feat(auth): add explicit handling for count_tokens endpoint errors and enhance model availability logic

- Implemented logic to differentiate between generic and explicit model-not-found errors for the `count_tokens` endpoint. - Introduced `recordAvailabilityNeutralResult` to ensure models remain active for generic endpoint errors. - Expanded test coverage to verify behavior for various 404 scenarios, including error propagation and model suspension handling. - Refactored error-parsing utilities to identify structured and nested model-not-found errors. Closes: #4410
2026-09-07 00:24:17 +08:00 · 2026-07-19 03:39:27 +08:00
parent 0b2ce80fcb
commit 36ed0ca5ce
2 changed files with 630 additions and 6 deletions
--- a/sdk/cliproxy/auth/conductor.go
+++ b/sdk/cliproxy/auth/conductor.go
@@ -2733,7 +2733,15 @@ func (m *Manager) executeCountMixedOnce(ctx context.Context, providers []string,
 				if ra := retryAfterFromError(errExec); ra != nil {
 					result.RetryAfter = ra
 				}
-				m.MarkResult(execCtx, result)
+				// Some Anthropic-compatible upstreams do not implement the
+				// count_tokens route and return a generic endpoint 404. Record
+				// the failure for hooks and metrics without suspending a model
+				// that remains usable through the messages endpoint.
+				if isCountTokensEndpointNotFoundError(errExec, execReq.Model) {
+					m.recordAvailabilityNeutralResult(execCtx, result)
+				} else {
+					m.MarkResult(execCtx, result)
+				}
 				if isRequestInvalidError(errExec) {
 					return cliproxyexecutor.Response{}, errExec
 				}
@@ -3891,6 +3899,30 @@ func (m *Manager) MarkResult(ctx context.Context, result Result) {
 	m.publishErrorEvent(result, authSnapshot)
 }

+func (m *Manager) recordAvailabilityNeutralResult(ctx context.Context, result Result) {
+	if result.AuthID == "" {
+		return
+	}
+
+	var authSnapshot *Auth
+	m.mu.Lock()
+	if auth, ok := m.auths[result.AuthID]; ok && auth != nil {
+		now := time.Now()
+		auth.recordRecentRequest(now, result.Success)
+		if result.Success {
+			auth.Success++
+		} else {
+			auth.Failed++
+		}
+		_ = m.persist(ctx, auth)
+		authSnapshot = auth.Clone()
+	}
+	m.mu.Unlock()
+
+	m.hook.OnResult(ctx, result)
+	m.publishErrorEvent(result, authSnapshot)
+}
+
 func ensureModelState(auth *Auth, model string) *ModelState {
 	if auth == nil || model == "" {
 		return nil
@@ -4096,9 +4128,15 @@ func resultErrorFromError(err error) *Error {
 	if err == nil {
 		return nil
 	}
-	resultErr := &Error{
-		Message:    err.Error(),
-		HTTPStatus: statusCodeFromError(err),
+	var sourceErr *Error
+	var resultErr *Error
+	if errors.As(err, &sourceErr) && sourceErr != nil {
+		resultErr = cloneError(sourceErr)
+	} else {
+		resultErr = &Error{Message: err.Error()}
+	}
+	if resultErr.HTTPStatus == 0 {
+		resultErr.HTTPStatus = statusCodeFromError(err)
 	}
 	if isRequestScopedError(err) || isRequestInvalidError(err) {
 		resultErr.Code = requestScopedErrorCode
@@ -4296,6 +4334,199 @@ func isRequestScopedResultError(err *Error) bool {
 	return err != nil && (err.IsRequestScoped() || isRequestScopedNotFoundResultError(err))
 }

+func isCountTokensEndpointNotFoundError(err error, requestedModel string) bool {
+	if err == nil || statusCodeFromError(err) != http.StatusNotFound {
+		return false
+	}
+	baseModel := thinking.ParseSuffix(requestedModel).ModelName
+	return !isExplicitModelNotFoundError(err, baseModel)
+}
+
+func isExplicitModelNotFoundError(err error, requestedModel string) bool {
+	if err == nil {
+		return false
+	}
+	if authErr, ok := err.(*Error); ok && authErr != nil {
+		if isModelNotFoundIdentifier(authErr.Code) || isStructuredModelNotFoundError(authErr.Message, requestedModel) {
+			return true
+		}
+	} else if isStructuredModelNotFoundError(err.Error(), requestedModel) {
+		return true
+	}
+
+	switch wrapped := err.(type) {
+	case interface{ Unwrap() []error }:
+		for _, nested := range wrapped.Unwrap() {
+			if isExplicitModelNotFoundError(nested, requestedModel) {
+				return true
+			}
+		}
