@sema-agent/server 3.7.1 → 3.8.0

package/dist/budget.js

@@ -139,6 +139,12 @@ export function createTracer(metrics, costQuota, modelUsage, fleetUsage, fleetLe
                 if (principal)
                     costQuota?.add(principal, e.costMicroUsd);
             }
+            else if (e.costMicroUsd === undefined) {
+                // D5:cost 整键缺席 = 未知(RB-368)——计数供 /metrics/summary 的 costUnknown/unpricedCalls
+                // 信号(costUsd 在 unpriced 部署恒 0,没有这个信号消费端会读成「零花费」)。显式 0 =
+                // declared free,走上面 truthy 为假的静默臂,不计入 unpriced。
+                metrics.inc("brain_calls_unpriced_total", { model: modelId(e.model) });
+            }
             // ┌─ #59 裁定 / RB-368 结案([2052]① cli 报案 → [2073] 立案 → [2076] 工单挂 core → [2083] 到货) ─┐
             // 病灶报案:下面三处曾是 `e.costMicroUsd ?? 0`(账本 + fleetUsage + fleetLease),把「成本未知」记成
             // 「免费」,与 `pricingConfigured=false` 的部署组合 ⇒ 全部账目假 0。

package/dist/observability/metrics.d.ts

@@ -34,6 +34,9 @@ export interface MetricsSummary {
     tokensTotal: number;
     costUsd: number;
     costUsdByModel: Record<string, number>;
+    /** true ⇒ 窗口内有 cost 未知的 brain.call(unpriced 部署),costUsd/costUsdByModel 是下界(D5)。 */
+    costUnknown: boolean;
+    unpricedCalls: number;
     taskDurationAvgSec: number | null;
     brainFirstTokenAvgMs: number | null;
     brainCallAvgMs: number | null;

package/dist/observability/metrics.js

@@ -163,6 +163,11 @@ export class Metrics {
             tokensTotal: ctotal("task_tokens_total"),
             costUsd: ctotal("model_cost_micro_usd_total") / 1e6,
             costUsdByModel: Object.fromEntries(Object.entries(cby("model_cost_micro_usd_total", "model")).map(([k, v]) => [k, v / 1e6])),
+            // D5([2122] 挂账,additive v1):unpriced 部署下 costUsd 恒 0 而消费端读成「零花费」。
+            // costUnknown=true ⇒ costUsd/costUsdByModel 是**下界**(RB-368 缺席=未知语义;显式 0 = declared
+            // free 不计入 unpriced)。计数点在 budget.ts 的 brain.call 臂。
+            costUnknown: ctotal("brain_calls_unpriced_total") > 0,
+            unpricedCalls: ctotal("brain_calls_unpriced_total"),
             taskDurationAvgSec: havg("task_duration_seconds"),
             brainFirstTokenAvgMs: havg("brain_first_token_ms"),
             brainCallAvgMs: havg("brain_call_latency_ms"),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sema-agent/server",
-  "version": "3.7.1",
+  "version": "3.8.0",
   "description": "Sema Server — the server/API implementation layer for Sema, wiring core, registry, model providers, and cloud agent execution. Built on @sema-agent/core.",
   "type": "module",
   "license": "BUSL-1.1",
@@ -46,6 +46,7 @@
     "dev:bake-runner": "tsx src/bake-runner/main.ts",
     "test": "vitest run",
     "test:db-integration": "vitest run integration",
+    "perf:baseline": "tsx scripts/perf-baseline.mjs",
     "build:binary": "bun build --compile src/main.ts --outfile dist/sema-server",
     "build:binary:linux-x64": "bun build --compile --target=bun-linux-x64 src/main.ts --outfile dist/sema-server-linux-x64",
     "build:binary:darwin-arm64": "bun build --compile --target=bun-darwin-arm64 src/main.ts --outfile dist/sema-server-darwin-arm64",