diff --git a/src/app/api/biomarkers/[id]/route.ts b/src/app/api/biomarkers/[id]/route.ts
index d3aec468b..506333ad3 100644
--- a/src/app/api/biomarkers/[id]/route.ts
+++ b/src/app/api/biomarkers/[id]/route.ts
@@ -17,6 +17,7 @@ import {
   decryptContextFromBytes,
   encryptContextToBytes,
 } from "@/lib/labs/biomarker-store";
+import { matchAnalyteKey } from "@/lib/labs/analyte-match";
 import { annotate } from "@/lib/logging/context";
 import { updateBiomarkerSchema } from "@/lib/validations/biomarkers";
 import { effectiveBound, isInvertedRange } from "@/lib/validations/labs";
@@ -176,6 +177,16 @@ export const PUT = apiHandler(
     if (d.upperBound !== undefined) data.upperBound = d.upperBound;
     if (d.panel !== undefined) data.panel = d.panel;
     if (d.hidden !== undefined) data.hidden = d.hidden;
+    // The analyte a marker is follows from its name and unit, the same match
+    // the creation paths use. A rename must move it, or a marker renamed from
+    // "Glucose" to "Cholesterol" would keep converting readings with the
+    // glucose factor; a name that now matches nothing clears it.
+    if (d.name !== undefined || d.unit !== undefined) {
+      data.analyteKey = matchAnalyteKey({
+        name: d.name ?? existing.name,
+        unit: d.unit ?? existing.unit,
+      });
+    }
 
     // Inverted-range guard for a PARTIAL bound update. The schema refine only
     // fires when both bounds arrive together; moving a single bound past the
diff --git a/src/lib/labs/analyte-key-store.ts b/src/lib/labs/analyte-key-store.ts
index f11720e53..dfefa416d 100644
--- a/src/lib/labs/analyte-key-store.ts
+++ b/src/lib/labs/analyte-key-store.ts
@@ -1,37 +1,53 @@
 /**
  * Link a marker to its analyte for the unit conversion.
  *
- * The key is set when a marker is created and, for a marker that already
- * existed, the first time a reading needs it. It is never overwritten: a key a
- * person (or an earlier pass) set stands, and `updateMany` is gated on the
- * column still being null, so two concurrent writers converge. A marker whose
- * name matches no analyte keeps a null key and gets no conversion.
+ * There is no manual choice of analyte: the key is a function of the marker's
+ * name and unit (`matchAnalyteKey`). It is set when a marker is created and
+ * re-derived whenever its name or unit changes, and this step brings a stored
+ * key back in line with that function the first time a reading needs it. So a
+ * marker that never had a key gets one, and a key that no longer agrees with
+ * the name (written before a rename re-derived it, or in an older form) is
+ * replaced, or cleared when the name now matches nothing. A key that agrees is
+ * left alone, so the step is idempotent.
+ *
+ * The write is gated on the column still holding what the caller read, so two
+ * concurrent writers converge, and a caller holding a stale copy re-reads the
+ * row rather than writing over a newer value.
  */
 import type { PrismaClient } from "@/generated/prisma/client";
 
 import { matchAnalyteKey } from "./analyte-match";
 
+interface MarkerKeyState {
+  id: string;
+  name: string;
+  unit: string;
+  analyteKey?: string | null;
+}
+
 export async function ensureAnalyteKey(
   client: Pick<PrismaClient, "biomarker">,
-  marker: {
-    id: string;
-    name: string;
-    unit: string;
-    analyteKey?: string | null;
-  },
+  marker: MarkerKeyState,
 ): Promise<string | null> {
-  if (marker.analyteKey) return marker.analyteKey;
-  const key = matchAnalyteKey({ name: marker.name, unit: marker.unit });
-  if (!key) return null;
-  const { count } = await client.biomarker.updateMany({
-    where: { id: marker.id, analyteKey: null },
-    data: { analyteKey: key },
-  });
-  if (count > 0) return key;
-  // Another writer set it first; read what stands.
-  const current = await client.biomarker.findUnique({
-    where: { id: marker.id },
-    select: { analyteKey: true },
-  });
-  return current?.analyteKey ?? null;
+  let current: MarkerKeyState = marker;
+  // Two rounds: the caller's copy, then the row as it stands if that copy was
+  // stale. A third writer racing both is vanishingly rare; the derived key is
+  // still the right answer for the name this call saw.
+  for (let round = 0; round < 2; round++) {
+    const derived = matchAnalyteKey({ name: current.name, unit: current.unit });
+    const stored = current.analyteKey ?? null;
+    if (stored === derived) return derived;
+    const { count } = await client.biomarker.updateMany({
+      where: { id: current.id, analyteKey: stored },
+      data: { analyteKey: derived },
+    });
+    if (count > 0) return derived;
+    const row = await client.biomarker.findUnique({
+      where: { id: current.id },
+      select: { id: true, name: true, unit: true, analyteKey: true },
+    });
+    if (!row) return null;
+    current = row;
+  }
+  return matchAnalyteKey({ name: current.name, unit: current.unit });
 }
diff --git a/src/lib/labs/ocr-extract.ts b/src/lib/labs/ocr-extract.ts
index 657445fd3..59740c81a 100644
--- a/src/lib/labs/ocr-extract.ts
+++ b/src/lib/labs/ocr-extract.ts
@@ -146,16 +146,18 @@ export async function annotateRow(
   // Biomarker match — case-insensitive on the catalog identity, presence-only.
   const existingBiomarker = await prisma.biomarker.findFirst({
     where: { userId, name: { equals: analyte, mode: "insensitive" } },
-    select: { id: true, name: true, unit: true, analyteKey: true },
+    select: { id: true, name: true, unit: true },
   });
   // The analyte the marker is, so the review can say whether a row in another
-  // unit converts. Read only: the key is filled when a reading is written.
+  // unit converts. Derived from the name and unit exactly as the commit derives
+  // it (a stored key is only that derivation, written down), so the review and
+  // the commit cannot disagree. Read only: the key is stored when a reading is
+  // written.
   const markerAnalyteKey = existingBiomarker
-    ? (existingBiomarker.analyteKey ??
-      matchAnalyteKey({
+    ? matchAnalyteKey({
         name: existingBiomarker.name,
         unit: existingBiomarker.unit,
-      }))
+      })
     : null;
 
