fix(admission): measure request bodies with the active cost budget (#13762)

Merged. Good catch: the feature extractor was silently capped by the size estimator's 256 KiB default, so every request above it reported the same flattened body/token estimate and admission decided on a number that was not the request. Deriving the measurement limit from the active cost budget — runtime overrides included — is the right place to fix it.

Validated as a combined board first (this PR merged with the 21 siblings of the same wave on the release tip): eslint with the frozen suppressions, typecheck:core, check:open-sse-typecheck, complexity, cognitive-complexity, changelog-integrity, i18n new-key coverage, docs-counts, docs-sync, migration-numbering, provider-consistency and a duplicate-identifier audit all green, plus 176 passing / 0 failing focused node:test cases across the 25 test files the wave touches and the dashboard test under Vitest (2/0). Then re-validated alone on the fresh tip before this merge: conflicts re-resolved, file sizes rebaselined for this PR's own growth, eslint and this PR's focused tests re-run.

Thank you.

open-sse/services/admission/requestFeatures.ts

@@ -4,11 +4,13 @@
  */
 
 import { estimateSizeFast } from "../../utils/estimateSize.ts";
-import type { AdmissionCostFeatures } from "./types.ts";
+import { resolveCostConfig } from "./cost.ts";
+import type { AdmissionCostConfig, AdmissionCostFeatures } from "./types.ts";
 
 export type AdmissionFeatureExtractionContext = {
   /** When set, wins over any body/wrapped stream field. */
   streaming?: boolean;
+  cost?: Partial<AdmissionCostConfig>;
 };
 
 /**
@@ -159,7 +161,8 @@ export function extractAdmissionCostFeatures(
   body: unknown,
   context?: AdmissionFeatureExtractionContext
 ): AdmissionCostFeatures {
-  const bodyBytes = estimateSizeFast(body);
+  const cost = resolveCostConfig(context?.cost);
+  const bodyBytes = estimateSizeFast(body, cost.bodyBytesPerUnit * cost.maxRequestCost);
   const layers = featureLayers(body);
   const draft: FeatureDraft = {
     messageCount: 0,

open-sse/services/admission/runtime.ts

@@ -5,6 +5,7 @@
 
 import { AdaptiveAdmissionController } from "./controller.ts";
 import { validateConfig } from "./config.ts";
+import { resolveCostConfig } from "./cost.ts";
 import { extractAdmissionCostFeatures } from "./requestFeatures.ts";
 import {
   type AdaptiveAdmissionConfig,
@@ -323,6 +324,7 @@ function classifyHttpOutcome(status: number, signal?: AbortSignal): AdmissionRel
 
 class AdaptiveAdmissionRuntimeImpl implements AdaptiveAdmissionRuntime {
   private readonly controller: AdaptiveAdmissionController;
+  private readonly costConfig: ReturnType<typeof resolveCostConfig>;
   private readonly checkResourcePressure: () => ResourcePressureGuardResult | null;
   private readonly getResourcePressureObservation: () => ResourcePressureObservation;
   private readonly onPressureObserved?: (pressure: AdmissionPressure) => void;
@@ -338,6 +340,7 @@ class AdaptiveAdmissionRuntimeImpl implements AdaptiveAdmissionRuntime {
 
   constructor(options: AdaptiveAdmissionRuntimeOptions, config: AdaptiveAdmissionConfig) {
     this.controller = new AdaptiveAdmissionController(config, options.clock);
+    this.costConfig = resolveCostConfig(config.cost);
     this.checkResourcePressure = options.checkResourcePressure ?? checkResourcePressureGuard;
     this.getResourcePressureObservation =
       options.getResourcePressureObservation ?? getResourcePressureObservation;
@@ -372,7 +375,10 @@ class AdaptiveAdmissionRuntimeImpl implements AdaptiveAdmissionRuntime {
 
     const features = extractAdmissionCostFeatures(
       input.body,
-      input.streaming === undefined ? undefined : { streaming: input.streaming }
+      {
+        streaming: input.streaming,
+        cost: this.costConfig,
+      }
     );
     let result: AdmissionAcquireResult;
     try {

tests/unit/adaptive-admission-features.test.ts

@@ -37,6 +37,25 @@ describe("bounded request feature extraction", () => {
     }
   });
 
+  it("measures admission bodies beyond the default size-estimator limit", () => {
+    const body = {
+      messages: Array.from({ length: 200 }, () => ({ role: "user", content: "x".repeat(10_000) })),
+    };
+    const features = extractAdmissionCostFeatures(body);
+
+    assert.ok((features.bodyBytes ?? 0) > 1_000_000);
+    assert.ok((features.estimatedInputTokens ?? 0) > 250_000);
+  });
+
+  it("uses the active admission cost budget when measuring bodies", () => {
+    const body = { payload: "x".repeat(17_000_000) };
+    const features = extractAdmissionCostFeatures(body, {
+      cost: { bodyBytesPerUnit: 1_000_000, maxRequestCost: 20 },
+    });
+
+    assert.ok((features.bodyBytes ?? 0) > 16_384_000);
+  });
+
   it("extracts production-realistic Chat, Responses, Gemini, and Antigravity shapes", () => {
     // OpenAI Chat Completions — stream omitted defaults false (higher non-stream class).
     const chat = extractAdmissionCostFeatures({
@@ -146,7 +165,7 @@ describe("bounded request feature extraction", () => {
   it("bounds tool scans and never touches entries beyond the budget (conservative count)", () => {
     // Huge leading string makes estimateSizeFast byte-exit before walking tools,
     // so only countTools can touch the tools proxy — proving its scan bound alone.
-    const sizePad = "x".repeat(300_000);
+    const sizePad = "x".repeat(17_000_000);
 
     let accesses = 0;
     const tools = new Proxy([] as unknown[], {