feat: implement tiered billing expression evaluation and related functionality

- Added support for tiered billing expressions in the billing system.
- Introduced new types and functions for handling billing expressions, including caching and execution.
- Updated existing billing logic to accommodate tiered billing scenarios.
- Enhanced request handling to support incoming billing expression requests.
- Added tests for tiered billing functionality to ensure correctness.

service/billing_session.go

@@ -27,6 +27,8 @@ type BillingSession struct {
 	funding          FundingSource
 	preConsumedQuota int  // 实际预扣额度（信任用户可能为 0）
 	tokenConsumed    int  // 令牌额度实际扣减量
+	extraReserved    int  // 发送前补充预扣的额度（订阅退款时需要单独回滚）
+	trusted          bool // 是否命中信任额度旁路
 	fundingSettled   bool // funding.Settle 已成功，资金来源已提交
 	settled          bool // Settle 全部完成（资金 + 令牌）
 	refunded         bool // Refund 已调用
@@ -97,6 +99,8 @@ func (s *BillingSession) Refund(c *gin.Context) {
 	tokenKey := s.relayInfo.TokenKey
 	isPlayground := s.relayInfo.IsPlayground
 	tokenConsumed := s.tokenConsumed
+	extraReserved := s.extraReserved
+	subscriptionId := s.relayInfo.SubscriptionId
 	funding := s.funding
 
 	gopool.Go(func() {
@@ -104,6 +108,11 @@ func (s *BillingSession) Refund(c *gin.Context) {
 		if err := funding.Refund(); err != nil {
 			common.SysLog("error refunding billing source: " + err.Error())
 		}
+		if extraReserved > 0 && funding.Source() == BillingSourceSubscription && subscriptionId > 0 {
+			if err := model.PostConsumeUserSubscriptionDelta(subscriptionId, -int64(extraReserved)); err != nil {
+				common.SysLog("error refunding subscription extra reserved quota: " + err.Error())
+			}
+		}
 		// 2) 退还令牌额度
 		if tokenConsumed > 0 && !isPlayground {
 			if err := model.IncreaseTokenQuota(tokenId, tokenKey, tokenConsumed); err != nil {
@@ -140,6 +149,34 @@ func (s *BillingSession) GetPreConsumedQuota() int {
 	return s.preConsumedQuota
 }
 
+func (s *BillingSession) Reserve(targetQuota int) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if s.settled || s.refunded || s.trusted || targetQuota <= s.preConsumedQuota {
+		return nil
+	}
+
+	delta := targetQuota - s.preConsumedQuota
+	if delta <= 0 {
+		return nil
+	}
+
+	if err := s.reserveFunding(delta); err != nil {
+		return err
+	}
+	if err := s.reserveToken(delta); err != nil {
+		s.rollbackFundingReserve(delta)
+		return err
+	}
+
+	s.preConsumedQuota += delta
+	s.tokenConsumed += delta
+	s.extraReserved += delta
+	s.syncRelayInfo()
+	return nil
+}
+
 // ---------------------------------------------------------------------------
 // PreConsume — 统一预扣费入口（含信任额度旁路）
 // ---------------------------------------------------------------------------
@@ -151,6 +188,7 @@ func (s *BillingSession) preConsume(c *gin.Context, quota int) *types.NewAPIErro
 
 	// ---- 信任额度旁路 ----
 	if s.shouldTrust(c) {
+		s.trusted = true
 		effectiveQuota = 0
 		logger.LogInfo(c, fmt.Sprintf("用户 %d 额度充足, 信任且不需要预扣费 (funding=%s)", s.relayInfo.UserId, s.funding.Source()))
 	} else if effectiveQuota > 0 {
@@ -191,6 +229,55 @@ func (s *BillingSession) preConsume(c *gin.Context, quota int) *types.NewAPIErro
 	return nil
 }
 
