diff --git a/runtime/src/cabi/flatten.ts b/runtime/src/cabi/flatten.ts
index 2f11e00..ae8d6be 100644
--- a/runtime/src/cabi/flatten.ts
+++ b/runtime/src/cabi/flatten.ts
@@ -18,7 +18,9 @@ export const MAX_FLAT_PARAMS = 16;
 export const MAX_FLAT_ASYNC_PARAMS = 4;
 // Mutable to mirror run_tests.py toggling definitions.MAX_FLAT_RESULTS.
 export let MAX_FLAT_RESULTS = 1;
-export function setMaxFlatResults(n: number): number {
+// Test-only mirror of run_tests.py's constant toggling; not for production
+// use (naming convention: see `schedulerSeedForTesting`).
+export function setMaxFlatResultsForTesting(n: number): number {
   const prev = MAX_FLAT_RESULTS;
   MAX_FLAT_RESULTS = n;
   return prev;
diff --git a/runtime/src/cabi/handles.ts b/runtime/src/cabi/handles.ts
index b490d12..b7bbec0 100644
--- a/runtime/src/cabi/handles.ts
+++ b/runtime/src/cabi/handles.ts
@@ -191,9 +191,9 @@ function isThenable(v: unknown): v is PromiseLike<unknown> {
 /**
  * Invoke a resource destructor through the reference's entry bracket.
  *
- * definitions.py `canon_resource_drop` (line 2318) does not call `rt.dtor`
+ * definitions.py `canon_resource_drop` (line 2319) does not call `rt.dtor`
  * directly. It builds the dtor into a function instance and calls it through
- * `Store.lift` / `Store.lower` (lines 2326-2333):
+ * `Store.lift` / `Store.lower` (lines 2330-2333):
  *
  * ```python
  *   dtor = rt.dtor or (lambda rep: [])
@@ -310,7 +310,7 @@ export function canonResourceDrop(
   trapIf(rh.numLends !== 0, "handle still lent out");
   if (rh.own) {
     assert_(rh.borrowScope === null);
-    // definitions.py line 2325-2333: the dtor runs through the store's
+    // definitions.py line 2326-2333: the dtor runs through the store's
     // lift/lower bracket. SCOPE NOTE (#85): the call below is a JS frame
     // inside the drop trampoline, so a *guest*-initiated drop whose dtor
     // suspends traps under the JSPI frame rule. That is deterministic and
diff --git a/runtime/src/cabi/strings.ts b/runtime/src/cabi/strings.ts
index b9b50c6..852f6f5 100644
--- a/runtime/src/cabi/strings.ts
+++ b/runtime/src/cabi/strings.ts
@@ -232,16 +232,19 @@ export function storeString(
 ): void {
   const mem = requireMemory(cx.opts);
   const [begin, taggedCodeUnits] = storeStringIntoRange(cx, v);
+  // Write order matches the reference (store_string, definitions.py:1613-1616):
+  // begin pointer first, then tagged length. Unobservable here (no trap can
+  // intervene between the two writes), but kept in step for parity.
   storeInt(
     mem,
-    mem.ptrSize() === 4 ? Number(taggedCodeUnits) : taggedCodeUnits,
-    ptr + mem.ptrSize(),
+    mem.ptrSize() === 4 ? begin : BigInt(begin),
+    ptr,
     mem.ptrSize(),
   );
   storeInt(
     mem,
-    mem.ptrSize() === 4 ? begin : BigInt(begin),
-    ptr,
+    mem.ptrSize() === 4 ? Number(taggedCodeUnits) : taggedCodeUnits,
+    ptr + mem.ptrSize(),
     mem.ptrSize(),
   );
 }
diff --git a/runtime/src/exec/boundary.ts b/runtime/src/exec/boundary.ts
index faaa3a9..f5f21f9 100644
--- a/runtime/src/exec/boundary.ts
+++ b/runtime/src/exec/boundary.ts
@@ -1847,6 +1847,12 @@ const CALLBACK_CODE_MAX = 2;
 export function unpackCallbackResult(
   packed: number,
 ): [code: CallbackCode, waitableSetIndex: number] {
+  // Reference parity insurance only: callers already guarantee this range via
+  // core-result normalization before calling in.
