@danceiny/gotry 0.0.1-rc.19 → 0.0.1-rc.20

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/clients/amap.d.ts

@@ -0,0 +1,53 @@
+/** Amap (高德) Web Service API client. */
+import type { GeocodeResult, LngLat, PoiResult, RouteResult } from '../types.js';
+/** 配额类错误（QPS 或日配额超限）。retryable=true 时短期内重试可能成功
+ *  （QPS 超限），false 表示重试无意义（日配额已用尽）。 */
+export declare class AmapQuotaError extends Error {
+    readonly infocode: string;
+    readonly retryable: boolean;
+    constructor(infocode: string, message: string, retryable: boolean);
+}
+export interface AmapClientOptions {
+    key: string;
+    timeoutMs: number;
+    /** 每秒最大高德请求数（默认 2，低于高德常见 3 QPS 上限，留余量防 10021）。 */
+    maxQps?: number;
+}
+export declare class AmapClient {
+    private readonly opts;
+    private readonly limiter;
+    private readonly geoCache;
+    private readonly routeCache;
+    private readonly poiCache;
+    constructor(opts: AmapClientOptions);
+    /**
+     * 带配额保护的统一请求入口：限速排队 → 缓存命中 → 请求 → 可重试错误退避
+     * 重试 → 写入缓存。
+     */
+    private request;
+    /** 裸 GET：拼 URL、带 key、超时 + abort、错误分类。 */
+    private rawGet;
+    /**
+     * Plan a route. `mode` maps to the Amap endpoint.
+     * Transit requires city1/city2 (origin/destination city names).
+     */
+    route(origin: LngLat, destination: LngLat, mode: 'driving' | 'transit' | 'walking' | 'bicycling', opts: {
+        city1?: string;
+        city2?: string;
+    } | undefined, signal: AbortSignal): Promise<RouteResult>;
+    private transitRoute;
+    /** Forward geocode: address → coordinates. */
+    geocode(address: string, signal: AbortSignal): Promise<GeocodeResult>;
+    /** Reverse geocode: coordinates → address. */
+    reverseGeocode(location: LngLat, signal: AbortSignal): Promise<GeocodeResult>;
+    /** POI text search. */
+    poiSearch(keywords: string, opts: {
+        region?: string;
+        cityLimit?: boolean;
+    }, signal: AbortSignal): Promise<PoiResult[]>;
+    /** POI around search (nearest first by default). */
+    poiAround(location: LngLat, keywords: string, opts: {
+        radiusM?: number;
+        types?: string;
+    }, signal: AbortSignal): Promise<PoiResult[]>;
+}

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/clients/nominatim.d.ts

@@ -0,0 +1,17 @@
+/** Nominatim (OpenStreetMap) geocoding client — free fallback, 1 QPS policy. */
+import type { GeocodeResult, LngLat } from '../types.js';
+export interface NominatimClientOptions {
+    timeoutMs: number;
+    baseUrl?: string;
+    /** User-Agent identifying the client (required by Nominatim usage policy). */
+    userAgent: string;
+}
+export declare class NominatimClient {
+    private readonly opts;
+    constructor(opts: NominatimClientOptions);
+    private get;
+    /** Forward geocode. */
+    geocode(query: string, signal: AbortSignal): Promise<GeocodeResult>;
+    /** Reverse geocode. */
+    reverseGeocode(location: LngLat, signal: AbortSignal): Promise<GeocodeResult>;
+}

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/clients/osrm.d.ts

@@ -0,0 +1,15 @@
+/** OSRM public server client — free route fallback (no API key). */
+import type { LngLat, RouteResult } from '../types.js';
+export interface OsrmClientOptions {
+    timeoutMs: number;
+    baseUrl?: string;
+}
+export declare class OsrmClient {
+    private readonly opts;
+    constructor(opts: OsrmClientOptions);
+    /**
+     * Plan a route. `profile` maps to OSRM profiles: driving | walking | cycling.
+     * OSRM has no transit profile — callers must not request transit here.
+     */
+    route(origin: LngLat, destination: LngLat, profile: 'driving' | 'walking' | 'cycling', signal: AbortSignal): Promise<RouteResult>;
+}