+func (s *BillingSession) reserveFunding(delta int) error {
+	switch funding := s.funding.(type) {
+	case *WalletFunding:
+		if err := model.DecreaseUserQuota(funding.userId, delta); err != nil {
+			return types.NewError(err, types.ErrorCodeUpdateDataError, types.ErrOptionWithSkipRetry())
+		}
+		funding.consumed += delta
+		return nil
+	case *SubscriptionFunding:
+		if err := model.PostConsumeUserSubscriptionDelta(funding.subscriptionId, int64(delta)); err != nil {
+			return types.NewErrorWithStatusCode(
+				fmt.Errorf("订阅额度不足或未配置订阅: %s", err.Error()),
+				types.ErrorCodeInsufficientUserQuota,
+				http.StatusForbidden,
+				types.ErrOptionWithSkipRetry(),
+				types.ErrOptionWithNoRecordErrorLog(),
+			)
+		}
+		return nil
+	default:
+		return types.NewError(fmt.Errorf("unsupported funding source: %s", s.funding.Source()), types.ErrorCodeUpdateDataError, types.ErrOptionWithSkipRetry())
+	}
+}
+
+func (s *BillingSession) rollbackFundingReserve(delta int) {
+	switch funding := s.funding.(type) {
+	case *WalletFunding:
+		if err := model.IncreaseUserQuota(funding.userId, delta, false); err != nil {
+			common.SysLog("error rolling back wallet funding reserve: " + err.Error())
+		} else {
+			funding.consumed -= delta
+		}
+	case *SubscriptionFunding:
+		if err := model.PostConsumeUserSubscriptionDelta(funding.subscriptionId, -int64(delta)); err != nil {
+			common.SysLog("error rolling back subscription funding reserve: " + err.Error())
+		}
+	}
+}
+
+func (s *BillingSession) reserveToken(delta int) error {
+	if delta <= 0 || s.relayInfo.IsPlayground {
+		return nil
+	}
+	if err := PreConsumeTokenQuota(s.relayInfo, delta); err != nil {
+		return types.NewErrorWithStatusCode(err, types.ErrorCodePreConsumeTokenQuotaFailed, http.StatusForbidden, types.ErrOptionWithSkipRetry(), types.ErrOptionWithNoRecordErrorLog())
+	}
+	return nil
+}
+
 // shouldTrust 统一信任额度检查，适用于钱包和订阅。
 func (s *BillingSession) shouldTrust(c *gin.Context) bool {
 	// 异步任务（ForcePreConsume=true）必须预扣全额，不允许信任旁路
@@ -235,10 +322,10 @@ func (s *BillingSession) syncRelayInfo() {
 
 	if sub, ok := s.funding.(*SubscriptionFunding); ok {
 		info.SubscriptionId = sub.subscriptionId
-		info.SubscriptionPreConsumed = sub.preConsumed
+		info.SubscriptionPreConsumed = sub.preConsumed + int64(s.extraReserved)
 		info.SubscriptionPostDelta = 0
 		info.SubscriptionAmountTotal = sub.AmountTotal
-		info.SubscriptionAmountUsedAfterPreConsume = sub.AmountUsedAfter
+		info.SubscriptionAmountUsedAfterPreConsume = sub.AmountUsedAfter + int64(s.extraReserved)
 		info.SubscriptionPlanId = sub.PlanId
 		info.SubscriptionPlanTitle = sub.PlanTitle
 	} else {

service/log_info_generate.go

@@ -6,6 +6,7 @@ import (
 	"github.com/QuantumNous/new-api/common"
 	"github.com/QuantumNous/new-api/constant"
 	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/pkg/billingexpr"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
 	"github.com/QuantumNous/new-api/types"
 
