Retire the legacy capacity client wrapper

docs/internal/capacity.md

@@ -68,9 +68,8 @@ Preparation and validation never hold an active permit.
 
 The manager's `Acquire` operation owns that active-permit acquisition and
 returns an idempotent release function. The coordinator uses it around
-individual attempts. The earlier one-shot wrapper still delegates to the same
+individual attempts. There is one queue and one active limit per limited backend
-operation for focused internal callers and tests, but engine assembly no
+pool.
-longer uses it. There is one queue and one active limit per limited backend pool.
 
 When all active permits are occupied, calls join a mutex-protected FIFO waiter
 list. Releasing a permit transfers it directly to the oldest remaining waiter
@@ -121,7 +120,7 @@ validation, bounded admission, idempotent release, context handling, and
 unlimited admission. The
 [client tests](../../internal/capacity/client_test.go) own peak enforcement,
 FIFO transfer, canceled-waiter removal, grant/cancel races, independent pools,
-unlimited calls, passthrough behavior, and panic release.
+unlimited calls, and panic release using test-only one-shot calls.
 
 The [runner tests](../../internal/usecase/runner_test.go) own ordinary early
 admission, lease lifetime, failure release, and shared initial/repair

docs/roadmap/implementation.md

@@ -826,6 +826,11 @@ focused public tests pass on Go 1.27.3; migration guidance notes hash/cache impa
 exercise the real scheduling path. No production scheduling policy or public
 API changes; no-repair runner construction and other deliberate fallbacks remain.
 
+Completion: initial and repair scheduling use the real injected coordinator;
+the production capacity wrapper was removed. Capacity tests retain one-shot
+test-only scaffolding for manager scenarios. Race-enabled capacity, generation,
+and use-case tests pass on Go 1.27.3, and internal capacity guidance is updated.
+
 ## Stage 33: Reconcile Documentation And Validate The End State
 
 **Catalog:** all entries; complete the sprint, not a release publication.

internal/capacity/client.go

@@ -1,40 +0,0 @@
-package capacity
-
-import (
-	"context"
-
-	"go.maximumdirect.net/promptkit/internal/domain"
-	"go.maximumdirect.net/promptkit/internal/llm"
-)
-
-type client struct {
-	manager *Manager
-	next    llm.Client
-}
-
-// NewClient wraps next with configured active-generation limits. A nil manager
-// leaves next unchanged.
-func NewClient(manager *Manager, next llm.Client) llm.Client {
-	if manager == nil {
-		return next
-	}
-	return &client{
-		manager: manager,
-		next:    next,
-	}
-}
-
-func (c *client) Generate(
-	ctx context.Context,
-	req domain.GenerateRequest,
-) (*domain.GenerateResponse, error) {
-	if c.manager.getPool(req.Target.BackendID) == nil {
-		return c.next.Generate(ctx, req)
-	}
-	release, err := c.manager.Acquire(ctx, req.Target.BackendID)
-	if err != nil {
-		return nil, err
-	}
-	defer release()
-	return c.next.Generate(ctx, req)
-}

internal/capacity/client_test.go

@@ -3,7 +3,6 @@ package capacity
 import (
 	"context"
 	"errors"
-	"reflect"
 	"runtime"
 	"sync"
 	"sync/atomic"
@@ -24,6 +23,18 @@ type clientFunc func(
 	domain.GenerateRequest,
 ) (*domain.GenerateResponse, error)
 
+// newTestClient exercises one acquisition around a controlled call.
+func newTestClient(manager *Manager, next llm.Client) llm.Client {
+	return clientFunc(func(ctx context.Context, req domain.GenerateRequest) (*domain.GenerateResponse, error) {
+		release, err := manager.Acquire(ctx, req.Target.BackendID)
+		if err != nil {
+			return nil, err
+		}
+		defer release()
+		return next.Generate(ctx, req)
+	})
+}
+
 func (f clientFunc) Generate(
 	ctx context.Context,
 	req domain.GenerateRequest,
@@ -167,7 +178,7 @@ func TestClientLimitsPeakConcurrencyAndServesWaitersFIFO(t *testing.T) {
 		"second": secondRelease,
 		"third":  thirdRelease,
 	})
-	client := NewClient(manager, next)
+	client := newTestClient(manager, next)
 
 	first := generateAsync(client, context.Background(), "limited", "first")
 	if got := receiveStarted(t, next.started); got != "first" {
@@ -212,7 +223,7 @@ func TestClientPeakConcurrencyDoesNotExceedConfiguredLimit(t *testing.T) {
 		releases[string(rune('a'+i))] = gate
 	}
 	next := newBlockingClient(releases)
-	client := NewClient(manager, next)
+	client := newTestClient(manager, next)
 
 	results := make([]<-chan generateResult, 0, len(releases))
 	for id := range releases {
@@ -257,7 +268,7 @@ func TestClientRemovesCanceledWaiters(t *testing.T) {
 				"three":  make(chan struct{}),
 			}
 			next := newBlockingClient(releases)
-			client := NewClient(manager, next)
+			client := newTestClient(manager, next)
 