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/clients/photon.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Photon (photon.komoot.io) geocoding client — free, keyless, reachable on CN
+ * networks (unlike nominatim.openstreetmap.org). Backed by OpenStreetMap data.
+ *
+ * Usage policy: free public service; be reasonable with request rate. Add a
+ * `lang` parameter (default zh) for localized results.
+ */
+import type { GeocodeResult, LngLat } from '../types.js';
+export interface PhotonClientOptions {
+    timeoutMs: number;
+    baseUrl?: string;
+    lang?: 'zh' | 'en';
+}
+export declare class PhotonClient {
+    private readonly opts;
+    constructor(opts: PhotonClientOptions);
+    private get;
+    /** Forward geocode. */
+    geocode(query: string, signal: AbortSignal): Promise<GeocodeResult>;
+    /** Reverse geocode. */
+    reverseGeocode(location: LngLat, signal: AbortSignal): Promise<GeocodeResult>;
+}

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/config-file.d.ts

@@ -0,0 +1,33 @@
+/**
+ * dsh-map-tools persisted config file (~/.dsh-map-tools/config.json).
+ *
+ * Follows the modlens pattern: the plugin's API keys and provider choice live
+ * in its own file (shared across profiles, never in the DSH settings
+ * document), and the settings card reads/writes it through a loopback route.
+ */
+/** The file the plugin's settings card reads and writes. Overridable for tests. */
+export declare function configPath(): string;
+/** What the plugin persists and serves to the settings card. */
+export interface MapToolsFileConfig {
+    provider?: 'amap' | 'osm';
+    amapKey?: string;
+    timeoutMs?: number;
+    maxQps?: number;
+}
+/**
+ * Read the shared config, or a thrown error. Only a missing file reads as
+ * empty: an existing-but-unparsable file is somebody's configuration, and a
+ * card that treated it as empty would overwrite it on the next save.
+ */
+export declare function readConfig(): MapToolsFileConfig;
+/** Persist a patch onto the config file, then return the new whole. */
+export declare function applyConfig(patch: Partial<MapToolsFileConfig>): MapToolsFileConfig;
+/** Whether the config file exists (the card can offer "open config file"). */
+export declare function configFileExists(): boolean;
+/** Non-secret summary served to the settings card (keys are never echoed). */
+export declare function configSummary(): {
+    provider: MapToolsFileConfig['provider'];
+    timeoutMs?: number;
+    maxQps?: number;
+    hasAmapKey: boolean;
+};

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/config-route.d.ts

@@ -0,0 +1,20 @@
+/**
+ * Loopback settings-card route for dsh-map-tools.
+ *
+ * GET  /dsh-map-tools/config            → non-secret summary (has-amap-key, …)
+ * GET  /dsh-map-tools/config?open=true  → open the config file in the editor
+ * POST /dsh-map-tools/config            → apply a provider/key/timeout patch
+ *
+ * Same-origin loopback only (mirrors dsh's own /api fence and the modlens
+ * pattern); the two halves of the card read/write a real file.
+ */
+import type { Context } from '@deepseek-ai/cordis';
+/**
+ * Register the settings-card route under the web server, when one exists.
+ *
+ * @param ctx - plugin context.
+ * @param reload - rebuild the tools after the card saves (the file write alone
+ *   never reaches the registered tool instances; without this the new provider
+ *   or key only takes effect on the next plugin reload).
+ */
+export declare function installConfigRoute(ctx: Context, reload?: () => void): void;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/config.d.ts

@@ -0,0 +1,23 @@
+/** Plugin configuration schema. */
+import Schema from '@deepseek-ai/schemastery';
+export interface Config {
+    /**
+     * Provider selection:
+     * - 'amap': 高德地图（Web 服务 API key，免费申请，推荐）
+     * - 'osm': 免费 OSM/OSRM 兜底（无 key，能力有限）
+     */
+    provider: 'amap' | 'osm';
+    /** Amap (高德) Web Service API key. */
+    amapKey?: string;
+    /** Per-request timeout in milliseconds. */
+    timeoutMs: number;
+    /** Amap requests per second cap (高德个人 key QPS 上限低，默认 2 留余量). */
+    maxQps: number;
+    /** Default route mode when a generic route is requested. */
+    defaultMode: 'driving' | 'transit' | 'walking' | 'bicycling';
+    /** Language hint for providers that support it. */
+    language: 'zh' | 'en';
+}
+/** Amap apply URL (shown as a clickable link in the settings card). */
+export declare const AMAP_APPLY_URL = "https://console.amap.com/dev/key/app";
+export declare const Config: Schema<Config>;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/index.d.ts

