deepline 0.3.234 → 0.3.236

package/dist/bundling-sources/sdk/src/http.ts

@@ -204,6 +204,18 @@ function describeRequestTarget(path: string): string {
   return 'Deepline';
 }
 
+function networkDiagnosticUrl(value: string): string | undefined {
+  try {
+    const url = new URL(value);
+    if (url.protocol !== 'https:' && url.protocol !== 'http:') return undefined;
+    // Diagnostics never retain URL credentials, query values, or fragments.
+    return `${url.origin}${url.pathname}`;
+  } catch {
+    // Preserve the original transport failure for malformed request URLs.
+    return undefined;
+  }
+}
+
 function providerLabelFromToolId(toolId: string): string {
   const normalized = toolId.trim().toLowerCase();
   const provider = normalized.split(/[_:./-]+/)[0];
@@ -379,6 +391,10 @@ export class HttpClient {
     const baseUrl = this.config.baseUrl;
     const url = `${baseUrl}${path}`;
     const method = options?.method ?? 'GET';
+    const requestStartedAt = Date.now();
+    let fetchAttempts = 0;
+    let lastRequestUrl = url;
+    let networkFailedAt: number | undefined;
 
     const headers = this.authHeaders(options?.headers);
 
@@ -397,6 +413,7 @@ export class HttpClient {
     requestAttempts: for (let attempt = 0; attempt <= maxRetries; attempt++) {
       if (options?.signal?.aborted) {
         lastError = requestCancelledError(options.signal);
+        networkFailedAt = Date.now();
         break;
       }
       if (attempt > 0) {
@@ -410,6 +427,7 @@ export class HttpClient {
           await sleep(delayMs, options?.signal);
         } catch (error) {
           lastError = error instanceof Error ? error : new Error(String(error));
+          networkFailedAt = Date.now();
           break;
         }
       }
@@ -417,6 +435,7 @@ export class HttpClient {
       for (const candidateUrl of candidateUrls) {
         if (options?.signal?.aborted) {
           lastError = requestCancelledError(options.signal);
+          networkFailedAt = Date.now();
           break requestAttempts;
         }
         const controller = new AbortController();
@@ -432,17 +451,20 @@ export class HttpClient {
         });
 
         try {
+          const requestBody =
+            options?.formData !== undefined
+              ? typeof options.formData === 'function'
+                ? options.formData()
+                : options.formData
+              : options?.body !== undefined
+                ? JSON.stringify(options.body)
+                : undefined;
+          fetchAttempts += 1;
+          lastRequestUrl = candidateUrl;
           const response = await fetch(candidateUrl, {
             method,
             headers,
-            body:
-              options?.formData !== undefined
-                ? typeof options.formData === 'function'
-                  ? options.formData()
-                  : options.formData
-                : options?.body !== undefined
-                  ? JSON.stringify(options.body)
-                  : undefined,
+            body: requestBody,
             signal: controller.signal,
           });
 
@@ -605,6 +627,7 @@ export class HttpClient {
           return parsed as T;
         } catch (error) {
           clearTimeout(timeoutId);
+          networkFailedAt = Date.now();
           if (options?.signal?.aborted) {
             lastError = requestCancelledError(options.signal);
             break requestAttempts;
@@ -637,6 +660,19 @@ export class HttpClient {
       lastError,
       describeRequestTarget(path),
     );
+    const failedAt = networkFailedAt ?? Date.now();
+    const diagnosticUrl = networkDiagnosticUrl(lastRequestUrl);
+    const networkDetails = {
+      ...mappedNetworkError.details,
+      request: {
+        method,
+        ...(diagnosticUrl === undefined ? {} : { url: diagnosticUrl }),
+        failedAtUtc: new Date(failedAt).toISOString(),
+        // Elapsed time covers the entire logical request, including retries.
+        elapsedMs: Math.max(0, failedAt - requestStartedAt),
+        attempts: fetchAttempts,
+      },
+    };
     if (options?.toolId) {
       const code = mappedNetworkError.code;
       throw withErrorCause(
@@ -662,7 +698,7 @@ export class HttpClient {
                   ? 'unknown'
                   : 'unavailable',
             networkScope: 'client_to_deepline',
-            details: mappedNetworkError.details,
+            details: networkDetails,
             publicDetails: options.idempotencyKey
               ? { idempotencyKey: options.idempotencyKey }
               : null,
@@ -676,7 +712,7 @@ export class HttpClient {
         withCoworkNetworkHint(mappedNetworkError.message),
         undefined,
         mappedNetworkError.code,
-        mappedNetworkError.details,
+        networkDetails,
       ),
       lastError,
     );

package/dist/bundling-sources/sdk/src/release.ts

@@ -200,7 +200,7 @@ export const SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: '0.3.234',
+  version: '0.3.236',
   updateSummary:
     'Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.',
   packageCapabilities: {

package/dist/bundling-sources/shared_libs/play-runtime/play-run-recovery-policy.ts

