diff --git a/src/app/api/purchase-orders/by-number/[poNumber]/items/route.ts b/src/app/api/purchase-orders/by-number/[poNumber]/items/route.ts
index 777775e..9f96318 100644
--- a/src/app/api/purchase-orders/by-number/[poNumber]/items/route.ts
+++ b/src/app/api/purchase-orders/by-number/[poNumber]/items/route.ts
@@ -19,6 +19,40 @@ import { pbAdmin } from "@/lib/pb";
  *   200 { po: null, items: [], note: "po_not_found" } when PO isn't matched
  *   502 pb_lookup_failed
  */
+
+///
+// / Levenshtein distance and similarity for fuzzy PO matching.
+///
+function levenshteinDistance(a: string, b: string): number {
+  if (a === b) return 0;
+  if (a.length === 0) return b.length;
+  if (b.length === 0) return a.length;
+
+  let prev = new Array(b.length + 1);
+  let curr = new Array(b.length + 1);
+  for (let j = 0; j <= b.length; j++) prev[j] = j;
+
+  for (let i = 1; i <= a.length; i++) {
+    curr[0] = i;
+    for (let j = 1; j <= b.length; j++) {
+      const cost = a[i - 1] === b[j - 1] ? 0 : 1;
+      curr[j] = Math.min(
+        curr[j - 1] + 1,        // insertion
+        prev[j] + 1,            // deletion
+        prev[j - 1] + cost,     // substitution
+      );
+    }
+    [prev, curr] = [curr, prev];
+  }
+  return prev[b.length];
+}
+
+function levenshteinSimilarity(a: string, b: string): number {
+  const maxLen = Math.max(a.length, b.length);
+  if (maxLen === 0) return 1;
+  return 1 - levenshteinDistance(a, b) / maxLen;
+}
+
 export async function GET(
   request: Request,
   { params }: { params: Promise<{ poNumber: string }> },
@@ -64,9 +98,15 @@ export async function GET(
   }
 
   if (!po) {
-    // Fuzzy fallback: strip non-alphanumeric, lowercase, compare.
-    const normalised = sanitisedPoNumber.replace(/[^a-z0-9]/gi, "").toLowerCase();
-    if (normalised.length >= 3) {
+    // OCR is fuzzy: it sometimes drops characters ("MEGCP09239"
+    // instead of "MEGCPO9239"), adds prefixes/slashes ("MEGCP09150 /
+    // SRU 11987"), or picks up the wrong number from a noisy slip.
+    // We have ~15 POs in PocketBase (one supplier per PO), so it's
+    // cheap to score the OCR'd value against all of them and pick the
+    // best. Levenshtein on the normalised (alphanumeric, lowercase)
+    // string handles single-character insertions/deletions perfectly.
+    const target = sanitisedPoNumber.replace(/[^a-z0-9]/gi, "").toLowerCase();
+    if (target.length >= 4) {
       try {
         const candidates = await pb
           .collection("purchase_orders")
@@ -74,13 +114,25 @@ export async function GET(
             fields: "id,purchase_order_number,supplier",
             perPage: 500,
           });
-        const hit = candidates.find(
-          (c) =>
-            (c.purchase_order_number ?? "")
-              .replace(/[^a-z0-9]/gi, "")
-              .toLowerCase() === normalised,
-        );
-        if (hit) po = hit as unknown as { id: string; purchase_order_number: string; supplier: string };
+        let bestScore = 0;
+        let bestHit: { id: string; purchase_order_number: string; supplier: string } | null = null;
+        for (const c of candidates) {
+          const candidate = (c.purchase_order_number ?? "")
+            .replace(/[^a-z0-9]/gi, "")
+            .toLowerCase();
+          if (candidate.length === 0) continue;
+          const score = levenshteinSimilarity(target, candidate);
+          if (score > bestScore) {
+            bestScore = score;
+            bestHit = c as unknown as { id: string; purchase_order_number: string; supplier: string };
+          }
+        }
+        // Threshold tuned so single-char insert/delete still matches
+        // (similarity 0.9 for 10-char strings), but completely different
+        // POs (similarity < 0.5) don't.
+        if (bestHit && bestScore >= 0.75) {
+          po = bestHit;
+        }
       } catch (err) {
         console.error("[po-items] fuzzy lookup failed:", err);
       }