@@ -214,3 +215,42 @@ func GenerateMjOtherInfo(relayInfo *relaycommon.RelayInfo, priceData types.Price
 	appendRequestPath(nil, relayInfo, other)
 	return other
 }
+
+func GenerateTieredOtherInfo(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, result *billingexpr.TieredResult) map[string]interface{} {
+	other := make(map[string]interface{})
+	other["billing_mode"] = "tiered_expr"
+
+	snap := relayInfo.TieredBillingSnapshot
+	if snap != nil {
+		other["group_ratio"] = snap.GroupRatio
+		other["expr_hash"] = snap.ExprHash
+		other["estimated_prompt_tokens"] = snap.EstimatedPromptTokens
+		other["estimated_completion_tokens"] = snap.EstimatedCompletionTokens
+		other["estimated_quota_before_group"] = snap.EstimatedQuotaBeforeGroup
+		other["estimated_quota_after_group"] = snap.EstimatedQuotaAfterGroup
+		other["estimated_tier"] = snap.EstimatedTier
+	}
+
+	if result != nil {
+		other["actual_quota_before_group"] = result.ActualQuotaBeforeGroup
+		other["actual_quota_after_group"] = result.ActualQuotaAfterGroup
+		other["matched_tier"] = result.MatchedTier
+		other["crossed_tier"] = result.CrossedTier
+	}
+
+	other["frt"] = float64(relayInfo.FirstResponseTime.UnixMilli() - relayInfo.StartTime.UnixMilli())
+	if relayInfo.IsModelMapped {
+		other["is_model_mapped"] = true
+		other["upstream_model_name"] = relayInfo.UpstreamModelName
+	}
+
+	adminInfo := make(map[string]interface{})
+	adminInfo["use_channel"] = ctx.GetStringSlice("use_channel")
+	AppendChannelAffinityAdminInfo(ctx, adminInfo)
+	other["admin_info"] = adminInfo
+
+	appendRequestPath(ctx, relayInfo, other)
+	appendRequestConversionChain(relayInfo, other)
+	appendBillingInfo(relayInfo, other)
+	return other
+}

service/quota.go

@@ -13,6 +13,7 @@ import (
 	"github.com/QuantumNous/new-api/dto"
 	"github.com/QuantumNous/new-api/logger"
 	"github.com/QuantumNous/new-api/model"
+	"github.com/QuantumNous/new-api/pkg/billingexpr"
 	relaycommon "github.com/QuantumNous/new-api/relay/common"
 	"github.com/QuantumNous/new-api/setting/ratio_setting"
 	"github.com/QuantumNous/new-api/setting/system_setting"
@@ -157,6 +158,15 @@ func PreWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usag
 func PostWssConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelName string,
 	usage *dto.RealtimeUsage, extraContent string) {
 
+	// Tiered billing early return
+	if ok, tieredQuota, tieredResult := TryTieredSettle(relayInfo, billingexpr.TokenParams{
+		P: float64(usage.InputTokens),
+		C: float64(usage.OutputTokens),
+	}); ok {
+		postConsumeQuotaTieredService(ctx, relayInfo, modelName, usage.InputTokens, usage.OutputTokens, usage.TotalTokens, tieredQuota, tieredResult, extraContent)
+		return
+	}
+
 	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
 	textInputTokens := usage.InputTokenDetails.TextTokens
 	textOutTokens := usage.OutputTokenDetails.TextTokens
@@ -240,6 +250,18 @@ func PostClaudeConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 		ObserveChannelAffinityUsageCacheByRelayFormat(ctx, usage, relayInfo.GetFinalRequestRelayFormat())
 	}
 
+	// Tiered billing early return
+	if ok, tieredQuota, tieredResult := TryTieredSettle(relayInfo, billingexpr.TokenParams{
+		P:    float64(usage.PromptTokens),
+		C:    float64(usage.CompletionTokens),
+		CR:   float64(usage.PromptTokensDetails.CachedTokens),
+		CC:   float64(usage.PromptTokensDetails.CachedCreationTokens - usage.ClaudeCacheCreation1hTokens),
+		CC1h: float64(usage.ClaudeCacheCreation1hTokens),
+	}); ok {
+		postConsumeQuotaTieredService(ctx, relayInfo, relayInfo.OriginModelName, usage.PromptTokens, usage.CompletionTokens, usage.PromptTokens+usage.CompletionTokens, tieredQuota, tieredResult, "")
+		return
+	}
+
 	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
 	promptTokens := usage.PromptTokens
 	completionTokens := usage.CompletionTokens
@@ -360,6 +382,16 @@ func CalcOpenRouterCacheCreateTokens(usage dto.Usage, priceData types.PriceData)
 
 func PostAudioConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, usage *dto.Usage, extraContent string) {
 
