diff --git a/crates/console/prometeu-vm/src/heap.rs b/crates/console/prometeu-vm/src/heap.rs
index b4ba5468..e5f22099 100644
--- a/crates/console/prometeu-vm/src/heap.rs
+++ b/crates/console/prometeu-vm/src/heap.rs
@@ -290,4 +290,34 @@ mod tests {
         // We can't access internal vector here, but stability is implied by handle not changing.
         assert_eq!(a.0, a.0); // placeholder sanity check
     }
+
+    #[test]
+    fn sweep_reclaims_unrooted_cycle() {
+        let mut heap = Heap::new();
+
+        // Build a 2-node cycle A <-> B using internal mutation (module-private access).
+        let a = heap.allocate_array(vec![]);
+        let b = heap.allocate_array(vec![]);
+
+        // Make A point to B and B point to A.
+        if let Some(slot) = heap.objects.get_mut(a.0 as usize) {
+            if let Some(obj) = slot.as_mut() {
+                obj.array_elems = Some(vec![Value::HeapRef(b)]);
+                obj.header.payload_len = 1;
+            }
+        }
+        if let Some(slot) = heap.objects.get_mut(b.0 as usize) {
+            if let Some(obj) = slot.as_mut() {
+                obj.array_elems = Some(vec![Value::HeapRef(a)]);
+                obj.header.payload_len = 1;
+            }
+        }
+
+        // No roots: perform sweep directly; both should be reclaimed.
+        heap.sweep();
+
+        assert!(!heap.is_valid(a));
+        assert!(!heap.is_valid(b));
+        assert_eq!(heap.len(), 0);
+    }
 }
diff --git a/crates/console/prometeu-vm/src/virtual_machine.rs b/crates/console/prometeu-vm/src/virtual_machine.rs
index 5c7e5199..ff9d95e0 100644
--- a/crates/console/prometeu-vm/src/virtual_machine.rs
+++ b/crates/console/prometeu-vm/src/virtual_machine.rs
@@ -2537,9 +2537,10 @@ mod tests {
     fn test_gc_keeps_roots_and_collects_unreachable_at_frame_sync() {
         use crate::object::ObjectKind;
 
-        // ROM: FRAME_SYNC; HALT
+        // ROM: FRAME_SYNC; FRAME_SYNC; HALT
         let mut rom = Vec::new();
         rom.extend_from_slice(&(OpCode::FrameSync as u16).to_le_bytes());
+        rom.extend_from_slice(&(OpCode::FrameSync as u16).to_le_bytes());
         rom.extend_from_slice(&(OpCode::Halt as u16).to_le_bytes());
 