@@ -43,6 +43,7 @@ export type PlayRunRecoveryReason =
   | 'modal_runner_start_failed_before_execution'
   | 'daytona_sandbox_transition_pending_before_execution'
   | 'daytona_sandbox_acquisition_unavailable_before_execution'
+  | 'runtime_scheduler_unavailable_before_execution'
   | 'runtime_initialization_transport_before_execution'
   | 'runtime_traffic_hold_before_execution'
   | 'runtime_traffic_throttled_before_execution';
@@ -211,6 +212,27 @@ export function recoveryForRuntimeTrafficPolicy(input: {
   });
 }
 
+/**
+ * Scheduler admission must be durable before a provider create. If that
+ * control-plane read/write is temporarily unavailable, defer through the
+ * recovery ledger instead of turning an unstarted Play into a runner failure.
+ */
+export function recoveryForRuntimeSchedulerAdmissionFailure(
+  error: unknown,
+): PlayRunRecoveryError {
+  return new PlayRunRecoveryError({
+    decision: {
+      stage: 'pre_code',
+      action: 'defer',
+      reason: 'runtime_scheduler_unavailable_before_execution',
+      dependency: 'runtime',
+    },
+    message:
+      'Runtime scheduler admission is temporarily unavailable before customer execution began.',
+    cause: error,
+  });
+}
+
 /**
  * Daytona acknowledged a named sandbox but has not finished creating it.
  * Resume that durable provider operation after a bounded defer; another

package/dist/bundling-sources/shared_libs/play-runtime/runner-backends/backends/daytona.ts

@@ -38,7 +38,10 @@ import {
   type OneShotDaytonaSandboxLifecycle,
   validateDaytonaExecutionContext,
 } from './daytona-lifecycle';
-import { recoveryForDaytonaSandboxTransitionPending } from '../../play-run-recovery-policy';
+import {
+  isPlayRunRecoveryError,
+  recoveryForDaytonaSandboxTransitionPending,
+} from '../../play-run-recovery-policy';
 import { stageDaytonaRunnerPayload } from './daytona-payload-transport';
 import {
   DaytonaRunnerInitializationError,
@@ -1819,6 +1822,9 @@ export const daytonaPlayRunnerBackend: PlayRunnerBackend = {
       if (error instanceof DaytonaSandboxAcquisitionUnavailableError) {
         throw error;
       }
+      if (isPlayRunRecoveryError(error)) {
+        throw error;
+      }
       emitDaytonaStage(callbacks, config.context, 'execute:error', {
         sandboxId: sandboxCleanup.currentSandbox()?.id ?? null,
         runnerAttempt: Math.max(0, Math.floor(config.context.runAttempt ?? 0)),

package/dist/bundling-sources/shared_libs/play-runtime/scheduler-backends/absurd-pool.ts

@@ -11,6 +11,7 @@
  * cross-call session state.
  */
 import { createRequire } from 'node:module';
+import { AsyncLocalStorage } from 'node:async_hooks';
 import { readFileSync } from 'node:fs';
 import path from 'node:path';
 import {
@@ -30,6 +31,18 @@ import {
 
 export { ABSURD_SCHEMA_VERSION };
 
+const absurdSchedulerQuerySignal = new AsyncLocalStorage<AbortSignal>();
+
+/** Applies cancellation to this async operation's Absurd SQL calls only.
+ * The SDK has no signal argument; keep cancellation in its Deepline adapter,
+ * without affecting other tasks sharing the same queryable or pool. */
+export function runWithAbsurdSchedulerQuerySignal<T>(
+  signal: AbortSignal,
+  operation: () => Promise<T>,
+): Promise<T> {
+  return absurdSchedulerQuerySignal.run(signal, operation);
+}
+
 /**
  * Builds an SDK `Queryable` backed by the hardened runtime scheduler pool.
  * Every query runs on a freshly checked-out pooled client and releases it,
@@ -43,13 +56,19 @@ export function absurdQueryableForRuntimeScheduler(
       sql: string,
       params?: unknown[],
     ): Promise<QueryResult<Row>> => {
-      return withRuntimeSchedulerClient(poolOptions, async (client) => {
-        const result = await client.query<Row & Record<string, unknown>>(
-          sql,
-          params,
-        );
-        return { rows: result.rows as Row[] };
-      });
+      const signal = absurdSchedulerQuerySignal.getStore();
+      return withRuntimeSchedulerClient(
+        poolOptions,
+        async (client) => {
+          const result = await client.query<Row & Record<string, unknown>>(
+            sql,
+            params,
+          );
+          return { rows: result.rows as Row[] };
+        },
+        undefined,
+        { signal },
+      );
     },
   };
 }

package/dist/bundling-sources/shared_libs/play-runtime/scheduler-backends/absurd-runtime-worker.ts