+	case interface{ Unwrap() error }:
+		return isExplicitModelNotFoundError(wrapped.Unwrap(), requestedModel)
+	}
+	return false
+}
+
+func isStructuredModelNotFoundError(message, requestedModel string) bool {
+	var payload any
+	if errJSON := json.Unmarshal([]byte(strings.TrimSpace(message)), &payload); errJSON != nil {
+		return false
+	}
+	return containsStructuredModelNotFound(payload, requestedModel)
+}
+
+func containsStructuredModelNotFound(value any, requestedModel string) bool {
+	switch typed := value.(type) {
+	case map[string]any:
+		notFoundType := false
+		exactModelReference := false
+		for key, item := range typed {
+			text, isString := item.(string)
+			if isString {
+				switch strings.ToLower(strings.TrimSpace(key)) {
+				case "code":
+					if isModelNotFoundIdentifier(text) {
+						return true
+					}
+				case "type":
+					if isModelNotFoundIdentifier(text) {
+						return true
+					}
+					notFoundType = notFoundType || isNotFoundErrorIdentifier(text)
+				case "error", "message", "detail", "error_description", "title":
+					if isExplicitModelNotFoundMessage(text, requestedModel) {
+						return true
+					}
+					exactModelReference = exactModelReference || isExactRequestedModelReference(text, requestedModel)
+				}
+			}
+			switch item.(type) {
+			case map[string]any, []any:
+				if containsStructuredModelNotFound(item, requestedModel) {
+					return true
+				}
+			}
+		}
+		return notFoundType && exactModelReference
+	case []any:
+		for _, item := range typed {
+			if text, isString := item.(string); isString && isExplicitModelNotFoundMessage(text, requestedModel) {
+				return true
+			}
+			if containsStructuredModelNotFound(item, requestedModel) {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+func isModelNotFoundIdentifier(value string) bool {
+	candidate := strings.ToLower(strings.TrimSpace(value))
+	if fragment := strings.LastIndex(candidate, "#"); fragment >= 0 && fragment+1 < len(candidate) {
+		candidate = candidate[fragment+1:]
+	} else {
+		if query := strings.Index(candidate, "?"); query >= 0 {
+			candidate = candidate[:query]
+		}
+		candidate = strings.TrimRight(candidate, "/")
+		if separator := strings.LastIndexAny(candidate, "/:"); separator >= 0 {
+			candidate = candidate[separator+1:]
+		}
+	}
+	normalized := strings.NewReplacer("-", "_", " ", "_").Replace(candidate)
+	switch normalized {
+	case "model_not_found", "model_not_found_error", "unknown_model", "model_does_not_exist", "model_not_exist":
+		return true
+	default:
+		return false
+	}
+}
+
+func isNotFoundErrorIdentifier(value string) bool {
+	normalized := strings.NewReplacer("-", "_", " ", "_").Replace(strings.ToLower(strings.TrimSpace(value)))
+	return normalized == "not_found" || normalized == "not_found_error"
+}
+
+func isExplicitModelNotFoundMessage(message, requestedModel string) bool {
+	lower := strings.Trim(strings.ToLower(strings.TrimSpace(message)), " .!;\t\r\n")
+	if lower == "" {
+		return false
+	}
+	normalized := strings.NewReplacer("-", "_", " ", "_").Replace(lower)
+	if strings.Contains(normalized, "model_not_found") || strings.Contains(normalized, "unknown_model") {
+		return true
+	}
+	for _, prefix := range []string{"no such model", "unknown model"} {
+		if lower != prefix && !strings.HasPrefix(lower, prefix+" ") && !strings.HasPrefix(lower, prefix+":") {
+			continue
+		}
+		remainder := strings.TrimSpace(strings.TrimPrefix(lower, prefix))
+		remainder = strings.TrimSpace(strings.TrimPrefix(remainder, ":"))
+		if remainder == "" {
+			return true
+		}
+		missingSuffix, matches := trimRequestedModelReference(remainder, requestedModel)
+		return matches && missingSuffix == ""
+	}
+	for _, prefix := range []string{"the requested model", "requested model", "the model", "model"} {
+		if lower != prefix && !strings.HasPrefix(lower, prefix+" ") && !strings.HasPrefix(lower, prefix+":") {
+			continue
+		}
+		remainder := strings.TrimSpace(strings.TrimPrefix(lower, prefix))
+		remainder = strings.TrimSpace(strings.TrimPrefix(remainder, ":"))
