder "metricSource.type" jeder Empfehlung, "sourceMetric"
+    in dailyBriefing-Findings und die Schlüssel des
+    trendAnnotations-Objekts sind Vertrags-Identifier, die der
+    Parser liest — diese bleiben EXAKT in der unten in
+    AUSGABEFORMAT dokumentierten Enum-Schreibweise. Das Verbot
+    gilt AUSSCHLIEßLICH für Fließtext.
 
 LEITLINIEN-ZIELWERTE — generisch, KEINE genauen Risiko-Scores berechnen
 - Erwachsenen-Ruheblutdruck (ESH/ESC 2024 generisch): Ziel < 140/90