@@ -50,6 +50,7 @@ import { runtimeLimitExceededCause } from '../run-failure';
 import {
   recoveryForDaytonaSandboxAcquisitionUnavailable,
   recoveryForRuntimeInitializationTransportFailure,
+  recoveryForRuntimeSchedulerAdmissionFailure,
   recoveryForRuntimeTrafficPolicy,
 } from '../play-run-recovery-policy';
 import type { PlayCheckpoint } from '../ctx-types';
@@ -208,9 +209,10 @@ import {
 } from './absurd-worker-required-env';
 import {
   acquirePgNotifyHub,
+  isConnectionCategoryError,
   isRuntimeSchedulerCapacityError,
+  isRuntimeSchedulerQueryCancellationError,
   isRuntimeSchedulerRetryableTransactionError,
-  isConnectionCategoryError,
   parseRuntimeSchedulerCapacityError,
   RuntimeSchedulerCapacityError,
   runtimeSchedulerCapacityRetryAfterMs,
@@ -219,6 +221,7 @@ import {
   type RuntimeSchedulerPoolOptions,
   withRuntimeSchedulerClient,
 } from './postgres-pool';
+import { runWithAbsurdSchedulerQuerySignal } from './absurd-pool';
 import {
   createPostgresSchedulerProgressReporter,
   type PostgresSchedulerProgressReporter,
@@ -548,7 +551,10 @@ export async function runWithAbsurdExecutionLeaseHeartbeat<T>(input: {
     lastHeartbeatAt = now;
     heartbeatState.inFlight = (async () => {
       try {
-        await input.ctx.heartbeat();
+        await runWithAbsurdSchedulerQuerySignal(
+          cancellationController.signal,
+          () => input.ctx.heartbeat(),
+        );
       } catch (error) {
         if (!isTransientHeartbeatTimeout(error) || stopped) throw error;
         console.warn('[absurd-runtime.worker] lease_heartbeat_timeout_retry', {
@@ -557,10 +563,22 @@ export async function runWithAbsurdExecutionLeaseHeartbeat<T>(input: {
         });
         await sleep(heartbeatTimeoutRetryDelayMs);
         if (stopped) return;
-        await input.ctx.heartbeat();
+        await runWithAbsurdSchedulerQuerySignal(
+          cancellationController.signal,
+          () => input.ctx.heartbeat(),
+        );
       }
     })()
       .catch((error) => {
+        // The handoff deadline owns this cancellation. A cancelled local SQL
+        // call does not prove that the engine cancelled the durable task.
+        if (
+          cancellationController.signal.aborted &&
+          error === cancellationController.signal.reason &&
+          error instanceof AbsurdExecutionOperationTimeoutError
+        ) {
+          return;
+        }
         const engineTerminal =
           error instanceof CancelledTask || error instanceof FailedTask;
         if (engineTerminal) {
@@ -593,9 +611,22 @@ export async function runWithAbsurdExecutionLeaseHeartbeat<T>(input: {
   timer.unref?.();
   let operationTimeout: ReturnType<typeof setTimeout> | null = null;
   try {
-    const operation = Promise.resolve().then(() =>
-      input.operation(cancellationController.signal),
-    );
+    // A completed backend still cannot park until its in-flight lease update
+    // settles. Keep that drain inside the same supervised handoff deadline.
+    const operation = (async () => {
+      try {
+        operationResult = await Promise.resolve().then(() =>
+          input.operation(cancellationController.signal),
+        );
+      } catch (error) {
+        operationError = error;
+      } finally {
+        stopped = true;
+        stopHeartbeatTimer();
+        await heartbeatState.inFlight;
+      }
+      return operationResult;
+    })();
     if (operationTimeoutMs !== null) {
       operationResult = await Promise.race([
         operation,
@@ -617,15 +648,11 @@ export async function runWithAbsurdExecutionLeaseHeartbeat<T>(input: {
       operationResult = await operation;
     }
   } catch (error) {
-    operationError = error;
+    operationError ??= error;
   } finally {
     if (operationTimeout) clearTimeout(operationTimeout);
     stopped = true;
     stopHeartbeatTimer();
-    const pendingHeartbeat = heartbeatState.inFlight;
-    if (pendingHeartbeat) {
-      await pendingHeartbeat.catch(() => undefined);
-    }
   }
   if (heartbeatFailure) {
     if (
@@ -3490,22 +3517,85 @@ async function executeAssignedAbsurdAttempt(input: {
                     );
                   }
                   const mode = runtimeSandboxCapacityAdmissionMode();
-                  const reserved = await deps.runWithClient((client) =>
-                    reserveRuntimeSandboxOperation(
