fix(labs): drop a lab's calculated-value note from the unit (#1313)

A read that copied "(calc)" into the unit minted a marker in "g/dL (calc)" that refused the next report's "g/dL". The note is dropped at the read and ignored in the unit comparison; any other bracket still counts.

CHANGELOG.md

@@ -24,6 +24,18 @@
   fell through to the international set with 112, which reaches the
   police there, not an ambulance.
 
+### Fixed
+
+- **A lab's "(calc)" note no longer becomes part of the unit.** Some
+  reports print "(calc)" beside a calculated value, and a document read
+  could copy it into the unit, so a marker was created in "g/dL (calc)"
+  or just "(calc)" and the next report's plain "g/dL" was refused as a
+  different unit. The read now drops that note (also "calculated",
+  "computed" and the German "berechnet"), a unit that was only the note
+  counts as no unit, and a marker already saved with the note accepts
+  the plain unit. Any other text in brackets still counts as part of
+  the unit. (#1313)
+
 ## [1.47.1] — 2026-11-07
 
 ### Added

src/lib/documents/__tests__/extract.test.ts

@@ -193,6 +193,36 @@ describe("runInboundExtraction — stated-status-only mapping", () => {
     expect(result.facts.map((f) => f.needsReview)).toEqual([true, false]);
   });
 
+  it('drops a lab\'s "(calc)" note from the unit, so two reports land in one', async () => {
+    const provider = fakeProvider(
+      JSON.stringify({
+        facts: [
+          {
+            type: "OBSERVATION",
+            label: "Globulin",
+            value: 2.6,
+            unit: "g/dL (calc)",
+            confidence: 0.95,
+            sourceText: "Globulin 2.6 g/dL (calc)",
+          },
+          {
+            type: "OBSERVATION",
+            label: "A/G Ratio",
+            value: 1.7,
+            unit: "(calc)",
+            confidence: 0.95,
+            sourceText: "A/G Ratio 1.7 (calc)",
+          },
+        ],
+      }),
+    );
+
+    const result = await runInboundExtraction(oneRead(provider));
+    expect(
+      result.facts.map((f) => (f.data as ObservationFactData).unit),
+    ).toEqual(["g/dL", null]);
+  });
+
   it("stores the document's real span, not the model's echo of it", async () => {
     // The OCR text says "Hämoglobin    14.2 g/dL"; the model echoes a tidied
     // rendering of it. The stored provenance must be the document's characters.

src/lib/documents/extract.ts

@@ -58,6 +58,7 @@ import {
   type OcrReadSurface,
 } from "@/lib/documents/ocr-read";
 import { isDimensionlessUnit } from "@/lib/labs/dimensionless";
+import { withoutLabUnitAnnotation } from "@/lib/labs/unit-normalise";
 import { annotate } from "@/lib/logging/context";
 import {
   INBOUND_CONFIDENCE_FLOOR,
@@ -168,7 +169,7 @@ function mapFactData(raw: ExtractedFactRaw): FactData | null {
       codeSystem: system === "LOINC" ? system : null,
       value: hasNumeric ? raw.value : null,
       valueText: hasNumeric ? null : raw.valueText,
-      unit: raw.unit,
+      unit: observationUnit(raw.unit),
       referenceLow: raw.referenceLow,
       referenceHigh: raw.referenceHigh,
       referenceText: raw.referenceText,
@@ -189,6 +190,17 @@ function mapFactData(raw: ExtractedFactRaw): FactData | null {
   return data;
 }
 