         let mut vm = new_test_vm(rom.clone(), vec![]);
@@ -2571,4 +2572,88 @@ mod tests {
         assert!(vm.heap.is_valid(rooted));
         assert!(!vm.heap.is_valid(unreachable));
     }
+
+    #[test]
+    fn test_gc_simple_allocation_collection_cycle() {
+        use crate::object::ObjectKind;
+
+        // ROM: FRAME_SYNC; FRAME_SYNC; HALT
+        let mut rom = Vec::new();
+        rom.extend_from_slice(&(OpCode::FrameSync as u16).to_le_bytes());
+        rom.extend_from_slice(&(OpCode::FrameSync as u16).to_le_bytes());
+        rom.extend_from_slice(&(OpCode::Halt as u16).to_le_bytes());
+
+        let mut vm = new_test_vm(rom.clone(), vec![]);
+        vm.program.functions = std::sync::Arc::from(vec![prometeu_bytecode::FunctionMeta {
+            code_offset: 0,
+            code_len: rom.len() as u32,
+            ..Default::default()
+        }]);
+
+        // Make GC trigger on any allocation delta
+        vm.gc_alloc_threshold = 1;
+
+        // Cycle 1: allocate one unreachable object
+        let _h1 = vm.heap.allocate_object(ObjectKind::Bytes, &[1]);
+        assert_eq!(vm.heap.len(), 1);
+
+        let mut native = MockNative;
+        let mut ctx = HostContext::new(None);
+
+        // FRAME_SYNC should collect it (first FRAME_SYNC)
+        match vm.step(&mut native, &mut ctx) {
+            Err(LogicalFrameEndingReason::FrameSync) => {}
+            other => panic!("Expected FrameSync, got {:?}", other),
+        }
+        assert_eq!(vm.heap.len(), 0);
+
+        // Cycle 2: allocate again and collect again deterministically
+        let _h2 = vm.heap.allocate_object(ObjectKind::Bytes, &[2]);
+        assert_eq!(vm.heap.len(), 1);
+        // Second FRAME_SYNC should also be reached deterministically
+        match vm.step(&mut native, &mut ctx) {
+            Err(LogicalFrameEndingReason::FrameSync) => {}
+            other => panic!("Expected FrameSync, got {:?}", other),
+        }
+        assert_eq!(vm.heap.len(), 0);
+    }
+
+    #[test]
+    fn test_gc_many_short_lived_objects_stress() {
+        use crate::object::ObjectKind;
+
+        // ROM: FRAME_SYNC; HALT
+        let mut rom = Vec::new();
+        rom.extend_from_slice(&(OpCode::FrameSync as u16).to_le_bytes());
+        rom.extend_from_slice(&(OpCode::Halt as u16).to_le_bytes());
+
+        let mut vm = new_test_vm(rom.clone(), vec![]);
+        vm.program.functions = std::sync::Arc::from(vec![prometeu_bytecode::FunctionMeta {
+            code_offset: 0,
+            code_len: rom.len() as u32,
+            ..Default::default()
+        }]);
+
+        // Deterministic: trigger collection when any growth since last sweep occurs
+        vm.gc_alloc_threshold = 1;
+
+        // Allocate many small, unreferenced objects
+        let count = 2048usize; // stress but still quick
+        for i in 0..count {
+            let byte = (i & 0xFF) as u8;
+            let _ = vm.heap.allocate_object(ObjectKind::Bytes, &[byte]);
+        }
+        assert_eq!(vm.heap.len(), count);
+
+        let mut native = MockNative;
+        let mut ctx = HostContext::new(None);
+
+        // Single FRAME_SYNC should reclaim all since there are no roots
+        match vm.step(&mut native, &mut ctx) {
+            Err(LogicalFrameEndingReason::FrameSync) => {}
+            other => panic!("Expected FrameSync, got {:?}", other),
+        }
+
+        assert_eq!(vm.heap.len(), 0, "All short-lived objects must be reclaimed deterministically");
+    }
 }
diff --git a/files/TODOs.md b/files/TODOs.md
index b324b30d..fe3eefb4 100644
--- a/files/TODOs.md
+++ b/files/TODOs.md
@@ -1,47 +1,3 @@
-# PR-3.8 — GC Smoke and Stress Tests
-
-### Briefing
-
-We need confidence that the GC behaves correctly under simple and stressed conditions.
-
-### Target
-
-* Add deterministic smoke and stress tests for the GC.
-
-### Work items
-
-* Add tests:
-
-  * Simple allocation and collection cycle.
-  * Many short-lived objects.
-  * Cyclic references.
-* Ensure tests are deterministic.
-
-### Acceptance checklist
-
-* [ ] Smoke tests pass.
-* [ ] Stress tests pass.
-* [ ] No nondeterministic failures.
-* [ ] `cargo test` passes.
-
-### Tests
-
-* New GC-specific tests.
-
-### Junie instructions
-
-**You MAY:**
-
-* Add deterministic tests.
-
-**You MUST NOT:**
-
-* Introduce random or timing-dependent tests.
-* Modify GC semantics to satisfy tests.
-
-**If unclear:**
-
-* Ask before changing test scenarios.
 # PR-4.1 — Define Canonical Stack Effect Table for Core ISA
 
 ### Briefing
@@ -396,3 +352,301 @@ We need a stable suite of valid and invalid bytecode samples to ensure verifier
 
 * Ask before changing test expectations.
 
