retrace-sdk 0.15.0 → 0.16.0

package/dist/interceptors/anthropic.js

@@ -3,6 +3,7 @@ import { genId, nowIso, truncateJson, wasTruncated } from "../utils.js";
 import { isReplaying, consumeCassetteEntry } from "../replay.js";
 import { emitAnthropicToolCalls, emitAnthropicToolResults, parseToolArgs, resetToolResultDedup, extractToolSchemas, extractSamplingParams } from "./tool-spans.js";
 import { dispatchRegisterOpenSpan, dispatchUnregisterOpenSpan } from "./_dispatch.js";
+import { UNKNOWN_MODEL_PRICING, costFromRate } from "../pricing.js";
 const PRICING = {
     "claude-opus-4.7": [5.0, 25.0],
     "claude-opus-4.6": [5.0, 25.0],
@@ -17,9 +18,10 @@ const PRICING = {
 function calcCost(model, inputTokens, outputTokens) {
     for (const [key, p] of Object.entries(PRICING)) {
         if (model.includes(key))
-            return (inputTokens * p[0] + outputTokens * p[1]) / 1_000_000;
+            return costFromRate(p, inputTokens, outputTokens);
     }
-    return 0;
+    // Unknown model — never cost $0 (that silently disarms USD budget ceilings). Conservative fallback.
+    return costFromRate(UNKNOWN_MODEL_PRICING, inputTokens, outputTokens);
 }
 let originalCreate = null;
 let installed = false;

package/dist/interceptors/gemini.js

@@ -2,6 +2,7 @@ import { SpanType } from "../trace.js";
 import { genId, nowIso, truncateJson } from "../utils.js";
 import { dispatchRegisterOpenSpan, dispatchUnregisterOpenSpan, captureActiveSpanEmit } from "./_dispatch.js";
 import { emitGeminiToolCalls, emitGeminiToolResults, resetToolResultDedup, extractToolSchemas, extractSamplingParams } from "./tool-spans.js";
+import { UNKNOWN_MODEL_PRICING, costFromRate } from "../pricing.js";
 const PRICING = {
     "gemini-3.1-flash-lite": [0.10, 0.40],
     "gemini-3.1-flash": [0.50, 3.0],
@@ -15,8 +16,10 @@ const PRICING = {
     "gemini-2.0-flash-lite": [0.05, 0.20],
 };
 function calcCost(model, inputTokens, outputTokens) {
-    const p = PRICING[model] || [0, 0];
-    return (inputTokens * p[0] + outputTokens * p[1]) / 1_000_000;
+    // Exact-match table (substring matching would mis-rank "…-flash" vs "…-flash-lite"); an unknown
+    // model falls back to a conservative non-zero rate so USD budget ceilings still engage.
+    const p = PRICING[model] ?? UNKNOWN_MODEL_PRICING;
+    return costFromRate(p, inputTokens, outputTokens);
 }
 let onSpanCallback = null;
 // eslint-disable-next-line @typescript-eslint/no-explicit-any

package/dist/interceptors/openai.js

@@ -5,6 +5,7 @@ import { getConfig } from "../config.js";
 import { RetraceRateLimitError, RetraceAuthError, RetraceConnectionError } from "../errors.js";
 import { emitOpenAIToolCalls, emitOpenAIToolResults, parseToolArgs, resetToolResultDedup, extractToolSchemas, extractSamplingParams } from "./tool-spans.js";
 import { dispatchRegisterOpenSpan, dispatchUnregisterOpenSpan } from "./_dispatch.js";
+import { UNKNOWN_MODEL_PRICING, costFromRate } from "../pricing.js";
 /** Hardcoded fallback pricing ($/1M tokens: [input, output]). Updated periodically. */
 const FALLBACK_PRICING = {
     "gpt-5.5-pro": [30.0, 180.0],
@@ -55,9 +56,10 @@ function calcCost(model, inputTokens, outputTokens) {
     const pricing = livePricing || FALLBACK_PRICING;
     for (const [key, p] of Object.entries(pricing)) {
         if (model.includes(key))
-            return (inputTokens * p[0] + outputTokens * p[1]) / 1_000_000;
+            return costFromRate(p, inputTokens, outputTokens);
     }
-    return 0;
+    // Unknown model — never cost $0 (that silently disarms USD budget ceilings). Conservative fallback.
+    return costFromRate(UNKNOWN_MODEL_PRICING, inputTokens, outputTokens);
 }
 let originalCreate = null;
 let installed = false;