+		if isMissingModelPhrase(remainder) {
+			return true
+		}
+		missingSuffix, matches := trimRequestedModelReference(remainder, requestedModel)
+		return matches && isMissingModelPhrase(missingSuffix)
+	}
+	return false
+}
+
+func isExactRequestedModelReference(message, requestedModel string) bool {
+	lower := strings.Trim(strings.ToLower(strings.TrimSpace(message)), " .!;\t\r\n")
+	for _, prefix := range []string{"the requested model", "requested model", "the model", "model"} {
+		if lower != prefix && !strings.HasPrefix(lower, prefix+" ") && !strings.HasPrefix(lower, prefix+":") {
+			continue
+		}
+		remainder := strings.TrimSpace(strings.TrimPrefix(lower, prefix))
+		remainder = strings.TrimSpace(strings.TrimPrefix(remainder, ":"))
+		suffix, matches := trimRequestedModelReference(remainder, requestedModel)
+		return matches && suffix == ""
+	}
+	return false
+}
+
+func trimRequestedModelReference(value, requestedModel string) (string, bool) {
+	model := strings.ToLower(strings.TrimSpace(requestedModel))
+	if model == "" {
+		return "", false
+	}
+	for _, candidate := range []string{model, "'" + model + "'", `"` + model + `"`, "`" + model + "`"} {
+		if value == candidate {
+			return "", true
+		}
+		if !strings.HasPrefix(value, candidate) {
+			continue
+		}
+		remainder := value[len(candidate):]
+		if remainder == "" || strings.ContainsRune(" :,", rune(remainder[0])) {
+			return strings.TrimLeft(remainder, " :,"), true
+		}
+	}
+	return "", false
+}
+
+func isMissingModelPhrase(value string) bool {
+	switch strings.Trim(value, " .!;\t\r\n") {
+	case "not found", "was not found", "could not be found", "does not exist", "doesn't exist", "not exist", "is unknown":
+		return true
+	default:
+		return false
+	}
+}
+
 // isRequestInvalidError returns true if the error represents a client request
 // error that should not be retried. Specifically, it treats 400 responses with
 // "invalid_request_error", request-scoped 404 item misses caused by `store=false`,
--- a/sdk/cliproxy/auth/conductor_overrides_test.go
+++ b/sdk/cliproxy/auth/conductor_overrides_test.go
@@ -2,6 +2,8 @@ package auth

 import (
 	"context"
+	"errors"
+	"fmt"
 	"net/http"
 	"sync"
 	"testing"
@@ -162,6 +164,7 @@ type authFallbackExecutor struct {
 	streamCalls       []string
 	executeErrors     map[string]error
 	streamFirstErrors map[string]error
+	countTokenErrors  map[string]error
 }

 func (e *authFallbackExecutor) Identifier() string {
@@ -200,8 +203,14 @@ func (e *authFallbackExecutor) Refresh(_ context.Context, auth *Auth) (*Auth, er
 	return auth, nil
 }

-func (e *authFallbackExecutor) CountTokens(context.Context, *Auth, cliproxyexecutor.Request, cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
-	return cliproxyexecutor.Response{}, &Error{HTTPStatus: 500, Message: "not implemented"}
+func (e *authFallbackExecutor) CountTokens(_ context.Context, auth *Auth, _ cliproxyexecutor.Request, _ cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
+	e.mu.Lock()
+	err := e.countTokenErrors[auth.ID]
+	e.mu.Unlock()
+	if err != nil {
+		return cliproxyexecutor.Response{}, err
+	}
+	return cliproxyexecutor.Response{Payload: []byte(auth.ID)}, nil
 }

 func (e *authFallbackExecutor) HttpRequest(context.Context, *Auth, *http.Request) (*http.Response, error) {
@@ -224,6 +233,27 @@ func (e *authFallbackExecutor) StreamCalls() []string {
 	return out
 }

+type resultCaptureHook struct {
+	NoopHook
+
+	mu      sync.Mutex
+	results []Result
+}
+
+func (h *resultCaptureHook) OnResult(_ context.Context, result Result) {
+	h.mu.Lock()
+	h.results = append(h.results, result)
+	h.mu.Unlock()
+}
+
+func (h *resultCaptureHook) Results() []Result {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	out := make([]Result, len(h.results))
+	copy(out, h.results)
+	return out
+}
+
 type retryAfterStatusError struct {
 	status     int
 	message    string
@@ -1269,6 +1299,369 @@ func TestManager_MarkResult_RequestScopedNotFoundDoesNotCooldownAuth(t *testing.