-                      client,
+                  const reservationInput = {
+                    orgId: launch.orgId,
+                    runId,
+                    provider,
+                    mode,
+                    attempt: generation,
+                    environment: 'production' as const,
+                    resumePendingDaytonaOperation: provider === 'daytona',
+                    operationKey: `sandbox-operation:${runId}:${generation}:${provider}:${providerAttempt ?? 1}`,
+                  };
+                  const reserve = () =>
+                    deps.runWithClient((client) =>
+                      reserveRuntimeSandboxOperation(
+                        client,
+                        reservationInput,
+                        options,
+                      ),
+                    );
+                  const deferForSchedulerFailure = (error: unknown): never => {
+                    console.warn(
+                      '[absurd-runtime.worker] sandbox_capacity_admission_deferred',
                       {
-                        orgId: launch.orgId,
                         runId,
                         provider,
-                        mode,
-                        attempt: generation,
-                        environment: 'production',
-                        resumePendingDaytonaOperation: provider === 'daytona',
-                        operationKey: `sandbox-operation:${runId}:${generation}:${provider}:${providerAttempt ?? 1}`,
+                        providerAttempt: providerAttempt ?? 1,
+                        error:
+                          error instanceof Error
+                            ? error.message
+                            : String(error),
                       },
-                      options,
-                    ),
-                  );
+                    );
+                    throw recoveryForRuntimeSchedulerAdmissionFailure(error);
+                  };
+                  const isRetryableSchedulerFailure = (error: unknown) =>
+                    isRuntimeSchedulerCapacityError(error) ||
+                    isConnectionCategoryError(error) ||
+                    isRuntimeSchedulerRetryableTransactionError(error) ||
+                    isRuntimeSchedulerQueryCancellationError(error);
+                  let reserved: Awaited<ReturnType<typeof reserve>> | null =
+                    null;
+                  try {
+                    reserved = await reserve();
+                  } catch (error) {
+                    if (isRuntimeSchedulerQueryCancellationError(error)) {
+                      // A query timeout has an ambiguous outcome. The stable
+                      // operation key makes this reservation safe to repeat:
+                      // a committed first attempt is read back, and a missing
+                      // one is inserted once.
+                      console.warn(
+                        '[absurd-runtime.worker] sandbox_capacity_admission_retry',
+                        {
+                          runId,
+                          provider,
+                          providerAttempt: providerAttempt ?? 1,
+                          error:
+                            error instanceof Error
+                              ? error.message
+                              : String(error),
+                        },
+                      );
+                      try {
+                        reserved = await reserve();
+                      } catch (retryError) {
+                        if (!isRetryableSchedulerFailure(retryError)) {
+                          throw retryError;
+                        }
+                        deferForSchedulerFailure(retryError);
+                      }
+                    } else if (isRetryableSchedulerFailure(error)) {
+                      deferForSchedulerFailure(error);
+                    } else {
+                      throw error;
+                    }
+                  }
+                  if (reserved === null) {
+                    throw new Error(
+                      'Runtime scheduler sandbox capacity reservation returned no result.',
+                    );
+                  }
                   if (!reserved.granted) {
                     throw new RuntimeSandboxCapacityLimitError({
                       orgId: launch.orgId,