package/dist/pricing.d.ts

@@ -0,0 +1,14 @@
+/**
+ * Conservative fallback pricing for models not present in an interceptor's price table.
+ *
+ * A model that matches no key (a brand-new release, a fine-tune, an OpenRouter alias) would
+ * otherwise be costed at $0 — and a $0 cost silently disarms USD budget ceilings in the enforcement
+ * gate (the gate trips on estimated USD per run). We instead cost unknown models at a conservative
+ * mid/high frontier rate so a budget cap still engages; the value self-corrects the moment the
+ * server price table (synced via /api/v1/pricing/models) learns the model.
+ *
+ * $/1M tokens: [input, output].
+ */
+export declare const UNKNOWN_MODEL_PRICING: readonly [number, number];
+/** Cost in USD from an [input, output] $/1M-token rate. */
+export declare function costFromRate(rate: readonly [number, number], inputTokens: number, outputTokens: number): number;

package/dist/pricing.js

@@ -0,0 +1,16 @@
+/**
+ * Conservative fallback pricing for models not present in an interceptor's price table.
+ *
+ * A model that matches no key (a brand-new release, a fine-tune, an OpenRouter alias) would
+ * otherwise be costed at $0 — and a $0 cost silently disarms USD budget ceilings in the enforcement
+ * gate (the gate trips on estimated USD per run). We instead cost unknown models at a conservative
+ * mid/high frontier rate so a budget cap still engages; the value self-corrects the moment the
+ * server price table (synced via /api/v1/pricing/models) learns the model.
+ *
+ * $/1M tokens: [input, output].
+ */
+export const UNKNOWN_MODEL_PRICING = [5.0, 15.0];
+/** Cost in USD from an [input, output] $/1M-token rate. */
+export function costFromRate(rate, inputTokens, outputTokens) {
+    return (inputTokens * rate[0] + outputTokens * rate[1]) / 1_000_000;
+}

package/dist/telemetry.js

@@ -14,7 +14,7 @@ import { getConfig } from "./config.js";
 const ANON_ID = Math.random().toString(16).slice(2, 18);
 const DISABLED = new Set(["0", "false", "no", "off"]);
 // Keep in sync with package.json version.
-const SDK_VERSION = "0.15.0";
+const SDK_VERSION = "0.16.0";
 function enabled() {
     return !DISABLED.has((process.env.RETRACE_TELEMETRY ?? "1").trim().toLowerCase());
 }

package/dist/transport.js

@@ -2,7 +2,7 @@ import { getConfig } from "./config.js";
 import { classifyServerSignal } from "./errors.js";
 // Client identifier sent on every request so the backend can attribute SDK usage/version.
 // Keep in sync with package.json on release.
-const CLIENT_ID = "typescript-sdk/0.15.0";
+const CLIENT_ID = "typescript-sdk/0.16.0";
 // ─── Runtime-agnostic WebSocket ──────────────────────────────────────────────
 // Prefer the global Web `WebSocket` (Node 20+, Bun, Deno, browsers, and every edge runtime); fall
 // back to the OPTIONAL `ws` package only on older Node that lacks a global. Both expose the standard

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "retrace-sdk",
-  "version": "0.15.0",
+  "version": "0.16.0",
   "description": "The execution replay engine for AI agents. Record, replay, fork, and share agent executions.",
   "type": "module",
   "main": "dist/index.js",