 	}
 }

+func TestManager_ExecuteCount_GenericRouteNotFoundDoesNotSuspendModel(t *testing.T) {
+	previous := quotaCooldownDisabled.Load()
+	quotaCooldownDisabled.Store(false)
+	t.Cleanup(func() { quotaCooldownDisabled.Store(previous) })
+
+	hook := &resultCaptureHook{}
+	m := NewManager(nil, nil, hook)
+	executor := &authFallbackExecutor{
+		id: "claude",
+		countTokenErrors: map[string]error{
+			"count-route-not-found-auth": &Error{
+				HTTPStatus: http.StatusNotFound,
+				Message:    "404 page not found",
+			},
+		},
+	}
+	m.RegisterExecutor(executor)
+
+	model := "count-route-not-found-model"
+	auth := &Auth{ID: "count-route-not-found-auth", Provider: "claude"}
+	reg := registry.GetGlobalRegistry()
+	reg.RegisterClient(auth.ID, auth.Provider, []*registry.ModelInfo{{ID: model}})
+	t.Cleanup(func() { reg.UnregisterClient(auth.ID) })
+
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	if _, errCount := m.ExecuteCount(context.Background(), []string{"claude"}, cliproxyexecutor.Request{Model: model}, cliproxyexecutor.Options{}); errCount == nil {
+		t.Fatal("expected count_tokens route 404 error")
+	}
+
+	updated, ok := m.GetByID(auth.ID)
+	if !ok || updated == nil {
+		t.Fatal("expected auth to remain registered")
+	}
+	if updated.Failed != 1 {
+		t.Fatalf("failed request count = %d, want 1", updated.Failed)
+	}
+	results := hook.Results()
+	if len(results) != 1 || results[0].Success || results[0].Error == nil || results[0].Error.HTTPStatus != http.StatusNotFound {
+		t.Fatalf("recorded results = %#v, want one failed 404", results)
+	}
+	if updated.Unavailable {
+		t.Fatal("expected route 404 to keep auth available")
+	}
+	if state := updated.ModelStates[model]; state != nil {
+		t.Fatalf("expected route 404 to avoid model cooldown state, got %#v", state)
+	}
+	if count := reg.GetModelCount(model); count != 1 {
+		t.Fatalf("available model count = %d, want 1", count)
+	}
+
+	resp, errExecute := m.Execute(context.Background(), []string{"claude"}, cliproxyexecutor.Request{Model: model}, cliproxyexecutor.Options{})
+	if errExecute != nil {
+		t.Fatalf("execute after count_tokens route 404: %v", errExecute)
+	}
+	if string(resp.Payload) != auth.ID {
+		t.Fatalf("execute payload = %q, want %q", string(resp.Payload), auth.ID)
+	}
+}
+
+func TestManager_ExecuteCount_ExplicitModelNotFoundSuspendsModel(t *testing.T) {
+	previous := quotaCooldownDisabled.Load()
+	quotaCooldownDisabled.Store(false)
+	t.Cleanup(func() { quotaCooldownDisabled.Store(previous) })
+
+	hook := &resultCaptureHook{}
+	m := NewManager(nil, nil, hook)
+	executor := &authFallbackExecutor{
+		id: "claude",
+		countTokenErrors: map[string]error{
+			"count-model-not-found-auth": &Error{
+				Code:       "model_not_found",
+				HTTPStatus: http.StatusNotFound,
+				Message:    `{"type":"error","error":{"type":"not_found_error","message":"model count-explicitly-missing-model was not found"}}`,
+			},
+		},
+	}
+	m.RegisterExecutor(executor)
+
+	model := "count-explicitly-missing-model"
+	auth := &Auth{ID: "count-model-not-found-auth", Provider: "claude"}
+	reg := registry.GetGlobalRegistry()
+	reg.RegisterClient(auth.ID, auth.Provider, []*registry.ModelInfo{{ID: model}})
+	t.Cleanup(func() { reg.UnregisterClient(auth.ID) })
+
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	if _, errCount := m.ExecuteCount(context.Background(), []string{"claude"}, cliproxyexecutor.Request{Model: model}, cliproxyexecutor.Options{}); errCount == nil {
+		t.Fatal("expected count_tokens model-not-found error")
+	}
+
+	updated, ok := m.GetByID(auth.ID)
+	if !ok || updated == nil {
+		t.Fatal("expected auth to remain registered")
+	}
+	state := updated.ModelStates[model]
+	if state == nil || !state.Unavailable {
+		t.Fatalf("expected model-not-found cooldown state, got %#v", state)
+	}
+	if state.LastError == nil || state.LastError.Code != "model_not_found" {
+		t.Fatalf("model state error = %#v, want preserved model_not_found code", state.LastError)
+	}
+	results := hook.Results()
+	if len(results) != 1 || results[0].Error == nil || results[0].Error.Code != "model_not_found" {
+		t.Fatalf("hook results = %#v, want preserved model_not_found code", results)