package/dist/cli/index.js

@@ -3454,7 +3454,7 @@ var SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: "0.3.234",
+  version: "0.3.236",
   updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.",
   packageCapabilities: {
     updatePreferences: 1
@@ -4133,6 +4133,15 @@ function describeRequestTarget(path) {
   }
   return "Deepline";
 }
+function networkDiagnosticUrl(value) {
+  try {
+    const url = new URL(value);
+    if (url.protocol !== "https:" && url.protocol !== "http:") return void 0;
+    return `${url.origin}${url.pathname}`;
+  } catch {
+    return void 0;
+  }
+}
 function providerLabelFromToolId(toolId) {
   const normalized = toolId.trim().toLowerCase();
   const provider = normalized.split(/[_:./-]+/)[0];
@@ -4233,6 +4242,10 @@ var HttpClient = class {
     const baseUrl = this.config.baseUrl;
     const url = `${baseUrl}${path}`;
     const method = options?.method ?? "GET";
+    const requestStartedAt = Date.now();
+    let fetchAttempts = 0;
+    let lastRequestUrl = url;
+    let networkFailedAt;
     const headers = this.authHeaders(options?.headers);
     if (options?.body !== void 0) {
       headers["Content-Type"] = "application/json";
@@ -4244,6 +4257,7 @@ var HttpClient = class {
     requestAttempts: for (let attempt = 0; attempt <= maxRetries; attempt++) {
       if (options?.signal?.aborted) {
         lastError = requestCancelledError(options.signal);
+        networkFailedAt = Date.now();
         break;
       }
       if (attempt > 0) {
@@ -4254,12 +4268,14 @@ var HttpClient = class {
           await sleep(delayMs, options?.signal);
         } catch (error) {
           lastError = error instanceof Error ? error : new Error(String(error));
+          networkFailedAt = Date.now();
           break;
         }
       }
       for (const candidateUrl of candidateUrls) {
         if (options?.signal?.aborted) {
           lastError = requestCancelledError(options.signal);
+          networkFailedAt = Date.now();
           break requestAttempts;
         }
         const controller = new AbortController();
@@ -4274,10 +4290,13 @@ var HttpClient = class {
           once: true
         });
         try {
+          const requestBody = options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0;
+          fetchAttempts += 1;
+          lastRequestUrl = candidateUrl;
           const response = await fetch(candidateUrl, {
             method,
             headers,
-            body: options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0,
+            body: requestBody,
             signal: controller.signal
           });
           options?.onResponse?.(response);
@@ -4381,6 +4400,7 @@ var HttpClient = class {
           return parsed;
         } catch (error) {
           clearTimeout(timeoutId);
+          networkFailedAt = Date.now();
           if (options?.signal?.aborted) {
             lastError = requestCancelledError(options.signal);
             break requestAttempts;
@@ -4411,6 +4431,19 @@ var HttpClient = class {
       lastError,
       describeRequestTarget(path)
     );
+    const failedAt = networkFailedAt ?? Date.now();
+    const diagnosticUrl = networkDiagnosticUrl(lastRequestUrl);
+    const networkDetails = {
+      ...mappedNetworkError.details,
+      request: {
+        method,
+        ...diagnosticUrl === void 0 ? {} : { url: diagnosticUrl },
+        failedAtUtc: new Date(failedAt).toISOString(),
+        // Elapsed time covers the entire logical request, including retries.
+        elapsedMs: Math.max(0, failedAt - requestStartedAt),
+        attempts: fetchAttempts
+      }
+    };
     if (options?.toolId) {
       const code = mappedNetworkError.code;
       throw withErrorCause(
@@ -4431,7 +4464,7 @@ var HttpClient = class {
             retryAfterMs: null,
             networkKind: code === "NETWORK_TIMEOUT" ? "timeout" : code === "NETWORK_ABORTED" ? "unknown" : "unavailable",
             networkScope: "client_to_deepline",
-            details: mappedNetworkError.details,
+            details: networkDetails,
             publicDetails: options.idempotencyKey ? { idempotencyKey: options.idempotencyKey } : null
           }
         ),
@@ -4443,7 +4476,7 @@ var HttpClient = class {
         withCoworkNetworkHint(mappedNetworkError.message),
         void 0,
         mappedNetworkError.code,
-        mappedNetworkError.details
+        networkDetails
       ),
       lastError
     );