+  assert_(
+    packed >= 0 && packed < 2 ** 32,
+    `unpack-callback-result: packed out of range: ${packed}`,
+  );
   const code = packed & 0xf;
   trapIf(code > CALLBACK_CODE_MAX, `invalid callback code ${code}`);
   return [code as CallbackCode, packed >>> 4];
diff --git a/runtime/src/intrinsics/async_builtins.ts b/runtime/src/intrinsics/async_builtins.ts
index cc8520c..b3b5a31 100644
--- a/runtime/src/intrinsics/async_builtins.ts
+++ b/runtime/src/intrinsics/async_builtins.ts
@@ -501,7 +501,8 @@ export function createSubtaskCancel(
             // reports the resolved state through the same tail as the
             // non-blocking path. Mirrors SITE 4 (stream_builtins.ts:305-323)
             // and `Waitable.waitForPendingEvent` (definitions.py:786-790,
-            // reached from canon_subtask_cancel :2491): `hasSyncWaiter` must
+            // reached from canon_subtask_cancel's `subtask.wait_for_pending_event()`
+            // call, :2484): `hasSyncWaiter` must
             // be set for the duration so a concurrent `waitable.join` on
             // this subtask traps (async_builtins.ts:362-365) instead of
             // racing the SUBTASK event away from this resume (#87).
diff --git a/runtime/src/intrinsics/context.ts b/runtime/src/intrinsics/context.ts
index 9173dd0..32e0996 100644
--- a/runtime/src/intrinsics/context.ts
+++ b/runtime/src/intrinsics/context.ts
@@ -19,7 +19,7 @@ import { UnsupportedFeatureError } from "./errors.ts";
 
 /**
  * Number of `i32` context slots per thread. definitions.py `Thread.storage`
- * is initialised `[0,0]` (line 346) and `canon_context_{get,set}` assert
+ * is initialised `[0,0]` (line 347) and `canon_context_{get,set}` assert
  * `i < len(thread.storage)` — so exactly two, matching the intrinsic names
  * `context-*-i32-0` and `context-*-i32-1`.
  */
diff --git a/runtime/src/intrinsics/fact_calls.ts b/runtime/src/intrinsics/fact_calls.ts
index 446d5df..524807f 100644
--- a/runtime/src/intrinsics/fact_calls.ts
+++ b/runtime/src/intrinsics/fact_calls.ts
@@ -644,7 +644,13 @@ export function createSyncStartCall(
       mode: ctx.suspensionMode,
       canBlock: ctx.calleeCanBlock?.(callee) ?? false,
       onCallerResults: (r) => {
-        callerResults = r ?? [];
+        // sync-start-call's callee always uses the async ABI (comment above),
+        // but the *caller* side here is the sync `canon_lower` path: the
+        // reference's on_resolve(None) case is reached only when a
+        // cancellation was requested, and a sync-lowered subtask has no
+        // handle and hence no cancel channel — so `r` can never be null here.
+        assert_(r !== null, "sync-start-call: caller results missing");
+        callerResults = r;
       },
       lenderScope,
     });
diff --git a/runtime/src/intrinsics/mod.ts b/runtime/src/intrinsics/mod.ts
index 501ec32..ef92640 100644
--- a/runtime/src/intrinsics/mod.ts
+++ b/runtime/src/intrinsics/mod.ts
@@ -203,7 +203,7 @@ export class SyncCallScope {
     this.lenders.push(h);
   }
 
-  /** definitions.py `Subtask.release_lenders`. */
+  /** definitions.py `Subtask.deliver_resolve` (lines 902-906): releases lenders at delivery time. */
   releaseLenders(): void {
     for (const h of this.lenders) h.numLends -= 1;
     this.lenders.length = 0;
diff --git a/runtime/src/intrinsics/stream_builtins.ts b/runtime/src/intrinsics/stream_builtins.ts
index 48684b7..58634da 100644
--- a/runtime/src/intrinsics/stream_builtins.ts
+++ b/runtime/src/intrinsics/stream_builtins.ts
@@ -171,6 +171,12 @@ function streamCopy(input: {
   const cx = new LiftLowerContext(cabiOptions(opts), inst, null);
   const buffer = new GuestBuffer(elem, cx, ptr, n);
 
+  // definitions.py `stream_copy`: `assert(not isinstance(stream_t, CharType))`
+  // — plan validation is expected to reject a char-typed stream before this
+  // point (streams of `char` are not a representable component type), so this
+  // documents that invariant rather than defending against a reachable case.
+  assert_(elem === null || elem.kind !== "char", "stream copy: char element");
+
   // definitions.py `stream_event`: the payload is computed at *delivery* time,
   // so `buffer.progress` reflects everything copied by the time the guest
   // looks — including copies that happened after the event was armed.
@@ -187,6 +193,12 @@ function streamCopy(input: {
       buffer.progress <= BUFFER_MAX_LENGTH,
       "stream progress out of packing range",
     );
+    // definitions.py `stream_copy`/`stream_event`: `assert(0 <= result < 2**4)`.