+	// Tiered billing early return
+	if ok, tieredQuota, tieredResult := TryTieredSettle(relayInfo, billingexpr.TokenParams{
+		P:  float64(usage.PromptTokens),
+		C:  float64(usage.CompletionTokens),
+		CR: float64(usage.PromptTokensDetails.CachedTokens),
+	}); ok {
+		postConsumeQuotaTieredService(ctx, relayInfo, relayInfo.OriginModelName, usage.PromptTokens, usage.CompletionTokens, usage.TotalTokens, tieredQuota, tieredResult, extraContent)
+		return
+	}
+
 	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
 	textInputTokens := usage.PromptTokensDetails.TextTokens
 	textOutTokens := usage.CompletionTokenDetails.TextTokens
@@ -607,3 +639,50 @@ func checkAndSendSubscriptionQuotaNotify(relayInfo *relaycommon.RelayInfo) {
 		}
 	})
 }
+
+func postConsumeQuotaTieredService(ctx *gin.Context, relayInfo *relaycommon.RelayInfo, modelName string,
+	promptTokens, completionTokens, totalTokens, quota int, tieredResult *TieredResultWrapper, extraContent string) {
+
+	useTimeSeconds := time.Now().Unix() - relayInfo.StartTime.Unix()
+	tokenName := ctx.GetString("token_name")
+	groupRatio := relayInfo.PriceData.GroupRatioInfo.GroupRatio
+
+	var logContent string
+	if totalTokens == 0 {
+		quota = 0
+		logContent = "上游没有返回计费信息（可能是上游超时）"
+		logger.LogError(ctx, fmt.Sprintf("tiered billing: total tokens is 0, userId %d, channelId %d, tokenId %d, model %s, pre-consumed %d",
+			relayInfo.UserId, relayInfo.ChannelId, relayInfo.TokenId, modelName, relayInfo.FinalPreConsumedQuota))
+	} else {
+		if groupRatio != 0 && quota == 0 {
+			quota = 1
+		}
+		model.UpdateUserUsedQuotaAndRequestCount(relayInfo.UserId, quota)
+		model.UpdateChannelUsedQuota(relayInfo.ChannelId, quota)
+	}
+
+	if err := SettleBilling(ctx, relayInfo, quota); err != nil {
+		logger.LogError(ctx, "error settling tiered billing: "+err.Error())
+	}
+
+	if extraContent != "" {
+		logContent += extraContent
+	}
+
+	other := GenerateTieredOtherInfo(ctx, relayInfo, tieredResult)
+
+	model.RecordConsumeLog(ctx, relayInfo.UserId, model.RecordConsumeLogParams{
+		ChannelId:        relayInfo.ChannelId,
+		PromptTokens:     promptTokens,
+		CompletionTokens: completionTokens,
+		ModelName:        modelName,
+		TokenName:        tokenName,
+		Quota:            quota,
+		Content:          logContent,
+		TokenId:          relayInfo.TokenId,
+		UseTimeSeconds:   int(useTimeSeconds),
+		IsStream:         relayInfo.IsStream,
+		Group:            relayInfo.UsingGroup,
+		Other:            other,
+	})
+}