package/dist/cli/index.mjs

@@ -3449,7 +3449,7 @@ var SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: "0.3.234",
+  version: "0.3.236",
   updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.",
   packageCapabilities: {
     updatePreferences: 1
@@ -4128,6 +4128,15 @@ function describeRequestTarget(path) {
   }
   return "Deepline";
 }
+function networkDiagnosticUrl(value) {
+  try {
+    const url = new URL(value);
+    if (url.protocol !== "https:" && url.protocol !== "http:") return void 0;
+    return `${url.origin}${url.pathname}`;
+  } catch {
+    return void 0;
+  }
+}
 function providerLabelFromToolId(toolId) {
   const normalized = toolId.trim().toLowerCase();
   const provider = normalized.split(/[_:./-]+/)[0];
@@ -4228,6 +4237,10 @@ var HttpClient = class {
     const baseUrl = this.config.baseUrl;
     const url = `${baseUrl}${path}`;
     const method = options?.method ?? "GET";
+    const requestStartedAt = Date.now();
+    let fetchAttempts = 0;
+    let lastRequestUrl = url;
+    let networkFailedAt;
     const headers = this.authHeaders(options?.headers);
     if (options?.body !== void 0) {
       headers["Content-Type"] = "application/json";
@@ -4239,6 +4252,7 @@ var HttpClient = class {
     requestAttempts: for (let attempt = 0; attempt <= maxRetries; attempt++) {
       if (options?.signal?.aborted) {
         lastError = requestCancelledError(options.signal);
+        networkFailedAt = Date.now();
         break;
       }
       if (attempt > 0) {
@@ -4249,12 +4263,14 @@ var HttpClient = class {
           await sleep(delayMs, options?.signal);
         } catch (error) {
           lastError = error instanceof Error ? error : new Error(String(error));
+          networkFailedAt = Date.now();
           break;
         }
       }
       for (const candidateUrl of candidateUrls) {
         if (options?.signal?.aborted) {
           lastError = requestCancelledError(options.signal);
+          networkFailedAt = Date.now();
           break requestAttempts;
         }
         const controller = new AbortController();
@@ -4269,10 +4285,13 @@ var HttpClient = class {
           once: true
         });
         try {
+          const requestBody = options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0;
+          fetchAttempts += 1;
+          lastRequestUrl = candidateUrl;
           const response = await fetch(candidateUrl, {
             method,
             headers,
-            body: options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0,
+            body: requestBody,
             signal: controller.signal
           });
           options?.onResponse?.(response);
@@ -4376,6 +4395,7 @@ var HttpClient = class {
           return parsed;
         } catch (error) {
           clearTimeout(timeoutId);
+          networkFailedAt = Date.now();
           if (options?.signal?.aborted) {
             lastError = requestCancelledError(options.signal);
             break requestAttempts;
@@ -4406,6 +4426,19 @@ var HttpClient = class {
       lastError,
       describeRequestTarget(path)
     );
+    const failedAt = networkFailedAt ?? Date.now();
+    const diagnosticUrl = networkDiagnosticUrl(lastRequestUrl);
+    const networkDetails = {
+      ...mappedNetworkError.details,
+      request: {
+        method,
+        ...diagnosticUrl === void 0 ? {} : { url: diagnosticUrl },
+        failedAtUtc: new Date(failedAt).toISOString(),
+        // Elapsed time covers the entire logical request, including retries.
+        elapsedMs: Math.max(0, failedAt - requestStartedAt),
+        attempts: fetchAttempts
+      }
+    };
     if (options?.toolId) {
       const code = mappedNetworkError.code;
       throw withErrorCause(
@@ -4426,7 +4459,7 @@ var HttpClient = class {
             retryAfterMs: null,
             networkKind: code === "NETWORK_TIMEOUT" ? "timeout" : code === "NETWORK_ABORTED" ? "unknown" : "unavailable",
             networkScope: "client_to_deepline",
-            details: mappedNetworkError.details,
+            details: networkDetails,
             publicDetails: options.idempotencyKey ? { idempotencyKey: options.idempotencyKey } : null
           }
         ),
@@ -4438,7 +4471,7 @@ var HttpClient = class {
         withCoworkNetworkHint(mappedNetworkError.message),
         void 0,
         mappedNetworkError.code,
-        mappedNetworkError.details
+        networkDetails
       ),
       lastError
     );