 			holder := generateAsync(client, context.Background(), "limited", "holder")
 			if got := receiveStarted(t, next.started); got != "holder" {
@@ -321,7 +332,7 @@ func TestClientGrantCancellationRaceDoesNotLeakPermit(t *testing.T) {
 			}
 			return &domain.GenerateResponse{Content: req.Prompt.SessionID}, nil
 		})
-		client := NewClient(manager, next)
+		client := newTestClient(manager, next)
 
 		holder := generateAsync(client, context.Background(), "limited", "holder")
 		waitForActiveCount(t, manager, "limited", 1)
@@ -394,7 +405,7 @@ func TestClientUsesIndependentPoolsAndUnlimitedFastPaths(t *testing.T) {
 		"alpha": alphaRelease,
 		"beta":  betaRelease,
 	})
-	client := NewClient(manager, next)
+	client := newTestClient(manager, next)
 
 	alpha := generateAsync(client, context.Background(), "alpha", "alpha")
 	beta := generateAsync(client, context.Background(), "beta", "beta")
@@ -423,63 +434,6 @@ func TestClientUsesIndependentPoolsAndUnlimitedFastPaths(t *testing.T) {
 			t.Fatalf("unlimited backend %q response=(%#v, %v)", backendID, response, err)
 		}
 	}
-	if got := NewClient(nil, next); got != next {
-		t.Fatal("nil manager did not return the wrapped client unchanged")
-	}
-}
-
-func TestClientPreservesRequestsResponsesAndErrors(t *testing.T) {
-	manager := newTestManager(t, map[string]domain.BackendCapacityPolicy{
-		"limited": {ConcurrencyLimit: 1},
-	})
-	request := domain.GenerateRequest{
-		Prompt: domain.RenderedPrompt{
-			SessionID: "session",
-			Messages: []domain.RenderedMessage{
-				{Role: "user", Content: "content"},
-			},
-		},
-		Target: domain.ExecutionTarget{
-			BackendID:   "limited",
-			Model:       "model",
-			ExtraParams: map[string]any{"key": "value"},
-		},
-	}
-	response := &domain.GenerateResponse{
-		Content: "output",
-		Usage:   domain.TokenUsage{TotalTokens: 7},
-	}
-	collaboratorErr := errors.New("collaborator failure")
-	tests := []struct {
-		name     string
-		response *domain.GenerateResponse
-		err      error
-	}{
-		{name: "successful response", response: response},
-		{name: "nil response"},
-		{name: "collaborator error", response: response, err: collaboratorErr},
-	}
-
-	for _, tc := range tests {
-		t.Run(tc.name, func(t *testing.T) {
-			var captured domain.GenerateRequest
-			next := clientFunc(func(
-				_ context.Context,
-				req domain.GenerateRequest,
-			) (*domain.GenerateResponse, error) {
-				captured = req
-				return tc.response, tc.err
-			})
-			gotResponse, gotErr := NewClient(manager, next).Generate(context.Background(), request)
-			if !reflect.DeepEqual(captured, request) {
-				t.Fatalf("request changed: %#v", captured)
-			}
-			if gotResponse != tc.response || gotErr != tc.err {
-				t.Fatalf("response=(%p, %v), want (%p, %v)",
-					gotResponse, gotErr, tc.response, tc.err)
-			}
-		})
-	}
 }
 
 func TestClientReleasesPermitDuringPanicUnwinding(t *testing.T) {
@@ -496,7 +450,7 @@ func TestClientReleasesPermitDuringPanicUnwinding(t *testing.T) {
 		}
 		return &domain.GenerateResponse{Content: "recovered"}, nil
 	})
-	client := NewClient(manager, next)
+	client := newTestClient(manager, next)
 	request := domain.GenerateRequest{
 		Target: domain.ExecutionTarget{BackendID: "limited"},
 	}

internal/usecase/runner_test.go

@@ -18,6 +18,7 @@ import (
 	"go.maximumdirect.net/promptkit/internal/capacity"
 	"go.maximumdirect.net/promptkit/internal/defaults"
 	"go.maximumdirect.net/promptkit/internal/domain"
+	"go.maximumdirect.net/promptkit/internal/generation"
 	"go.maximumdirect.net/promptkit/internal/llm"
 	"go.maximumdirect.net/promptkit/internal/prompt"
 	"go.maximumdirect.net/promptkit/internal/validate"
@@ -2398,7 +2399,7 @@ func TestRunnerSchedulesInitialAndRepairGenerationThroughOneBackendPool(t *testi
 	baseClient := &controlledRepairLLM{
 		events: make(chan controlledGeneration, 4),
 	}
-	scheduledClient := capacity.NewClient(manager, baseClient)
+	scheduledClient := generation.NewInjected(manager, baseClient)
 	admitter := &recordingRunAdmitter{next: manager}
 	runner := NewRunnerWithRepairer(
 		staticPromptRepo{def: promptDef(domain.FormatJSON, domain.ValidationJSON, 1)},