fix(coach): narrow the pt-BR population-norm patterns

A sentence the grounding check reads as population guidance has its
numbers exempted from the evidence check. Two of the pt-BR patterns
matched sentences about the reader's own readings: "(entre|de) ...
geralmente" fired on "Sua pressão de 158/97 geralmente fica...", and a
singular "médico ... recomenda" with a 30-character gap fired on "Seu
médico recomenda manter a pressão em 158/97". Either let an invented
figure through unchecked.

Every pt-BR entry now needs a population subject or a reference noun:
plural guideline, doctor or specialist subjects next to their verb,
plural adults needing something, and "a/o <metric> normal" without a
possessive. Portuguese month names join the written-date pass so "21 de
julho" is not graded as a figure, and quilos, por cento and passos
join the unit words.

Negative cases prove the reader's-own sentences are checked again;
positive cases keep the reference sentences exempt.

src/lib/ai/coach/__tests__/coach-prose-grounding.test.ts

@@ -600,6 +600,7 @@ describe("tokenizing prose written in the reader's language", () => {
     ["es", "El 21 de julio, su tensión era de 128 mmHg."],
     ["it", "Il 21 luglio la pressione era di 128 mmHg."],
     ["pl", "21 lipca ciśnienie wynosiło 128 mmHg."],
+    ["pt-BR", "Em 21 de julho, sua pressão era de 128 mmHg."],
   ] as const)("leaves a written %s date alone", (locale, prose) => {
     const findings = findUnverifiedCoachNumbers(
       prose,
@@ -623,3 +624,38 @@ describe("tokenizing prose written in the reader's language", () => {
     expect(findings.map((f) => f.value)).not.toContain(21);
   });
 });
+
+/**
+ * A sentence framed as population guidance has its numbers exempted from the
+ * check, so a framing pattern that also matches a sentence about the reader's
+ * own readings lets an invented figure through unchecked. The Brazilian
+ * Portuguese bank needs a population subject or a reference noun for that
+ * reason; the first two cases here were exempted by the broader patterns it
+ * started with.
+ */
+describe("Brazilian Portuguese population-norm framing", () => {
+  const payloads = [{ metric: "bp", latest: 128 }];
+
+  it.each([
+    "Sua pressão de 158/97 geralmente fica acima do seu padrão.",
+    "Seu médico recomenda manter a pressão em 158/97.",
+    "Entre as suas medições, a pressão normalmente fica em 158/97.",
+    "Para um adulto como você, 158/97 geralmente precisa de atenção.",
+  ])("still checks the numbers in: %s", (prose) => {
+    const values = findUnverifiedCoachNumbers(prose, payloads, "pt-BR").map(
+      (f) => f.value,
+    );
+    expect(values).toContain(158);
+    expect(values).toContain(97);
+  });
+
+  it.each([
+    "A faixa normal do pulso em repouso fica entre 60 e 100 bpm.",
+    "Os médicos recomendam de 7 a 9 horas de sono por noite.",
+    "A maioria dos adultos precisa de 7 a 9 horas de sono.",
+    "A pressão arterial normal de um adulto fica abaixo de 120/80 mmHg.",
+    "Os valores de referência da glicemia em jejum vão de 70 a 99 mg/dL.",
+  ])("exempts the reference figures in: %s", (prose) => {
+    expect(findUnverifiedCoachNumbers(prose, payloads, "pt-BR")).toEqual([]);
+  });
+});

src/lib/ai/coach/coach-prose-grounding.ts

@@ -156,6 +156,10 @@ const UNIT_WORDS: Readonly<Record<NumKind, readonly string[]>> = {
     "livres",
     "libra",
     "libras",
+    "quilo",
+    "quilos",
+    "quilograma",
+    "quilogramas",
     "funt",
     "funty",
   ],
@@ -170,6 +174,7 @@ const UNIT_WORDS: Readonly<Record<NumKind, readonly string[]>> = {
     "pour cent",
     "pourcent",
     "por ciento",
+    "por cento",
     "porciento",
     "per cento",
     "procent",
@@ -214,6 +219,7 @@ const UNIT_WORDS: Readonly<Record<NumKind, readonly string[]>> = {
     "pasos",
     "passo",
     "passi",
+    "passos",
     "krok",
     "kroki",
     "krokow",
@@ -626,6 +632,8 @@ const MONTHS = [
   "stycznia|lutego|marca|kwietnia|maja|czerwca|lipca|sierpnia|września|wrzesnia|października|pazdziernika|listopada|grudnia",
   // nl
   "januari|februari|maart|mei|augustus",
+  // pt-BR
+  "janeiro|fevereiro|março|marco|maio|junho|julho|setembro|outubro|novembro|dezembro",
 ].join("|");
 