service/tiered_settle.go

@@ -0,0 +1,36 @@
+package service
+
+import (
+	"github.com/QuantumNous/new-api/pkg/billingexpr"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+)
+
+// TieredResultWrapper wraps billingexpr.TieredResult for use at the service layer.
+type TieredResultWrapper = billingexpr.TieredResult
+
+// TryTieredSettle checks if the request uses tiered_expr billing and, if so,
+// computes the actual quota using the frozen BillingSnapshot. Returns:
+//   - ok=true, quota, result  when tiered billing applies
+//   - ok=false, 0, nil        when it doesn't (caller should fall through to existing logic)
+func TryTieredSettle(relayInfo *relaycommon.RelayInfo, params billingexpr.TokenParams) (ok bool, quota int, result *billingexpr.TieredResult) {
+	snap := relayInfo.TieredBillingSnapshot
+	if snap == nil || snap.BillingMode != "tiered_expr" {
+		return false, 0, nil
+	}
+
+	requestInput := billingexpr.RequestInput{}
+	if relayInfo.BillingRequestInput != nil {
+		requestInput = *relayInfo.BillingRequestInput
+	}
+
+	tr, err := billingexpr.ComputeTieredQuotaWithRequest(snap, params, requestInput)
+	if err != nil {
+		quota = relayInfo.FinalPreConsumedQuota
+		if quota <= 0 {
+			quota = snap.EstimatedQuotaAfterGroup
+		}
+		return true, quota, nil
+	}
+
+	return true, tr.ActualQuotaAfterGroup, &tr
+}