+---
+
+# PR-5.1 — Define Canonical Syscall Metadata Table
+
+### Briefing
+
+Syscalls must follow a unified, function-like ABI based on slot counts and capabilities. This PR introduces the canonical metadata table describing every syscall.
+
+### Target
+
+* Define a single authoritative metadata structure for syscalls.
+* Ensure all syscalls are described in terms of slot-based ABI.
+
+### Work items
+
+* Introduce a `SyscallMeta` struct containing:
+
+  * Syscall identifier.
+  * `arg_slots`.
+  * `ret_slots`.
+  * Capability flags.
+  * Determinism flags (if defined in spec).
+* Create a canonical syscall table/registry.
+* Ensure all existing syscalls are registered in this table.
+* Document the structure in code comments (English).
+
+### Acceptance checklist
+
+* [ ] All syscalls have a `SyscallMeta` entry.
+* [ ] Metadata includes arg and return slot counts.
+* [ ] No syscall is invoked without metadata.
+* [ ] `cargo test` passes.
+
+### Tests
+
+* Add a test ensuring all registered syscalls have metadata.
+
+### Junie instructions
+
+**You MAY:**
+
+* Introduce a syscall metadata struct.
+* Add a central registry for syscalls.
+
+**You MUST NOT:**
+
+* Change existing syscall semantics.
+* Add new syscalls.
+
+**If unclear:**
+
+* Ask before defining metadata fields.
+
+---
+
+# PR-5.2 — Implement Slot-Based Syscall Calling Convention
+
+### Briefing
+
+Syscalls must behave like functions using the stack-based slot ABI. This PR implements the unified calling convention.
+
+### Target
+
+* Ensure syscalls read arguments from the stack.
+* Ensure syscalls push return values based on `ret_slots`.
+
+### Work items
+
+* Modify the VM syscall dispatch logic to:
+
+  * Look up `SyscallMeta`.
+  * Pop `arg_slots` from the stack.
+  * Invoke the syscall.
+  * Push `ret_slots` results.
+* Remove any legacy argument handling paths.
+
+### Acceptance checklist
+
+* [ ] All syscalls use slot-based calling convention.
+* [ ] Argument and return slot counts are enforced.
+* [ ] No legacy calling paths remain.
+* [ ] `cargo test` passes.
+
+### Tests
+
+* Add tests:
+
+  * Syscall with correct arg/ret slots → passes.
+  * Syscall with insufficient args → trap or verifier failure.
+
+### Junie instructions
+
+**You MAY:**
+
+* Refactor syscall dispatch code.
+* Use metadata to enforce slot counts.
+
+**You MUST NOT:**
+
+* Change the stack model.
+* Add compatibility layers.
+
+**If unclear:**
+
+* Ask before modifying dispatch semantics.
+
+---
+
+# PR-5.3 — Enforce Capability Checks in Syscall Dispatch
+
+### Briefing
+
+Syscalls must be capability-gated. The runtime must verify that the current program has permission to invoke each syscall.
+
+### Target
+
+* Enforce capability checks before syscall execution.
+
+### Work items
+
+* Extend `SyscallMeta` with capability requirements.
+* Add capability data to the VM or execution context.
+* In syscall dispatch:
+
+  * Check capabilities.
+  * If missing, trigger `TRAP_INVALID_SYSCALL` or equivalent.
+
+### Acceptance checklist
+
+* [ ] Capability checks occur before syscall execution.
+* [ ] Missing capability leads to deterministic trap.
+* [ ] Valid capabilities allow execution.
+* [ ] `cargo test` passes.
+
+### Tests
+
+* Add tests:
+
+  * Syscall without capability → trap.
+  * Syscall with capability → success.
+
+### Junie instructions
+
+**You MAY:**
+
+* Add capability checks in dispatch.
+* Extend metadata with capability fields.
+
+**You MUST NOT:**
+
+* Change trap semantics.
+* Introduce dynamic or nondeterministic permission systems.
+
+**If unclear:**
+
+* Ask before defining capability behavior.
+
+---
+
+# PR-5.4 — Verifier Integration for Syscall Slot Rules
+
+### Briefing
+
+The verifier must ensure that syscall calls respect argument and return slot counts before runtime.
+
+### Target
+
+* Extend verifier to validate syscall usage.
+
+### Work items
+
+* At syscall call sites:
+
+  * Look up `SyscallMeta`.
+  * Ensure enough argument slots are available.
+  * Ensure stack shape after call matches `ret_slots`.
+* Emit verifier errors for mismatches.
+
+### Acceptance checklist
+
+* [ ] Verifier rejects incorrect syscall slot usage.
+* [ ] Correct programs pass.
+* [ ] Runtime traps are not required for verifier-detectable cases.
+* [ ] `cargo test` passes.
+
+### Tests
+
+* Add tests:
+
+  * Too few args for syscall → verifier error.
+  * Correct args/returns → passes.
+
+### Junie instructions
+
+**You MAY:**
+
+* Extend verifier with syscall checks.
+
+**You MUST NOT:**
+
+* Change runtime trap logic.
+* Add new trap categories.
+
+**If unclear:**
+
+* Ask before enforcing slot rules.
+
+---
+
+# PR-5.5 — Remove Legacy Syscall Entry Paths
+
+### Briefing
+
+Any old or experimental syscall entry paths must be removed so that the slot-based ABI is the only supported mechanism.
+
+### Target
+
+* Ensure only the new unified syscall dispatch path exists.
+
+### Work items
+
+* Search for legacy or alternate syscall invocation logic.
+* Remove or refactor them to use the canonical dispatch.
+* Update modules and exports accordingly.
+
+### Acceptance checklist
+
+* [ ] Only one syscall dispatch path remains.
+* [ ] No legacy syscall logic is present.
+* [ ] `cargo test` passes.
+
+### Tests
+
+* Existing tests only.
+
+### Junie instructions
+
+**You MAY:**
+
+* Remove legacy syscall code paths.
+* Refactor callers to use the unified dispatch.
+
+**You MUST NOT:**
+
+* Introduce new syscall semantics.
+* Keep compatibility shims.
+
+**If unclear:**
+
+* Ask before deleting anything that looks externally visible.
+
+---
+
+# PR-5.6 — Syscall Multi-Return Tests
+
+### Briefing
+
+We must ensure multi-return syscalls behave correctly with the slot-based ABI.
+
+### Target
+
+* Add deterministic tests covering multi-return behavior.
+
+### Work items
+
+* Create or adapt at least one syscall with `ret_slots > 1`.
+* Add tests:
+
+  * Verify correct stack results after syscall.
+  * Verify incorrect caller expectations fail verification.
+
+### Acceptance checklist
+
+* [ ] Multi-return syscalls behave correctly.
+* [ ] Verifier catches mismatches.
+* [ ] `cargo test` passes.
+
+### Tests
+
+* New multi-return syscall tests.
+
+### Junie instructions
+
+**You MAY:**
+
+* Add deterministic tests.
+* Use existing syscalls or create a simple test-only syscall.
+
+**You MUST NOT:**
+
+* Modify syscall semantics to satisfy tests.
+* Add nondeterministic behavior.
+
+**If unclear:**
+
+* Ask before introducing new test syscalls.
+
+