diff --git a/capacity_contract_test.go b/capacity_contract_test.go
index b4f0bea..2a57370 100644
--- a/capacity_contract_test.go
+++ b/capacity_contract_test.go
@@ -3,8 +3,6 @@ package promptkit_test
 import (
 	"context"
 	"errors"
-	"runtime"
-	"strings"
 	"sync"
 	"testing"
 	"time"
@@ -172,46 +170,6 @@ func TestUnlimitedBackendsRetainInjectedClientConcurrency(t *testing.T) {
 	}
 }
 
-func TestCancelWhileWaitingForGenerationCapacityPreservesErrorAndCapacity(t *testing.T) {
-	release := make(chan struct{})
-	client := newCapacityGateClient(release, 3)
-	engine := newBackendCapacityEngine(t, client, 1, capacityInt(1), nil)
-
-	firstResult := make(chan capacityRunResult, 1)
-	go runCapacityRequest(engine, context.Background(), promptkit.RunRequest{PromptID: "prompt"}, firstResult)
-	awaitCapacityRequest(t, client.started)
-
-	waitingContext := newObservedCancelContext()
-	secondResult := make(chan capacityRunResult, 1)
-	go runCapacityRequest(engine, waitingContext, promptkit.RunRequest{PromptID: "prompt"}, secondResult)
-	awaitCapacitySignal(t, waitingContext.waiting, "generation-capacity wait")
-	waitingContext.cancel()
-
-	outcome := awaitCapacityRun(t, secondResult)
-	if outcome.result != nil {
-		t.Fatalf("canceled capacity wait returned partial result: %+v", outcome.result)
-	}
-	if !errors.Is(outcome.err, context.Canceled) || !errors.Is(outcome.err, promptkit.ErrLLMGenerate) {
-		t.Fatalf("canceled capacity wait=%v, want context.Canceled and ErrLLMGenerate", outcome.err)
-	}
-	if _, _, calls := client.snapshot(); calls != 1 {
-		t.Fatalf("client calls=%d after canceled waiter, want only active call", calls)
-	}
-
-	close(release)
-	first := awaitCapacityRun(t, firstResult)
-	if first.err != nil || first.result == nil {
-		t.Fatalf("first run=(%+v, %v), want success", first.result, first.err)
-	}
-	result, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt"})
-	if err != nil || result == nil {
-		t.Fatalf("run after cancellation=(%+v, %v), want reusable capacity", result, err)
-	}
-	if _, _, calls := client.snapshot(); calls != 2 {
-		t.Fatalf("client calls=%d after reuse, want 2", calls)
-	}
-}
-
 func TestCapacityExceededSentinelContract(t *testing.T) {
 	if promptkit.ErrCapacityExceeded == nil {
 		t.Fatal("ErrCapacityExceeded is nil")
@@ -366,61 +324,6 @@ func (r *capacityArtifactReader) callCount() int {
 	return r.calls
 }
 
-type observedCancelContext struct {
-	done    chan struct{}
-	waiting chan struct{}
-	once    sync.Once
-}
-
-func newObservedCancelContext() *observedCancelContext {
-	return &observedCancelContext{
-		done:    make(chan struct{}),
-		waiting: make(chan struct{}, 1),
-	}
-}
-
-func (c *observedCancelContext) Deadline() (time.Time, bool) {
-	return time.Time{}, false
-}
-
-func (c *observedCancelContext) Done() <-chan struct{} {
-	var callers [8]uintptr
-	frames := runtime.CallersFrames(callers[:runtime.Callers(2, callers[:])])
-	for {
-		frame, more := frames.Next()
-		if strings.Contains(frame.Function, "internal/capacity.(*pool).acquire") {
-			select {
-			case c.waiting <- struct{}{}:
-			default:
-			}
-			break
-		}
-		if !more {
-			break
-		}
-	}
-	return c.done
-}
-
-func (c *observedCancelContext) Err() error {
-	select {
-	case <-c.done:
-		return context.Canceled
-	default:
-		return nil
-	}
-}
-
-func (c *observedCancelContext) Value(any) any {
-	return nil
-}
-
-func (c *observedCancelContext) cancel() {
-	c.once.Do(func() {
-		close(c.done)
-	})
-}
-
 func awaitCapacityRequest(
 	t *testing.T,
 	requests <-chan promptkit.GenerateRequest,
diff --git a/docs/internal/capacity.md b/docs/internal/capacity.md
index 35d6f37..1556806 100644
--- a/docs/internal/capacity.md
+++ b/docs/internal/capacity.md
@@ -94,4 +94,7 @@ The [runner tests](../../internal/usecase/runner_test.go) own early admission,
 lease lifetime, failure release, and shared initial/repair scheduling. The
 [external package capacity tests](../../capacity_contract_test.go) own the
 assembled public-engine behavior for configured limits, capacity errors,
-endpoint identity, engine independence, cancellation, and injected clients.
+endpoint identity, engine independence, and injected clients. The
+[root error-boundary tests](../../errors_internal_test.go) own preservation of
+the public generation category and context identity when generation is
+canceled.
diff --git a/errors_internal_test.go b/errors_internal_test.go
new file mode 100644
index 0000000..e3e1be6
--- /dev/null
+++ b/errors_internal_test.go
@@ -0,0 +1,22 @@
+package promptkit
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/usecase"
+)
+
+func TestMapPublicErrorPreservesGenerationCancellation(t *testing.T) {
+	internalErr := fmt.Errorf("%w: %w", usecase.ErrLLMGenerate, context.Canceled)
+
+	err := mapPublicError(internalErr)
+	if !errors.Is(err, ErrLLMGenerate) {
+		t.Fatalf("mapped error=%v, want ErrLLMGenerate", err)
+	}
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("mapped error=%v, want context.Canceled", err)
+	}
+}