service/tiered_settle_test.go

@@ -0,0 +1,401 @@
+package service
+
+import (
+	"testing"
+
+	"github.com/QuantumNous/new-api/pkg/billingexpr"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+)
+
+// Claude Sonnet-style tiered expression: standard vs long-context
+const sonnetTieredExpr = `p <= 200000 ? tier("standard", p * 1.5 + c * 7.5) : tier("long_context", p * 3 + c * 11.25)`
+
+// Simple flat expression
+const flatExpr = `tier("default", p * 2 + c * 10)`
+
+// Expression with cache tokens
+const cacheExpr = `tier("default", p * 2 + c * 10 + cr * 0.2 + cc * 2.5 + cc1h * 4)`
+
+// Expression with request probes
+const probeExpr = `param("service_tier") == "fast" ? tier("fast", p * 4 + c * 20) : tier("normal", p * 2 + c * 10)`
+
+func makeSnapshot(expr string, groupRatio float64, estPrompt, estCompletion int) *billingexpr.BillingSnapshot {
+	return &billingexpr.BillingSnapshot{
+		BillingMode:               "tiered_expr",
+		ExprString:                expr,
+		ExprHash:                  billingexpr.ExprHashString(expr),
+		GroupRatio:                groupRatio,
+		EstimatedPromptTokens:     estPrompt,
+		EstimatedCompletionTokens: estCompletion,
+	}
+}
+
+func makeRelayInfo(expr string, groupRatio float64, estPrompt, estCompletion int) *relaycommon.RelayInfo {
+	snap := makeSnapshot(expr, groupRatio, estPrompt, estCompletion)
+	cost, trace, _ := billingexpr.RunExpr(expr, billingexpr.TokenParams{P: float64(estPrompt), C: float64(estCompletion)})
+	snap.EstimatedQuotaBeforeGroup = cost
+	snap.EstimatedQuotaAfterGroup = billingexpr.QuotaRound(cost * groupRatio)
+	snap.EstimatedTier = trace.MatchedTier
+	return &relaycommon.RelayInfo{
+		TieredBillingSnapshot: snap,
+		FinalPreConsumedQuota: snap.EstimatedQuotaAfterGroup,
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Existing tests (preserved)
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettleUsesFrozenRequestInput(t *testing.T) {
+	exprStr := `param("service_tier") == "fast" ? tier("fast", p * 2) : tier("normal", p)`
+	relayInfo := &relaycommon.RelayInfo{
+		TieredBillingSnapshot: &billingexpr.BillingSnapshot{
+			BillingMode:               "tiered_expr",
+			ExprString:                exprStr,
+			ExprHash:                  billingexpr.ExprHashString(exprStr),
+			GroupRatio:                1.0,
+			EstimatedPromptTokens:     100,
+			EstimatedCompletionTokens: 0,
+			EstimatedQuotaAfterGroup:  100,
+		},
+		BillingRequestInput: &billingexpr.RequestInput{
+			Body: []byte(`{"service_tier":"fast"}`),
+		},
+	}
+
+	ok, quota, result := TryTieredSettle(relayInfo, billingexpr.TokenParams{P: 100})
+	if !ok {
+		t.Fatal("expected tiered settle to apply")
+	}
+	if quota != 200 {
+		t.Fatalf("quota = %d, want 200", quota)
+	}
+	if result == nil || result.MatchedTier != "fast" {
+		t.Fatalf("matched tier = %v, want fast", result)
+	}
+}
+
+func TestTryTieredSettleFallsBackToFrozenPreConsumeOnExprError(t *testing.T) {
+	relayInfo := &relaycommon.RelayInfo{
+		FinalPreConsumedQuota: 321,
+		TieredBillingSnapshot: &billingexpr.BillingSnapshot{
+			BillingMode:              "tiered_expr",
+			ExprString:               `invalid +-+ expr`,
+			ExprHash:                 billingexpr.ExprHashString(`invalid +-+ expr`),
+			GroupRatio:               1.0,
+			EstimatedQuotaAfterGroup: 123,
+		},
+	}
+
+	ok, quota, result := TryTieredSettle(relayInfo, billingexpr.TokenParams{P: 100})
+	if !ok {
+		t.Fatal("expected tiered settle to apply")
+	}
+	if quota != 321 {
+		t.Fatalf("quota = %d, want 321", quota)
+	}
+	if result != nil {
+		t.Fatalf("result = %#v, want nil", result)
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Pre-consume vs Post-consume consistency
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettle_PreConsumeMatchesPostConsume(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 1.0, 1000, 500)
+	params := billingexpr.TokenParams{P: 1000, C: 500}
+
+	ok, quota, _ := TryTieredSettle(info, params)
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// p*2 + c*10 = 2000 + 5000 = 7000
+	if quota != 7000 {
+		t.Fatalf("quota = %d, want 7000", quota)
+	}
+	if quota != info.FinalPreConsumedQuota {
+		t.Fatalf("pre-consume %d != post-consume %d", info.FinalPreConsumedQuota, quota)