+	}
+	remaining := time.Until(state.NextRetryAfter)
+	if remaining < 11*time.Hour || remaining > 12*time.Hour {
+		t.Fatalf("model-not-found cooldown = %v, want about 12h", remaining)
+	}
+	if count := reg.GetModelCount(model); count != 0 {
+		t.Fatalf("available model count = %d, want 0", count)
+	}
+}
+
+func TestIsCountTokensEndpointNotFoundError(t *testing.T) {
+	tests := []struct {
+		name  string
+		err   error
+		model string
+		want  bool
+	}{
+		{
+			name: "empty router 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound},
+			want: true,
+		},
+		{
+			name: "plain router 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: "404 page not found"},
+			want: true,
+		},
+		{
+			name: "wrapped router 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: "upstream request failed: 404 page not found"},
+			want: true,
+		},
+		{
+			name: "fastapi route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"detail":"Not Found"}`},
+			want: true,
+		},
+		{
+			name: "problem details route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"title":"Not Found","status":404}`},
+			want: true,
+		},
+		{
+			name: "nested generic route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"Not Found"}}`},
+			want: true,
+		},
+		{
+			name: "generic model api route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"type":"not_found_error","title":"Model API","detail":"Not Found"}`},
+			want: true,
+		},
+		{
+			name: "generic model metadata route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"model metadata route not found"}}`},
+			want: true,
+		},
+		{
+			name: "generic model provider 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"model provider was not found"}}`},
+			want: true,
+		},
+		{
+			name: "generic route with misleading metadata",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"message":"Not Found","request_id":"model_not_found"}`},
+			want: true,
+		},
+		{
+			name: "express count route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: "Cannot POST /v1/messages/count_tokens"},
+			want: true,
+		},
+		{
+			name: "html route 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: "<html><title>404 Not Found</title></html>"},
+			want: true,
+		},
+		{
+			name: "structured model 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"model claude-missing was not found"}}`},
+			want: false,
+		},
+		{
+			name: "anthropic exact model reference",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"model: claude-missing"}}`},
+			want: false,
+		},
+		{
+			name:  "anthropic model reference with thinking suffix",
+			err:   &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"model: claude-missing"}}`},
+			model: "claude-missing(high)",
+			want:  false,
+		},
+		{
+			name: "requested model does not exist",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"The requested model does not exist"}}`},
+			want: false,
+		},
+		{
+			name:  "requested quoted model could not be found",
+			err:   &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":{"type":"not_found_error","message":"The requested model 'foo' could not be found"}}`},
+			model: "foo",
+			want:  false,
+		},
+		{
+			name: "problem details model type uri",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"type":"https://example.com/problems/model-not-found","title":"Not Found","status":404}`},
+			want: false,
+		},
+		{
+			name: "structured model error string",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"error":"model claude-missing does not exist"}`},
+			want: false,
+		},
+		{
+			name: "model code with generic message",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"message":"Not Found","code":"model_not_found","model":"claude-missing"}`},
+			want: false,
+		},
+		{
+			name: "typed model not found code",
+			err:  &Error{Code: "model_not_found", HTTPStatus: http.StatusNotFound, Message: "Not Found"},
+			want: false,
+		},
+		{
+			name: "typed wrapper with structured model code",
+			err:  &Error{Code: "not_found", HTTPStatus: http.StatusNotFound, Message: `{"error":{"code":"model_not_found","message":"Not Found"}}`},