@@ -0,0 +1,18 @@
+/**
+ * dsh-map-tools — map & routing tools for DeepSeek Harness.
+ *
+ * Registers 7 native tools:
+ *   map_driving_route / map_transit_route / map_walking_route / map_bicycling_route
+ *   map_geocode / map_reverse_geocode / map_poi_search
+ *
+ * Data sources: OSM/OSRM/Nominatim free fallback by default (zero-key); when
+ * amapKey is configured (or provider='amap'), Amap (高德) is used for best
+ * CN data quality (transit + POI require Amap).
+ */
+import type { Context } from '@deepseek-ai/cordis';
+import { Config } from './config.js';
+import type { Config as ConfigType } from './config.js';
+export declare const name = "dsh-map-tools";
+export declare const inject: string[];
+export { Config };
+export declare function apply(ctx: Context, config: ConfigType): void;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/settings-ns.d.ts

@@ -0,0 +1,14 @@
+import type { Context } from '@deepseek-ai/cordis';
+import type { Config as ConfigType } from './config.js';
+/** Namespace the settings page keys this plugin's card to (plain string since DSH 0.1.2-rc.1). */
+export declare const MAP_TOOLS_NS = "dsh-map-tools";
+/**
+ * Wire the settings section so the card renders. The values live in the
+ * config file; the section carries only the composition entry as its base so
+ * the page has something to dispatch on.
+ *
+ * @param ctx - plugin context.
+ * @param entry - the composition entry config (schema defaults).
+ * @param reload - rebuild tools after a settings change.
+ */
+export declare function installSettingsNamespace(ctx: Context, entry: ConfigType, reload: () => void): void;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/tools/geocode.d.ts

@@ -0,0 +1,11 @@
+/** Geocoding tools: address → coordinates, and reverse. */
+import type { Context } from '@deepseek-ai/cordis';
+import type { AmapClient } from '../clients/amap.js';
+import type { NominatimClient } from '../clients/nominatim.js';
+import type { PhotonClient } from '../clients/photon.js';
+export interface GeocodeClients {
+    amap?: AmapClient;
+    nominatim?: NominatimClient;
+    photon?: PhotonClient;
+}
+export declare function registerGeocodeTools(ctx: Context, clients: GeocodeClients, disposers?: Array<() => void>): void;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/tools/poi.d.ts

@@ -0,0 +1,9 @@
+/** POI search tool. */
+import type { Context } from '@deepseek-ai/cordis';
+import type { AmapClient } from '../clients/amap.js';
+export interface PoiClients {
+    amap?: AmapClient;
+    /** Resolve an address (or `lng,lat`) to coordinates; throws with a helpful message. */
+    resolve: (text: string, signal: AbortSignal) => Promise<[number, number]>;
+}
+export declare function registerPoiTool(ctx: Context, clients: PoiClients, disposers?: Array<() => void>): void;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/tools/routes.d.ts

@@ -0,0 +1,16 @@
+/** Route planning tools: driving / transit / walking / bicycling. */
+import type { Context } from '@deepseek-ai/cordis';
+import { type AmapClient } from '../clients/amap.js';
+import type { OsrmClient } from '../clients/osrm.js';
+import type { LngLat } from '../types.js';
+/** Shared runtime handle handed to every tool (built by the plugin entry). */
+export interface MapClients {
+    amap?: AmapClient;
+    osrm?: OsrmClient;
+    /** Resolve an address (or `lng,lat`) to coordinates; throws with a helpful message. */
+    resolve: (text: string, signal: AbortSignal) => Promise<LngLat>;
+    /** Resolve the city name for a point (transit queries need city1/city2). */
+    resolveCity: (text: string, signal: AbortSignal) => Promise<string>;
+    defaultMode: 'driving' | 'transit' | 'walking' | 'bicycling';
+}
+export declare function registerRouteTools(ctx: Context, clients: MapClients, disposers?: Array<() => void>): void;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types/types.d.ts

