diff --git a/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.spec.ts b/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.spec.ts
index 1a700986a..0c01befe2 100644
--- a/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.spec.ts
+++ b/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.spec.ts
@@ -54,21 +54,65 @@ describe(aggregateCriteria.name, () => {
       expect(c.totalEphemeralStorage).toBe(1000n);
     });
 
-    it("skips ram-class storage from totalMemory (issue 4 territory)", () => {
+    it("rolls ram-class volumes into totalMemory and leaves storage totals untouched", () => {
       const c = aggregateCriteria(
         [
           makeUnit({
             memory: 100n,
-            count: 1,
-            storage: [{ name: "shm", quantity: 99_999n, attributes: [{ key: "class", value: "ram" }] }]
+            count: 3,
+            storage: [{ name: "shm", quantity: 50n, attributes: [{ key: "class", value: "ram" }] }]
           })
         ],
         makeRequirements()
       );
-      expect(c.totalMemory).toBe(100n);
+      expect(c.totalMemory).toBe((100n + 50n) * 3n);
       expect(c.totalEphemeralStorage).toBe(0n);
       expect(c.totalPersistentStorage).toBe(0n);
     });
+
+    it("sums multiple ram-class volumes on a single unit into totalMemory", () => {
+      const c = aggregateCriteria(
+        [
+          makeUnit({
+            memory: 100n,
+            count: 1,
+            storage: [
+              { name: "shm1", quantity: 25n, attributes: [{ key: "class", value: "ram" }] },
+              { name: "shm2", quantity: 75n, attributes: [{ key: "class", value: "ram" }] }
+            ]
+          })
+        ],
+        makeRequirements()
+      );
+      expect(c.totalMemory).toBe(200n);
+    });
+
+    it("partitions ram, ephemeral, and persistent volumes on the same unit without double-counting", () => {
+      const c = aggregateCriteria(
+        [
+          makeUnit({
+            memory: 100n,
+            count: 2,
+            storage: [
+              { name: "shm", quantity: 50n, attributes: [{ key: "class", value: "ram" }] },
+              { name: "scratch", quantity: 500n, attributes: [] },
+              {
+                name: "data",
+                quantity: 1000n,
+                attributes: [
+                  { key: "persistent", value: "true" },
+                  { key: "class", value: "beta2" }
+                ]
+              }
+            ]
+          })
+        ],
+        makeRequirements()
+      );
+      expect(c.totalMemory).toBe((100n + 50n) * 2n);
+      expect(c.totalEphemeralStorage).toBe(500n * 2n);
+      expect(c.totalPersistentStorage).toBe(1000n * 2n);
+    });
   });
 
   describe("maxPerReplica", () => {
@@ -88,6 +132,21 @@ describe(aggregateCriteria.name, () => {
       expect(c.maxPerReplicaMemory).toBe(50n);
     });
 
+    it("includes ram-class volume sizes when picking maxPerReplicaMemory", () => {
+      const c = aggregateCriteria(
+        [
+          makeUnit({ memory: 100n, count: 1 }),
+          makeUnit({
+            memory: 30n,
+            count: 1,
+            storage: [{ name: "shm", quantity: 200n, attributes: [{ key: "class", value: "ram" }] }]
+          })
+        ],
+        makeRequirements()
+      );
+      expect(c.maxPerReplicaMemory).toBe(230n);
+    });
+
     it("picks max of unit.gpu across units", () => {
       const c = aggregateCriteria([makeUnit({ gpu: 1n, count: 8 }), makeUnit({ gpu: 4n, count: 1 })], makeRequirements());
       expect(c.maxPerReplicaGpu).toBe(4n);
diff --git a/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.ts b/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.ts
index 82ae673e5..ab11ec10e 100644
--- a/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.ts
+++ b/apps/provider-inventory/src/repositories/bid-screening/bid-screening.aggregator.ts
@@ -37,27 +37,28 @@ export function aggregateCriteria(resourceUnits: RequestedResourceUnit[], requir
   for (const unit of resourceUnits) {
     const count = BigInt(unit.count);
     const cpu = unit.resources.cpu.units;
-    const memory = unit.resources.memory.quantity;
     const gpu = unit.resources.gpu.units;
 
-    totalCpu += cpu * count;
-    totalMemory += memory * count;
-    totalGpu += gpu * count;
-
-    if (cpu > maxPerReplicaCpu) maxPerReplicaCpu = cpu;
-    if (memory > maxPerReplicaMemory) maxPerReplicaMemory = memory;
-    if (gpu > maxPerReplicaGpu) maxPerReplicaGpu = gpu;
-
+    let effectiveMemory = unit.resources.memory.quantity;
     for (const vol of unit.resources.storage) {
       const parsed = parseStorageAttributes(vol.attributes);
       if (parsed.classification === "persistent") {
         totalPersistentStorage += vol.quantity * count;
       } else if (parsed.classification === "ephemeral") {
         totalEphemeralStorage += vol.quantity * count;
+      } else if (parsed.classification === "ram") {
+        effectiveMemory += vol.quantity;
       }
-      // ram volumes intentionally skipped — issue 4 will add them to totalMemory
     }
 
+    totalCpu += cpu * count;
+    totalMemory += effectiveMemory * count;
+    totalGpu += gpu * count;
+
+    if (cpu > maxPerReplicaCpu) maxPerReplicaCpu = cpu;
+    if (effectiveMemory > maxPerReplicaMemory) maxPerReplicaMemory = effectiveMemory;
+    if (gpu > maxPerReplicaGpu) maxPerReplicaGpu = gpu;
+
     units.push({
       gpuTokens: collectGpuTokens(unit.resources.gpu),
       persistentClasses: collectPersistentStorageTokens(unit.resources.storage)
diff --git a/apps/provider-inventory/src/repositories/bid-screening/bid-screening.repository.integration.ts b/apps/provider-inventory/src/repositories/bid-screening/bid-screening.repository.integration.ts
index 6fd60dfc6..71ebbc5e6 100644
--- a/apps/provider-inventory/src/repositories/bid-screening/bid-screening.repository.integration.ts
+++ b/apps/provider-inventory/src/repositories/bid-screening/bid-screening.repository.integration.ts
@@ -38,6 +38,37 @@ describe(BidScreeningRepository.name, () => {
 
       expect(owners(rows)).toEqual(["akash1roomy"]);
     });
+
+    it("excludes providers whose memory headroom covers the bare unit but not the combined memory + ram-class volume demand", async () => {
+      // Workload requests 100 bytes of memory + a 200-byte ram-class volume → 300 bytes per replica.
+      const replicaMemory = 100n;
+      const ramVolume = 200n;
+      const effectivePerReplica = replicaMemory + ramVolume;
+
+      await seed({
+        owner: "akash1tightOnMemory",
+        totalAvailableMemory: effectivePerReplica - 1n,
+        maxNodeFreeMemory: effectivePerReplica - 1n
+      });
+      await seed({
+        owner: "akash1fits",
+        totalAvailableMemory: effectivePerReplica,
+        maxNodeFreeMemory: effectivePerReplica
+      });
+
+      const rows = await repository.findCandidates(
+        [
+          unit({
+            memory: replicaMemory,
+            count: 1,
+            storage: [{ name: "shm", quantity: ramVolume, attributes: [{ key: "class", value: "ram" }] }]
+          })
+        ],
+        requirements()
+      );
+
+      expect(owners(rows)).toEqual(["akash1fits"]);
+    });
   });
 
   describe("signedBy", () => {