+	}
+}
+
+func TestTryTieredSettle_PostConsumeOverPreConsume(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 1.0, 1000, 500)
+	preConsumed := info.FinalPreConsumedQuota // 7000
+
+	// Actual usage is higher than estimated
+	params := billingexpr.TokenParams{P: 2000, C: 1000}
+	ok, quota, _ := TryTieredSettle(info, params)
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// p*2 + c*10 = 4000 + 10000 = 14000
+	if quota != 14000 {
+		t.Fatalf("quota = %d, want 14000", quota)
+	}
+	if quota <= preConsumed {
+		t.Fatalf("expected supplement: actual %d should > pre-consumed %d", quota, preConsumed)
+	}
+}
+
+func TestTryTieredSettle_PostConsumeUnderPreConsume(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 1.0, 1000, 500)
+	preConsumed := info.FinalPreConsumedQuota // 7000
+
+	// Actual usage is lower than estimated
+	params := billingexpr.TokenParams{P: 100, C: 50}
+	ok, quota, _ := TryTieredSettle(info, params)
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// p*2 + c*10 = 200 + 500 = 700
+	if quota != 700 {
+		t.Fatalf("quota = %d, want 700", quota)
+	}
+	if quota >= preConsumed {
+		t.Fatalf("expected refund: actual %d should < pre-consumed %d", quota, preConsumed)
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Tiered boundary conditions
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettle_ExactBoundary(t *testing.T) {
+	info := makeRelayInfo(sonnetTieredExpr, 1.0, 200000, 1000)
+
+	// p == 200000 => standard tier (p <= 200000)
+	ok, quota, result := TryTieredSettle(info, billingexpr.TokenParams{P: 200000, C: 1000})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// standard: p*1.5 + c*7.5 = 300000 + 7500 = 307500
+	if quota != 307500 {
+		t.Fatalf("quota = %d, want 307500", quota)
+	}
+	if result.MatchedTier != "standard" {
+		t.Fatalf("tier = %s, want standard", result.MatchedTier)
+	}
+}
+
+func TestTryTieredSettle_BoundaryPlusOne(t *testing.T) {
+	info := makeRelayInfo(sonnetTieredExpr, 1.0, 200000, 1000)
+
+	// p == 200001 => crosses to long_context tier
+	ok, quota, result := TryTieredSettle(info, billingexpr.TokenParams{P: 200001, C: 1000})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// long_context: p*3 + c*11.25 = 600003 + 11250 = 611253
+	if quota != 611253 {
+		t.Fatalf("quota = %d, want 611253", quota)
+	}
+	if result.MatchedTier != "long_context" {
+		t.Fatalf("tier = %s, want long_context", result.MatchedTier)
+	}
+	if !result.CrossedTier {
+		t.Fatal("expected CrossedTier = true")
+	}
+}
+
+func TestTryTieredSettle_ZeroTokens(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 1.0, 0, 0)
+
+	ok, quota, result := TryTieredSettle(info, billingexpr.TokenParams{P: 0, C: 0})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	if quota != 0 {
+		t.Fatalf("quota = %d, want 0", quota)
+	}
+	if result == nil {
+		t.Fatal("result should not be nil")
+	}
+}
+
+func TestTryTieredSettle_HugeTokens(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 1.0, 10000000, 5000000)
+
+	ok, quota, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 10000000, C: 5000000})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// p*2 + c*10 = 20000000 + 50000000 = 70000000
+	if quota != 70000000 {
+		t.Fatalf("quota = %d, want 70000000", quota)
+	}
+}
+
+func TestTryTieredSettle_CacheTokensAffectSettlement(t *testing.T) {
+	info := makeRelayInfo(cacheExpr, 1.0, 1000, 500)
+
+	// Without cache tokens
+	ok1, quota1, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if !ok1 {
+		t.Fatal("expected tiered settle")
+	}
+	// p*2 + c*10 + cr*0.2 + cc*2.5 + cc1h*4 = 2000 + 5000 + 0 + 0 + 0 = 7000
+
+	// With cache tokens
+	ok2, quota2, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500, CR: 10000, CC: 5000, CC1h: 2000})
+	if !ok2 {
+		t.Fatal("expected tiered settle")
+	}
+	// 2000 + 5000 + 10000*0.2 + 5000*2.5 + 2000*4 = 2000 + 5000 + 2000 + 12500 + 8000 = 29500
+
+	if quota2 <= quota1 {
+		t.Fatalf("cache tokens should increase quota: without=%d, with=%d", quota1, quota2)
+	}
+	if quota1 != 7000 {
+		t.Fatalf("no-cache quota = %d, want 7000", quota1)
+	}
+	if quota2 != 29500 {
+		t.Fatalf("cache quota = %d, want 29500", quota2)