@@ -0,0 +1,55 @@
+/** Shared types for dsh-map-tools. */
+/** A normalized geo point: [longitude, latitude]. */
+export type LngLat = [number, number];
+/** One route leg/step instruction, normalized across providers. */
+export interface RouteStep {
+    /** Human-readable instruction, e.g. "沿北京路向西行驶 300 米". */
+    instruction: string;
+    /** Distance in meters. */
+    distanceM: number;
+    /** Duration in seconds. */
+    durationS: number;
+}
+/** Normalized route result, provider-agnostic. */
+export interface RouteResult {
+    /** Provider that produced this result: 'amap' | 'osrm'. */
+    provider: 'amap' | 'osrm';
+    /** Total distance in meters. */
+    distanceM: number;
+    /** Total duration in seconds. */
+    durationS: number;
+    /** Compact polyline string (Amap format) or GeoJSON line, depending on provider. */
+    polyline: string;
+    /** Waypoint coordinates [lng, lat] along the route (sampled). */
+    points: LngLat[];
+    /** Step-by-step instructions. */
+    steps: RouteStep[];
+}
+/** Normalized geocode result. */
+export interface GeocodeResult {
+    /** Provider: 'amap' | 'nominatim' | 'photon'. */
+    provider: 'amap' | 'nominatim' | 'photon';
+    /** Matched display name. */
+    formatted: string;
+    /** Normalized [lng, lat]. */
+    location: LngLat;
+    /** Optional structured fields. */
+    city?: string;
+    district?: string;
+    adcode?: string;
+}
+/** Normalized POI search result. */
+export interface PoiResult {
+    name: string;
+    location: LngLat;
+    /** POI category/type label. */
+    type?: string;
+    address?: string;
+    tel?: string;
+    /** Distance in meters from the query location (around-search only). */
+    distanceM?: number;
+}
+/** One coordinate pair from a `lng,lat` or `lat,lng` string. */
+export declare function parseLngLat(text: string): LngLat | null;
+/** Format a pair as `lng,lat`. */
+export declare function formatLngLat([lng, lat]: LngLat): string;

package/ts/dsh-runtime/vendor/dsh-map-tools/lib/types.js

@@ -0,0 +1,16 @@
+/** Shared types for dsh-map-tools. */
+/** One coordinate pair from a `lng,lat` or `lat,lng` string. */
+export function parseLngLat(text) {
+    const m = /^\s*(-?\d+(?:\.\d+)?)\s*,\s*(-?\d+(?:\.\d+)?)\s*$/.exec(text);
+    if (!m)
+        return null;
+    const lng = Number(m[1]);
+    const lat = Number(m[2]);
+    if (lng < -180 || lng > 180 || lat < -90 || lat > 90)
+        return null;
+    return [lng, lat];
+}
+/** Format a pair as `lng,lat`. */
+export function formatLngLat([lng, lat]) {
+    return `${lng},${lat}`;
+}

package/ts/dsh-runtime/vendor/dsh-map-tools/package.json

@@ -0,0 +1,91 @@
+{
+  "name": "dsh-map-tools",
+  "description": "Map & routing tools for DeepSeek Harness: driving/transit/walking/bicycling route planning, geocoding, reverse geocoding and POI search via Amap (高德) or free OSM/OSRM.",
+  "version": "0.5.1",
+  "type": "module",
+  "main": "lib/index.js",
+  "types": "lib/types/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./lib/types/index.d.ts",
+      "default": "./lib/index.js"
+    },
+    "./client": "./client/client.js",
+    "./cordis.patch.yml": "./cordis.patch.yml",
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "lib",
+    "client",
+    "cordis.patch.yml",
+    "LICENSE"
+  ],
+  "keywords": [
+    "deepseek",
+    "harness",
+    "dsh",
+    "dsh-plugin",
+    "map",
+    "routing",
+    "path-planning",
+    "geocoding",
+    "amap",
+    "gaode",
+    "osm",
+    "osrm"
+  ],
+  "license": "MIT",
+  "author": "HorusJiang",
+  "engines": {
+    "node": ">=20"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/HorusJiang/dsh-map-tools.git"
+  },
+  "bugs": {
+    "url": "https://github.com/HorusJiang/dsh-map-tools/issues"
+  },
+  "homepage": "https://github.com/HorusJiang/dsh-map-tools#readme",
+  "peerDependencies": {
+    "@deepseek-ai/cordis": "^4.0.2",
+    "@deepseek-ai/dsh-settings": ">=0.1.2-rc.1",
+    "@deepseek-ai/dsh-tools": ">=0.1.2-rc.1"
+  },
+  "dependencies": {
+    "@deepseek-ai/schemastery": "^3.18.2"
+  },
+  "devDependencies": {
+    "@deepseek-ai/cordis": "^4.0.2",
+    "@deepseek-ai/dsh-settings": ">=0.1.2-rc.1",
+    "@deepseek-ai/dsh-tools": ">=0.1.2-rc.1",
+    "@deepseek-ai/schemastery": "^3.18.2",
+    "@types/node": "^22.0.0",
+    "typescript": "^5.6.0",
+    "vitest": "^4.1.10"
+  },
+  "publishConfig": {
+    "registry": "https://registry.npmjs.org/"
+  },
+  "dsh": {
+    "bundle": {
+      "patch": "./cordis.patch.yml"
+    },
+    "client": {
+      "platform": "web"
+    },
+    "compatibility": {
+      "dshReleases": {
+        "0.1.2-rc.1": "compatible",
+        "0.1.3-alpha.1": "compatible"
+      }
+    }
+  },
+  "scripts": {
+    "build": "tsc -p tsconfig.json",
+    "test": "vitest run",
+    "prepack": "npm run build",
+    "hooks": "node scripts/install-hooks.mjs",
+    "check:secrets": "node scripts/check-secrets.mjs"
+  }
+}