package/dist/index.js

@@ -1267,7 +1267,7 @@ var SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: "0.3.234",
+  version: "0.3.236",
   updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.",
   packageCapabilities: {
     updatePreferences: 1
@@ -1909,6 +1909,15 @@ function describeRequestTarget(path) {
   }
   return "Deepline";
 }
+function networkDiagnosticUrl(value) {
+  try {
+    const url = new URL(value);
+    if (url.protocol !== "https:" && url.protocol !== "http:") return void 0;
+    return `${url.origin}${url.pathname}`;
+  } catch {
+    return void 0;
+  }
+}
 function providerLabelFromToolId(toolId) {
   const normalized = toolId.trim().toLowerCase();
   const provider = normalized.split(/[_:./-]+/)[0];
@@ -2009,6 +2018,10 @@ var HttpClient = class {
     const baseUrl = this.config.baseUrl;
     const url = `${baseUrl}${path}`;
     const method = options?.method ?? "GET";
+    const requestStartedAt = Date.now();
+    let fetchAttempts = 0;
+    let lastRequestUrl = url;
+    let networkFailedAt;
     const headers = this.authHeaders(options?.headers);
     if (options?.body !== void 0) {
       headers["Content-Type"] = "application/json";
@@ -2020,6 +2033,7 @@ var HttpClient = class {
     requestAttempts: for (let attempt = 0; attempt <= maxRetries; attempt++) {
       if (options?.signal?.aborted) {
         lastError = requestCancelledError(options.signal);
+        networkFailedAt = Date.now();
         break;
       }
       if (attempt > 0) {
@@ -2030,12 +2044,14 @@ var HttpClient = class {
           await sleep(delayMs, options?.signal);
         } catch (error) {
           lastError = error instanceof Error ? error : new Error(String(error));
+          networkFailedAt = Date.now();
           break;
         }
       }
       for (const candidateUrl of candidateUrls) {
         if (options?.signal?.aborted) {
           lastError = requestCancelledError(options.signal);
+          networkFailedAt = Date.now();
           break requestAttempts;
         }
         const controller = new AbortController();
@@ -2050,10 +2066,13 @@ var HttpClient = class {
           once: true
         });
         try {
+          const requestBody = options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0;
+          fetchAttempts += 1;
+          lastRequestUrl = candidateUrl;
           const response = await fetch(candidateUrl, {
             method,
             headers,
-            body: options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0,
+            body: requestBody,
             signal: controller.signal
           });
           options?.onResponse?.(response);
@@ -2157,6 +2176,7 @@ var HttpClient = class {
           return parsed;
         } catch (error) {
           clearTimeout(timeoutId);
+          networkFailedAt = Date.now();
           if (options?.signal?.aborted) {
             lastError = requestCancelledError(options.signal);
             break requestAttempts;
@@ -2187,6 +2207,19 @@ var HttpClient = class {
       lastError,
       describeRequestTarget(path)
     );
+    const failedAt = networkFailedAt ?? Date.now();
+    const diagnosticUrl = networkDiagnosticUrl(lastRequestUrl);
+    const networkDetails = {
+      ...mappedNetworkError.details,
+      request: {
+        method,
+        ...diagnosticUrl === void 0 ? {} : { url: diagnosticUrl },
+        failedAtUtc: new Date(failedAt).toISOString(),
+        // Elapsed time covers the entire logical request, including retries.
+        elapsedMs: Math.max(0, failedAt - requestStartedAt),
+        attempts: fetchAttempts
+      }
+    };
     if (options?.toolId) {
       const code = mappedNetworkError.code;
       throw withErrorCause(
@@ -2207,7 +2240,7 @@ var HttpClient = class {
             retryAfterMs: null,
             networkKind: code === "NETWORK_TIMEOUT" ? "timeout" : code === "NETWORK_ABORTED" ? "unknown" : "unavailable",
             networkScope: "client_to_deepline",
-            details: mappedNetworkError.details,
+            details: networkDetails,
             publicDetails: options.idempotencyKey ? { idempotencyKey: options.idempotencyKey } : null
           }
         ),
@@ -2219,7 +2252,7 @@ var HttpClient = class {
         withCoworkNetworkHint(mappedNetworkError.message),
         void 0,
         mappedNetworkError.code,
-        mappedNetworkError.details
+        networkDetails
       ),
       lastError
     );