 /** Parse a raw number token honouring the reply language's separators. */
@@ -635,7 +643,7 @@ function parseLocaleNumber(raw: string, locale: Locale): number | null {
     // EN: "," groups thousands, "." is the decimal point.
     normalised = raw.replace(/,/g, "");
   } else {
-    // DE / fr / es / it / pl / nl: "." groups thousands, "," is the decimal
+    // DE / fr / es / it / pl / nl / pt-BR: "." groups thousands, "," is the decimal
     // point.
     normalised = raw.replace(/\./g, "").replace(/,/g, ".");
   }
@@ -880,18 +888,23 @@ const POPULATION_NORM_PATTERNS: Record<Locale, readonly RegExp[]> = {
     /\b(?:volwassenen|mensen)\s+(?:hebben\s+)?(?:meestal|doorgaans|gewoonlijk|normaal\s+gesproken|over\s+het\s+algemeen)\s+(?:nodig|hebben)\b/i,
     /\b(?:referentie|normaal|aanbevolen)\w*\s*(?:bereik|waarden?)\b/i,
     /\bligt\s+(?:normaal\s+gesproken|meestal|doorgaans|gewoonlijk)\s+(?:tussen|rond|op)\b/i,
-  ], // No `\b` next to Hangul (it is an ASCII \w boundary), and the object sits
-  // between subject and verb in Korean, so each entry allows a gap where the
-  // Latin banks put \s+.
+  ],
+  // pt-BR. Every entry needs a population subject or a reference noun: a
+  // reading of the reader's own ("Sua pressão de 158/97 geralmente fica…",
+  // "Seu médico recomenda…") is a claim about their data, and an exempted
+  // sentence is one whose numbers are never checked.
   "pt-BR": [
     /\bfaixa\s+(?:de\s+refer[eê]ncia|normal)\b/i,
     /\bvalor(?:es)?\s+de\s+refer[eê]ncia\b/i,
     /\b(?:adultos?|pessoas?)\s+saud[áa]ve(?:l|is)\b/i,
     /\bmaioria\s+(?:dos\s+)?(?:adultos?|das\s+pessoas)\b/i,
-    /\b(?:diretrizes|guias|m[eé]dicos?|especialistas?)\b[^.?!]{0,30}\b(?:recomend(?:am|a)|suger(?:em|e)|indic(?:am|a))\b/i,
-    /\b(?:adultos?|pessoas?)\b[^.?!]{0,30}\b(?:geralmente|normalmente|em\s+geral)\b[^.?!]{0,30}\b(?:precis(?:am|a)|devem|necessit(?:am|a))\b/i,
-    /\b(?:entre|de)\s+[^.?!]{0,30}\b(?:normalmente|geralmente|em\s+geral)\b/i,
+    /\b(?:as\s+|os\s+)?(?:diretrizes|m[eé]dicos|especialistas)\s+(?:geralmente\s+|em\s+geral\s+)?(?:dizem|recomendam|aconselham|sugerem|indicam|consideram)\b/i,
+    /\b(?:os\s+)?(?:adultos|as\s+pessoas)\s+(?:geralmente\s+|normalmente\s+|em\s+geral\s+)?(?:precisam|necessitam)\s+de\b/i,
+    /\b(?:a|o)\s+(?:frequ[êe]ncia\s+card[íi]aca|pulso|press[ãa]o(?:\s+arterial)?|glicemia|glicose)(?:\s+de\s+repouso)?\s+(?:normal|saud[áa]vel|t[íi]pica|t[íi]pico)\b/i,
   ],
+  // No `\b` next to Hangul (it is an ASCII \w boundary), and the object sits
+  // between subject and verb in Korean, so each entry allows a gap where the
+  // Latin banks put \s+.
   ko: [
     /정상\s*범위(?:는|가|를|로|의)/,
     /(?:참고|기준|권장)\s*(?:범위|치|값)/,