package/ts/package.json

@@ -11,11 +11,11 @@
   },
   "dependencies": {
     "@deepseek-ai/cordis": "4.0.2",
+    "@deepseek-ai/dsh-session": "0.1.2-alpha.3",
     "@deepseek-ai/dsh-tools": "0.1.2-alpha.3",
     "@deepseek-ai/schemastery": "^3.18.2",
     "@types/better-sqlite3": "^9.6.0",
     "better-sqlite3": "^13.0.3",
-    "dsh-map-tools": "0.5.1",
     "z3-solver": "^5.2.0"
   },
   "devDependencies": {
@@ -24,5 +24,13 @@
     "puppeteer-core": "^25.9.0",
     "tsx": "^4.19.0",
     "typescript": "^5.6.0"
+  },
+  "overrides": {
+    "@deepseek-ai/dsh-brand": "0.1.2-alpha.3",
+    "@deepseek-ai/dsh-llm": "0.1.2-alpha.3",
+    "@deepseek-ai/dsh-timeout": "0.1.2-alpha.3",
+    "@deepseek-ai/dsh-typert-protocol": "0.1.2-alpha.3",
+    "@deepseek-ai/dsh-util-crypto": "0.1.2-alpha.3",
+    "@deepseek-ai/dsh-util-values": "0.1.2-alpha.3"
   }
 }

package/ts/src/benchmark-agent-conformance.ts

@@ -22,6 +22,8 @@ export const BENCHMARK_CONFORMANCE_STATE_UNAVAILABLE = 'BENCHMARK_CONFORMANCE_ST
 
 const IDENTIFIER = /^[A-Za-z][A-Za-z0-9_.-]*$/
 const MAX_TERMINAL_BYTES = 1024 * 1024
+const BENCHMARK_TOOL_RESULT_SCHEMA_VERSION = 'gotry_benchmark_tool_result_v1'
+const BENCHMARK_DOMAIN_RECOVERIES = new Set(['none', 'retry_same', 'revise_arguments', 'choose_alternative'])
 
 export interface TerminalOutputConfig {
   tag: string
@@ -46,6 +48,8 @@ interface TurnState {
   retryCount: number
   retryMode: RetryMode
   successfulResultStep?: number
+  domainOutcomeStep?: number
+  lastBridgeResponseStep?: number
   bridgeCallFailed: boolean
   bridgeFailureCode?: string
   retryRedispatchAttempted: boolean
@@ -150,7 +154,9 @@ function assistantText(data: Record<string, unknown>): { text: string; interrupt
   return { text, interrupted: data.interrupted === true }
 }
 