+			want: false,
+		},
+		{
+			name: "wrapped structured model code",
+			err: fmt.Errorf("upstream failed: %w", &requestScopedStatusError{
+				status:  http.StatusNotFound,
+				message: `{"error":{"code":"model_not_found","message":"Not Found"}}`,
+			}),
+			want: false,
+		},
+		{
+			name: "joined structured model code",
+			err: errors.Join(
+				errors.New("upstream failed"),
+				&requestScopedStatusError{
+					status:  http.StatusNotFound,
+					message: `{"error":{"code":"model_not_found","message":"Not Found"}}`,
+				},
+			),
+			want: false,
+		},
+		{
+			name: "outer generic inner model 404",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: `{"message":"Not Found","error":{"type":"not_found_error","message":"model claude-missing does not exist"}}`},
+			want: false,
+		},
+		{
+			name: "unstructured model text",
+			err:  &Error{HTTPStatus: http.StatusNotFound, Message: "model claude-missing was not found"},
+			want: true,
+		},
+		{
+			name: "non 404",
+			err:  &Error{HTTPStatus: http.StatusInternalServerError, Message: "404 page not found"},
+			want: false,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			model := tc.model
+			if model == "" {
+				model = "claude-missing"
+			}
+			if got := isCountTokensEndpointNotFoundError(tc.err, model); got != tc.want {
+				t.Fatalf("isCountTokensEndpointNotFoundError() = %v, want %v", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestManager_Execute_GenericRouteNotFoundStillSuspendsModel(t *testing.T) {
+	previous := quotaCooldownDisabled.Load()
+	quotaCooldownDisabled.Store(false)
+	t.Cleanup(func() { quotaCooldownDisabled.Store(previous) })
+
+	m := NewManager(nil, nil, nil)
+	executor := &authFallbackExecutor{
+		id: "claude",
+		executeErrors: map[string]error{
+			"messages-route-not-found-auth": &Error{
+				HTTPStatus: http.StatusNotFound,
+				Message:    "404 page not found",
+			},
+		},
+	}
+	m.RegisterExecutor(executor)
+
+	model := "messages-route-not-found-model"
+	auth := &Auth{ID: "messages-route-not-found-auth", Provider: "claude"}
+	reg := registry.GetGlobalRegistry()
+	reg.RegisterClient(auth.ID, auth.Provider, []*registry.ModelInfo{{ID: model}})
+	t.Cleanup(func() { reg.UnregisterClient(auth.ID) })
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	if _, errExecute := m.Execute(context.Background(), []string{"claude"}, cliproxyexecutor.Request{Model: model}, cliproxyexecutor.Options{}); errExecute == nil {
+		t.Fatal("expected messages route 404")
+	}
+
+	updated, ok := m.GetByID(auth.ID)
+	if !ok || updated == nil {
+		t.Fatal("expected auth to remain registered")
+	}
+	state := updated.ModelStates[model]
+	if state == nil || !state.Unavailable || state.NextRetryAfter.IsZero() {
+		t.Fatalf("expected ordinary messages 404 to suspend model, got %#v", state)
+	}
+}
+
+func TestManager_RecordResult_AvailabilityNeutralSkipsSchedulerUpdate(t *testing.T) {
+	m := NewManager(nil, nil, nil)
+	auth := &Auth{ID: "availability-neutral-auth", Provider: "claude"}
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	m.scheduler.mu.Lock()
+	provider := m.scheduler.providers[auth.Provider]
+	if provider == nil || provider.auths[auth.ID] == nil {
+		m.scheduler.mu.Unlock()
+		t.Fatal("expected scheduler auth metadata")
+	}
+	before := provider.auths[auth.ID].auth
+	m.scheduler.mu.Unlock()
+
+	m.recordAvailabilityNeutralResult(context.Background(), Result{
+		AuthID:   auth.ID,
+		Provider: auth.Provider,
+		Model:    "availability-neutral-model",
+		Success:  false,
+		Error:    &Error{HTTPStatus: http.StatusNotFound, Message: "404 page not found"},
+	})
+
+	updated, ok := m.GetByID(auth.ID)
+	if !ok || updated == nil || updated.Failed != 1 {
+		t.Fatalf("updated auth = %#v, want one recorded failure", updated)
+	}
+	m.scheduler.mu.Lock()
+	after := m.scheduler.providers[auth.Provider].auths[auth.ID].auth
+	m.scheduler.mu.Unlock()
+	if after != before {
+		t.Fatal("availability-neutral result unexpectedly replaced scheduler auth snapshot")
+	}
+}
+
 func TestManager_RequestScopedNotFoundStopsRetryWithoutSuspendingAuth(t *testing.T) {
 	m := NewManager(nil, nil, nil)
 	executor := &authFallbackExecutor{