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Request probe tests
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettle_RequestProbeInfluencesBilling(t *testing.T) {
+	info := makeRelayInfo(probeExpr, 1.0, 1000, 500)
+	info.BillingRequestInput = &billingexpr.RequestInput{
+		Body: []byte(`{"service_tier":"fast"}`),
+	}
+
+	ok, quota, result := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// fast: p*4 + c*20 = 4000 + 10000 = 14000
+	if quota != 14000 {
+		t.Fatalf("quota = %d, want 14000", quota)
+	}
+	if result.MatchedTier != "fast" {
+		t.Fatalf("tier = %s, want fast", result.MatchedTier)
+	}
+}
+
+func TestTryTieredSettle_NoRequestInput_FallsBackToDefault(t *testing.T) {
+	info := makeRelayInfo(probeExpr, 1.0, 1000, 500)
+	// No BillingRequestInput set — param("service_tier") returns nil, not "fast"
+
+	ok, quota, result := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// normal: p*2 + c*10 = 2000 + 5000 = 7000
+	if quota != 7000 {
+		t.Fatalf("quota = %d, want 7000", quota)
+	}
+	if result.MatchedTier != "normal" {
+		t.Fatalf("tier = %s, want normal", result.MatchedTier)
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Group ratio tests
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettle_GroupRatioScaling(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 1.5, 1000, 500)
+
+	ok, quota, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	// cost = 7000, after group = round(7000 * 1.5) = 10500
+	if quota != 10500 {
+		t.Fatalf("quota = %d, want 10500", quota)
+	}
+}
+
+func TestTryTieredSettle_GroupRatioZero(t *testing.T) {
+	info := makeRelayInfo(flatExpr, 0, 1000, 500)
+
+	ok, quota, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if !ok {
+		t.Fatal("expected tiered settle")
+	}
+	if quota != 0 {
+		t.Fatalf("quota = %d, want 0 (group ratio = 0)", quota)
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Ratio mode (negative tests) — TryTieredSettle must return false
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettle_RatioMode_NilSnapshot(t *testing.T) {
+	info := &relaycommon.RelayInfo{
+		TieredBillingSnapshot: nil,
+	}
+
+	ok, _, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if ok {
+		t.Fatal("expected TryTieredSettle to return false when snapshot is nil")
+	}
+}
+
+func TestTryTieredSettle_RatioMode_WrongBillingMode(t *testing.T) {
+	info := &relaycommon.RelayInfo{
+		TieredBillingSnapshot: &billingexpr.BillingSnapshot{
+			BillingMode: "ratio",
+			ExprString:  flatExpr,
+			ExprHash:    billingexpr.ExprHashString(flatExpr),
+			GroupRatio:  1.0,
+		},
+	}
+
+	ok, _, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if ok {
+		t.Fatal("expected TryTieredSettle to return false for ratio billing mode")
+	}
+}
+
+func TestTryTieredSettle_RatioMode_EmptyBillingMode(t *testing.T) {
+	info := &relaycommon.RelayInfo{
+		TieredBillingSnapshot: &billingexpr.BillingSnapshot{
+			BillingMode: "",
+			ExprString:  flatExpr,
+			ExprHash:    billingexpr.ExprHashString(flatExpr),
+			GroupRatio:  1.0,
+		},
+	}
+
+	ok, _, _ := TryTieredSettle(info, billingexpr.TokenParams{P: 1000, C: 500})
+	if ok {
+		t.Fatal("expected TryTieredSettle to return false for empty billing mode")
+	}
+}
+
+// ---------------------------------------------------------------------------
+// Fallback tests
+// ---------------------------------------------------------------------------
+
+func TestTryTieredSettle_ErrorFallbackToEstimatedQuotaAfterGroup(t *testing.T) {
+	info := &relaycommon.RelayInfo{
+		FinalPreConsumedQuota: 0,
+		TieredBillingSnapshot: &billingexpr.BillingSnapshot{
+			BillingMode:              "tiered_expr",
+			ExprString:               `invalid expr!!!`,
+			ExprHash:                 billingexpr.ExprHashString(`invalid expr!!!`),
+			GroupRatio:               1.0,
+			EstimatedQuotaAfterGroup: 999,
+		},
+	}
+
+	ok, quota, result := TryTieredSettle(info, billingexpr.TokenParams{P: 100})
+	if !ok {
+		t.Fatal("expected tiered settle to apply")
+	}
+	// FinalPreConsumedQuota is 0, should fall back to EstimatedQuotaAfterGroup
+	if quota != 999 {
+		t.Fatalf("quota = %d, want 999", quota)
+	}
+	if result != nil {
+		t.Fatal("result should be nil on error fallback")
+	}
+}