+/**
+ * The printed unit without a lab's "(calc)" note, so the same marker read
+ * from two reports, one with the note and one without, lands in one unit.
+ * A unit that was only the note is no unit.
+ */
+function observationUnit(unit: string | null): string | null {
+  if (unit === null) return null;
+  const bare = withoutLabUnitAnnotation(unit);
+  return bare === "" ? null : bare;
+}
+
 /** A staged fact ready to persist (before it gets an id). */
 export interface StagedFactInput {
   factType: InboundFactType;
@@ -306,7 +318,7 @@ function factKey(raw: ExtractedFactRaw, readDate: string | null): string {
     raw.label.trim().toLowerCase(),
     raw.value ?? "",
     (raw.valueText ?? "").trim().toLowerCase(),
-    (raw.unit ?? "").trim().toLowerCase(),
+    (observationUnit(raw.unit) ?? "").toLowerCase(),
     (raw.referenceText ?? "").trim().toLowerCase(),
     (raw.dose ?? "").trim().toLowerCase(),
     normaliseDate(raw.effectiveDate) ?? readDate ?? "",

src/lib/labs/__tests__/unit-normalise.test.ts

@@ -1,6 +1,10 @@
 import { describe, expect, it } from "vitest";
 
-import { normaliseLabUnit, sameLabUnit } from "../unit-normalise";
+import {
+  normaliseLabUnit,
+  sameLabUnit,
+  withoutLabUnitAnnotation,
+} from "../unit-normalise";
 
 describe("sameLabUnit — spellings of one unit", () => {
   it.each([
@@ -88,3 +92,19 @@ describe("normaliseLabUnit", () => {
     expect(normaliseLabUnit("MG/DL")).toBe("mg/dL");
   });
 });
+
+describe("a calculated-value note in the unit", () => {
+  it("is not part of the unit", () => {
+    expect(sameLabUnit("g/dL (calc)", "g/dL")).toBe(true);
+    expect(sameLabUnit("mmol/l (berechnet)", "mmol/L")).toBe(true);
+    expect(sameLabUnit("mg/dL (Calc.)", "MG/DL")).toBe(true);
+    expect(withoutLabUnitAnnotation("(calc)")).toBe("");
+    expect(withoutLabUnitAnnotation(" g/dL (calculated) ")).toBe("g/dL");
+  });
+
+  it("keeps any other bracket, so it still surfaces as a mismatch", () => {
+    expect(sameLabUnit("mg/dL (fasting)", "mg/dL")).toBe(false);
+    expect(withoutLabUnitAnnotation("U/L (37 °C)")).toBe("U/L (37 °C)");
+    expect(withoutLabUnitAnnotation("(calc) g/dL")).toBe("(calc) g/dL");
+  });
+});

src/lib/labs/unit-normalise.ts

@@ -21,9 +21,31 @@
  * does not recognise is compared exactly, so an unfamiliar spelling surfaces as
  * a mismatch rather than being guessed at.
  *
+ * A lab's note that a value was calculated rather than measured (`g/dL
+ * (calc)`, `(berechnet)`) is not part of the unit and is dropped first; see
+ * `withoutLabUnitAnnotation`.
+ *
  * Pure and dependency-free.
  */
 
+/**
+ * A trailing note in brackets that says the lab calculated the value. Some
+ * reports print it in the unit column, and a read that copies the column
+ * verbatim then minted a marker in "g/dL (calc)", which refused the next
+ * report's plain "g/dL". Only these words are dropped, so a bracket that
+ * carries meaning stays and still surfaces as a mismatch.
+ */
+const CALCULATED_NOTE =
+  /\s*\((?:calc\.?|calculated|computed|berechnet|ber\.?)\)\s*$/i;
+
+/**
+ * The unit without a trailing calculated-value note, trimmed: `g/dL (calc)`
+ * reads `g/dL`, and a unit that was only the note reads `""`.
+ */
+export function withoutLabUnitAnnotation(raw: string): string {
+  return raw.replace(CALCULATED_NOTE, "").trim();
+}
+
 /** Units that may be recognised when printed entirely in capitals. */
 const SHOUTED_UNITS: ReadonlyMap<string, string> = new Map(
   [
@@ -79,7 +101,7 @@ function foldLitre(text: string): string {
  * means the same thing, and the (trimmed) string itself otherwise.
  */
 export function normaliseLabUnit(raw: string): string {
-  const spaced = foldSpacing(raw);
+  const spaced = foldSpacing(withoutLabUnitAnnotation(raw));
 
   const hasLowercase = /[a-zµμ]/.test(spaced);
   const hasUppercase = /[A-Z]/.test(spaced);