-function bridgeResultStatus(data: Record<string, unknown>, callId: string): { ok: true } | { ok: false; code: string } | undefined {
+type BridgeResultStatus = { kind: 'result' } | { kind: 'domain' } | { kind: 'failure'; code: string }
+
+function bridgeResultStatus(data: Record<string, unknown>, callId: string): BridgeResultStatus | undefined {
   if (!plainObject(data.message) || !plainObject(data.message.source)) return undefined
   if (data.message.source.callId !== callId || !Array.isArray(data.message.content)) return undefined
   const block = data.message.content.find(candidate => plainObject(candidate)
@@ -164,8 +170,28 @@ function bridgeResultStatus(data: Record<string, unknown>, callId: string): { ok
   try {
     const parsed: unknown = JSON.parse(text)
     if (!plainObject(parsed)) return undefined
-    if (parsed.ok === true) return { ok: true }
+    const keys = Object.keys(parsed).sort().join(',')
+    if (parsed.ok === true && keys === 'ok,result' && (plainObject(parsed.result) || Array.isArray(parsed.result))) return { kind: 'result' }
+    if (parsed.ok === true && keys === 'ok,outcome' && plainObject(parsed.outcome)) {
+      const outcomeKeys = Object.keys(parsed.outcome).sort().join(',')
+      if (outcomeKeys === 'code,recovery,schema_version,status'
+        && parsed.outcome.schema_version === BENCHMARK_TOOL_RESULT_SCHEMA_VERSION
+        && (parsed.outcome.status === 'miss' || parsed.outcome.status === 'error')
+        && typeof parsed.outcome.code === 'string'
+        && IDENTIFIER.test(parsed.outcome.code)
+        && typeof parsed.outcome.recovery === 'string'
+        && BENCHMARK_DOMAIN_RECOVERIES.has(parsed.outcome.recovery)) return { kind: 'domain' }
+      return undefined
+    }
     if (parsed.ok === false && typeof parsed.error === 'string') {
+      const exactFailure = parsed.error === 'runner_failed'
+        ? (keys === 'error,ok' || (keys === 'error,exit_code,ok,signal'
+          && (parsed.exit_code === null || (typeof parsed.exit_code === 'number' && Number.isInteger(parsed.exit_code)))
+          && (parsed.signal === null || typeof parsed.signal === 'string')))
+        : parsed.error === 'invalid_arguments'
+          ? keys === 'error,ok,reason' && typeof parsed.reason === 'string'
+          : keys === 'error,ok'
+      if (!exactFailure) return undefined
       const code = parsed.error === 'timed_out'
         ? BENCHMARK_BRIDGE_TIMED_OUT
         : parsed.error === 'runner_failed'
@@ -175,7 +201,7 @@ function bridgeResultStatus(data: Record<string, unknown>, callId: string): { ok
             : parsed.error === 'output_truncated'
               ? BENCHMARK_BRIDGE_OUTPUT_TRUNCATED
             : BENCHMARK_BRIDGE_CALL_FAILED
-      return { ok: false, code }
+      return { kind: 'failure', code }
     }
     return undefined
   } catch {
@@ -249,10 +275,15 @@ export function createBenchmarkAgentConformance(projection: BenchmarkBridgeProje
         const callId = event.data.message.source.callId
         if (typeof callId !== 'string' || !state.validCallIds.has(callId)) return
         const resultStatus = bridgeResultStatus(event.data, callId)
-        if (resultStatus?.ok === true) {
+        if (typeof event.data.step === 'number') state.lastBridgeResponseStep = event.data.step
+        if (resultStatus?.kind === 'result') {
           state.successfulResultStep = typeof event.data.step === 'number'
             ? event.data.step
             : state.successfulResultStep
+        } else if (resultStatus?.kind === 'domain') {
+          state.domainOutcomeStep = typeof event.data.step === 'number'
+            ? event.data.step
+            : state.domainOutcomeStep
         } else {
           state.bridgeCallFailed = true
           state.bridgeFailureCode = resultStatus?.code ?? BENCHMARK_BRIDGE_CALL_FAILED
@@ -274,10 +305,10 @@ export function createBenchmarkAgentConformance(projection: BenchmarkBridgeProje
       if (state.retryRedispatchAttempted) {
         return { kind: 'reject', code: BENCHMARK_BRIDGE_RETRY_CALL_NOT_ALLOWED }
       }
-      if (state.successfulResultStep === undefined) {
-        if (state.bridgeCallFailed) {
-          return { kind: 'reject', code: state.bridgeFailureCode ?? BENCHMARK_BRIDGE_CALL_FAILED }
-        }
+      if (state.bridgeCallFailed && state.successfulResultStep === undefined) {
+        return { kind: 'reject', code: state.bridgeFailureCode ?? BENCHMARK_BRIDGE_CALL_FAILED }
+      }
+      if (state.successfulResultStep === undefined && state.domainOutcomeStep === undefined) {
         if (state.retryCount >= MAX_CONFORMANCE_RETRIES) {
           return { kind: 'reject', code: BENCHMARK_BRIDGE_CALL_REQUIRED }
         }
@@ -289,7 +320,7 @@ export function createBenchmarkAgentConformance(projection: BenchmarkBridgeProje
       const terminal = state.lastAssistant
       const terminalIsValid = terminal !== undefined
         && !terminal.interrupted
-        && terminal.step > state.successfulResultStep
+        && terminal.step > state.lastBridgeResponseStep!
         && parseBenchmarkTerminal(terminal.text, projection.terminal).ok
       if (terminalIsValid) return { kind: 'accept' }
       if (state.retryCount >= MAX_CONFORMANCE_RETRIES) {