   // Duplicate check: a live lab_results row with the same analyte (fuzzy,
diff --git a/tests/integration/labs-analyte-key.test.ts b/tests/integration/labs-analyte-key.test.ts
index df12eb7c3..595c70261 100644
--- a/tests/integration/labs-analyte-key.test.ts
+++ b/tests/integration/labs-analyte-key.test.ts
@@ -1,15 +1,19 @@
 /**
- * A marker is linked to its analyte, once, and the link is never overwritten.
+ * A marker is linked to its analyte, and the link follows its name and unit.
  *
  * `Biomarker.analyteKey` says which analyte a marker is, so a reading in another
- * unit can be converted (#1095). It is set when a marker is created, by all
- * three creation paths, and, for a marker that already exists, the first time
- * it is needed. Against a real Postgres:
+ * unit can be converted (#1095). There is no manual choice: the key is derived
+ * from the name and unit. It is set when a marker is created, by all three
+ * creation paths, re-derived when a rename or unit change moves it, and, for a
+ * marker that already exists, brought in line the first time it is needed.
+ * Against a real Postgres:
  *   - a marker created by name is linked when the name is a known analyte and
  *     its unit is one the analyte reaches, and left null otherwise;
  *   - the key reaches the API response;
- *   - `ensureAnalyteKey` fills a null key once and never replaces a key that is
- *     set, even by a different match, and two concurrent calls converge;
+ *   - a rename moves the key, or clears it, and the next reading in another
+ *     unit converts with the new analyte's factor;
+ *   - `ensureAnalyteKey` fills a null key, replaces one that disagrees with the
+ *     name, and two concurrent calls converge;
  *   - the mint path used by the lab writes and the legacy backfill set it too.
  */
 import { NextRequest } from "next/server";
@@ -100,6 +104,109 @@ describe("POST /api/biomarkers", () => {
   });
 });
 