+    // `CopyResult` is a fixed 0..2 enum so this can't fire; kept for parity.
+    assert_(
+      result >= 0 && result < 2 ** 4,
+      "stream event: packed result out of 4-bit range",
+    );
     return [eventCode, i, (result | (buffer.progress << 4)) >>> 0];
   };
 
@@ -391,7 +403,7 @@ function takeCancelEvent(
   );
   // UPSTREAM DIVERGENCE (definitions.py is wrong here; wasmtime is right).
   //
-  // `cancel_copy` (definitions.py line 2645) returns an already-armed pending
+  // `cancel_copy` (definitions.py line 2654) returns an already-armed pending
   // event verbatim, so cancelling a stream write that had been partially
   // satisfied yields COMPLETED with the copied count. wasmtime instead
   // *supersedes* an undelivered stream COMPLETED with CANCELLED, keeping the
diff --git a/runtime/src/task/scheduler.ts b/runtime/src/task/scheduler.ts
index 88c8811..61f5655 100644
--- a/runtime/src/task/scheduler.ts
+++ b/runtime/src/task/scheduler.ts
@@ -744,7 +744,6 @@ export interface SchedulableThread {
  */
 export class Store {
   readonly waiting: SchedulableThread[] = [];
-  nestingDepth = 0;
 
   /**
    * Host-import promises this store is waiting on. Non-empty means progress
diff --git a/runtime/src/task/streams.ts b/runtime/src/task/streams.ts
index d1ff356..9688056 100644
--- a/runtime/src/task/streams.ts
+++ b/runtime/src/task/streams.ts
@@ -469,7 +469,8 @@ export class SharedFutureImpl implements SharedBase {
 
   read(inst: unknown, dstBuffer: GuestBuffer, onCopyDone: OnCopyDone): void {
     // #84 leg (c): the reader arrives AFTER the writable side was abandoned.
-    // definitions.py:1141 asserts `not self.dropped` here because the drop
+    // definitions.py:1154 (SharedFutureImpl.read) asserts `not self.dropped`
+    // here because the drop
     // trap keeps that unreachable; for our abandoned state the honest answer
     // is the same trap the parked reader gets, delivered synchronously.
     if (this.dropped && this.abandonReason !== null) {
diff --git a/runtime/src/task/subtask.ts b/runtime/src/task/subtask.ts
index 804fc5a..a62488b 100644
--- a/runtime/src/task/subtask.ts
+++ b/runtime/src/task/subtask.ts
@@ -144,10 +144,16 @@ export class Subtask extends Waitable {
 
   /**
    * definitions.py `canon_lower`'s `on_progress`/`subtask_event` closure
-   * (line 2296). The event payload is computed **at delivery time** and
+   * (lines 2297-2298). The event payload is computed **at delivery time** and
    * delivering it is what runs `deliver_resolve` — so the lent handles are
    * released exactly when the guest observes the resolution, not when it
    * happens.
+   *
+   * The `!this.resolveDelivered()` guard has no reference analogue: it exists
+   * so this can coexist with `unwindLenders()` (which may itself have already
+   * delivered the resolve on an abandoned path). The reference's
+   * `subtask_event` calls `deliver_resolve()` unconditionally and would
+   * assert on a double delivery.
    */
   setSubtaskPendingEvent(subtaski: number): void {
     this.setPendingEvent(() => {
@@ -159,7 +165,7 @@ export class Subtask extends Waitable {
 
 /**
  * Pack a `canon_lower` async return value: `state | (subtaski << 4)`
- * (definitions.py line 2308, with the accompanying asserts on the ranges).
+ * (definitions.py line 2306, with the accompanying asserts on the ranges).
  */
 export function packSubtaskResult(
   state: SubtaskState,
diff --git a/runtime/src/task/thread.ts b/runtime/src/task/thread.ts
index 96bbbdd..d254f5e 100644
--- a/runtime/src/task/thread.ts
+++ b/runtime/src/task/thread.ts
@@ -42,6 +42,10 @@ export class Thread implements SchedulableThread {
    * write *this*, not per-task state: two threads of the same task have
    * independent context. wit-bindgen 0.60 keeps its async task pointer in
    * slot 0.
+   *
+   * Slots are plain JS numbers: a `context.set` of an i64 value above
+   * 2^53-1 would lose precision. Moot while memory64/threads support is
+   * deferred (issue #12) — revisit this when that issue's closure lands.
    */
   readonly storage: number[] = [0, 0];