+async function patch(id: string, body: Record<string, unknown>) {
+  const s = await prisma.session.create({
+    data: { userId: USER, expiresAt: new Date(Date.now() + 3_600_000) },
+  });
+  cookieJar.clear();
+  cookieJar.set("healthlog_session", s.id);
+  const { PUT } = await import("@/app/api/biomarkers/[id]/route");
+  return PUT(
+    new NextRequest(`http://localhost/api/biomarkers/${id}`, {
+      method: "PUT",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify(body),
+    }),
+    { params: Promise.resolve({ id }) },
+  );
+}
+
+async function postReading(body: Record<string, unknown>) {
+  const s = await prisma.session.create({
+    data: { userId: USER, expiresAt: new Date(Date.now() + 3_600_000) },
+  });
+  cookieJar.clear();
+  cookieJar.set("healthlog_session", s.id);
+  const { POST } = await import("@/app/api/labs/route");
+  return POST(
+    new NextRequest("http://localhost/api/labs", {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify(body),
+    }),
+  );
+}
+
+describe("PUT /api/biomarkers/[id]", () => {
+  it("moves the key with a rename, and the next reading converts with the new analyte", async () => {
+    const created = await post({ name: "Blood sugar", unit: "mg/dL" });
+    const { data: marker } = (await created.json()) as {
+      data: { id: string; analyteKey: string | null };
+    };
+    expect(marker.analyteKey).toBe("fasting-glucose");
+
+    const res = await patch(marker.id, { name: "Cholesterol" });
+    expect(res.status).toBe(200);
+    const body = (await res.json()) as {
+      data: { analyteKey: string | null };
+    };
+    expect(body.data.analyteKey).toBe("total-cholesterol");
+    expect(
+      (await prisma.biomarker.findUnique({ where: { id: marker.id } }))
+        ?.analyteKey,
+    ).toBe("total-cholesterol");
+
+    // 5.2 mmol/L of cholesterol is 201.07 mg/dL (C27H46O, 386.664 g/mol). With
+    // the glucose key left over it would have been stored as 93.7.
+    const reading = await postReading({
+      biomarkerId: marker.id,
+      value: 5.2,
+      unit: "mmol/L",
+      takenAt: "2026-05-03T08:00:00.000Z",
+    });
+    expect(reading.status).toBe(201);
+    const { data: dto } = (await reading.json()) as {
+      data: { value: number; unit: string; converted: boolean };
+    };
+    expect(dto.converted).toBe(true);
+    expect(dto.unit).toBe("mg/dL");
+    expect(dto.value.toFixed(2)).toBe("201.07");
+  });
+
+  it("clears the key when the new name matches no analyte", async () => {
+    const created = await post({ name: "Harnstoff", unit: "mg/dL" });
+    const { data: marker } = (await created.json()) as {
+      data: { id: string; analyteKey: string | null };
+    };
+    expect(marker.analyteKey).toBe("urea");
+    const res = await patch(marker.id, { name: "Harnstoff (eigene Notiz)" });
+    expect(res.status).toBe(200);
+    const body = (await res.json()) as { data: { analyteKey: string | null } };
+    expect(body.data.analyteKey).toBeNull();
+  });
+
+  it("re-derives the key when the unit changes on a marker without readings", async () => {
+    const created = await post({ name: "Glukose", unit: "mEq/L" });
+    const { data: marker } = (await created.json()) as {
+      data: { id: string; analyteKey: string | null };
+    };
+    expect(marker.analyteKey).toBeNull();
+    const res = await patch(marker.id, { unit: "mg/dL" });
+    expect(res.status).toBe(200);
+    const body = (await res.json()) as { data: { analyteKey: string | null } };
+    expect(body.data.analyteKey).toBe("fasting-glucose");
+  });
+
+  it("leaves the key alone on an edit that touches neither name nor unit", async () => {
+    const created = await post({ name: "Kalzium gesamt", unit: "mmol/L" });
+    const { data: marker } = (await created.json()) as { data: { id: string } };
+    const res = await patch(marker.id, { lowerBound: 2.1 });
+    expect(res.status).toBe(200);
+    const body = (await res.json()) as { data: { analyteKey: string | null } };
+    expect(body.data.analyteKey).toBe("calcium-total");
+  });
+});
+
 describe("ensureAnalyteKey", () => {
   it("fills a null key once", async () => {
     const marker = await prisma.biomarker.create({
@@ -113,7 +220,12 @@ describe("ensureAnalyteKey", () => {
     ).toBe("uric-acid");
   });
 
-  it("never replaces a key that is set, even when the name would match another", async () => {
+  // This used to pin the opposite: a set key was never replaced. With no
+  // manual choice of analyte, a key that disagrees with the name can only be
+  // left over from a rename (or written in an older form), and keeping it
+  // converted readings with the wrong analyte's factor. So the key follows the
+  // name.
+  it("replaces a key that disagrees with the marker's name", async () => {
     const marker = await prisma.biomarker.create({
       data: {
         userId: USER,
@@ -122,16 +234,47 @@ describe("ensureAnalyteKey", () => {
         analyteKey: "hba1c",
       },
     });
-    // The caller read the row with the key set.
-    expect(await ensureAnalyteKey(prisma, marker)).toBe("hba1c");
-    // A caller holding a stale copy with a null key does not overwrite either.
-    expect(
-      await ensureAnalyteKey(prisma, { ...marker, analyteKey: null }),
-    ).toBe("hba1c");
+    expect(await ensureAnalyteKey(prisma, marker)).toBe("fasting-glucose");
     expect(
       (await prisma.biomarker.findUnique({ where: { id: marker.id } }))
         ?.analyteKey,
-    ).toBe("hba1c");
+    ).toBe("fasting-glucose");
+  });
+
+  it("clears a key when the name matches no analyte", async () => {
+    const marker = await prisma.biomarker.create({
+      data: {
+        userId: USER,
+        name: "Some other marker",
+        unit: "mg/dL",
+        analyteKey: "fasting-glucose",
+      },
+    });
+    expect(await ensureAnalyteKey(prisma, marker)).toBeNull();
+    expect(
+      (await prisma.biomarker.findUnique({ where: { id: marker.id } }))
+        ?.analyteKey,
+    ).toBeNull();
+  });
+
+  it("re-reads the row when the caller's copy is stale", async () => {
+    const marker = await prisma.biomarker.create({
+      data: {
+        userId: USER,
+        name: "Urea",
+        unit: "mg/dL",
+        analyteKey: "urea",
+      },
+    });
+    // A copy read before the key was set: the gated write misses, the row as it
+    // stands already agrees, and nothing is overwritten.
+    expect(
+      await ensureAnalyteKey(prisma, { ...marker, analyteKey: null }),
+    ).toBe("urea");
+    expect(
+      (await prisma.biomarker.findUnique({ where: { id: marker.id } }))
+        ?.analyteKey,
+    ).toBe("urea");
   });
 
   it("converges when two callers fill the same marker", async () => {