package/dist/index.mjs

@@ -1171,7 +1171,7 @@ var SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: "0.3.234",
+  version: "0.3.236",
   updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.",
   packageCapabilities: {
     updatePreferences: 1
@@ -1813,6 +1813,15 @@ function describeRequestTarget(path) {
   }
   return "Deepline";
 }
+function networkDiagnosticUrl(value) {
+  try {
+    const url = new URL(value);
+    if (url.protocol !== "https:" && url.protocol !== "http:") return void 0;
+    return `${url.origin}${url.pathname}`;
+  } catch {
+    return void 0;
+  }
+}
 function providerLabelFromToolId(toolId) {
   const normalized = toolId.trim().toLowerCase();
   const provider = normalized.split(/[_:./-]+/)[0];
@@ -1913,6 +1922,10 @@ var HttpClient = class {
     const baseUrl = this.config.baseUrl;
     const url = `${baseUrl}${path}`;
     const method = options?.method ?? "GET";
+    const requestStartedAt = Date.now();
+    let fetchAttempts = 0;
+    let lastRequestUrl = url;
+    let networkFailedAt;
     const headers = this.authHeaders(options?.headers);
     if (options?.body !== void 0) {
       headers["Content-Type"] = "application/json";
@@ -1924,6 +1937,7 @@ var HttpClient = class {
     requestAttempts: for (let attempt = 0; attempt <= maxRetries; attempt++) {
       if (options?.signal?.aborted) {
         lastError = requestCancelledError(options.signal);
+        networkFailedAt = Date.now();
         break;
       }
       if (attempt > 0) {
@@ -1934,12 +1948,14 @@ var HttpClient = class {
           await sleep(delayMs, options?.signal);
         } catch (error) {
           lastError = error instanceof Error ? error : new Error(String(error));
+          networkFailedAt = Date.now();
           break;
         }
       }
       for (const candidateUrl of candidateUrls) {
         if (options?.signal?.aborted) {
           lastError = requestCancelledError(options.signal);
+          networkFailedAt = Date.now();
           break requestAttempts;
         }
         const controller = new AbortController();
@@ -1954,10 +1970,13 @@ var HttpClient = class {
           once: true
         });
         try {
+          const requestBody = options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0;
+          fetchAttempts += 1;
+          lastRequestUrl = candidateUrl;
           const response = await fetch(candidateUrl, {
             method,
             headers,
-            body: options?.formData !== void 0 ? typeof options.formData === "function" ? options.formData() : options.formData : options?.body !== void 0 ? JSON.stringify(options.body) : void 0,
+            body: requestBody,
             signal: controller.signal
           });
           options?.onResponse?.(response);
@@ -2061,6 +2080,7 @@ var HttpClient = class {
           return parsed;
         } catch (error) {
           clearTimeout(timeoutId);
+          networkFailedAt = Date.now();
           if (options?.signal?.aborted) {
             lastError = requestCancelledError(options.signal);
             break requestAttempts;
@@ -2091,6 +2111,19 @@ var HttpClient = class {
       lastError,
       describeRequestTarget(path)
     );
+    const failedAt = networkFailedAt ?? Date.now();
+    const diagnosticUrl = networkDiagnosticUrl(lastRequestUrl);
+    const networkDetails = {
+      ...mappedNetworkError.details,
+      request: {
+        method,
+        ...diagnosticUrl === void 0 ? {} : { url: diagnosticUrl },
+        failedAtUtc: new Date(failedAt).toISOString(),
+        // Elapsed time covers the entire logical request, including retries.
+        elapsedMs: Math.max(0, failedAt - requestStartedAt),
+        attempts: fetchAttempts
+      }
+    };
     if (options?.toolId) {
       const code = mappedNetworkError.code;
       throw withErrorCause(
@@ -2111,7 +2144,7 @@ var HttpClient = class {
             retryAfterMs: null,
             networkKind: code === "NETWORK_TIMEOUT" ? "timeout" : code === "NETWORK_ABORTED" ? "unknown" : "unavailable",
             networkScope: "client_to_deepline",
-            details: mappedNetworkError.details,
+            details: networkDetails,
             publicDetails: options.idempotencyKey ? { idempotencyKey: options.idempotencyKey } : null
           }
         ),
@@ -2123,7 +2156,7 @@ var HttpClient = class {
         withCoworkNetworkHint(mappedNetworkError.message),
         void 0,
         mappedNetworkError.code,
-        mappedNetworkError.details
+        networkDetails
       ),
       lastError
     );

package/dist/release.d.mts

@@ -149,7 +149,7 @@ type SdkRelease = {
     supportPolicy: SdkSupportPolicy;
 };
 declare const SDK_RELEASE: {
-    readonly version: "0.3.234";
+    readonly version: "0.3.236";
     readonly updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.";
     readonly packageCapabilities: {
         readonly updatePreferences: 1;

package/dist/release.d.ts

@@ -149,7 +149,7 @@ type SdkRelease = {
     supportPolicy: SdkSupportPolicy;
 };
 declare const SDK_RELEASE: {
-    readonly version: "0.3.234";
+    readonly version: "0.3.236";
     readonly updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.";
     readonly packageCapabilities: {
         readonly updatePreferences: 1;

package/dist/release.js

@@ -74,7 +74,7 @@ var SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: "0.3.234",
+  version: "0.3.236",
   updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.",
   packageCapabilities: {
     updatePreferences: 1

package/dist/release.mjs

@@ -48,7 +48,7 @@ var SDK_RELEASE = {
   // getters keep their established compatibility behavior.
   // 0.3.1 deprecated the legacy `deepline enrich` command in favor of Plays.
   // 0.3.90 is the first deliberately versioned SDK release for API v3.
-  version: "0.3.234",
+  version: "0.3.236",
   updateSummary: "Play grep now returns concise results with the matched fields and terms. Automatic CLI updates are enabled by default; use `deepline settings autoupdate off` to opt out, `deepline settings autoupdate on` to re-enable updates, or `deepline settings autoupdate pin <version>` to hold an exact release.",
   packageCapabilities: {
     updatePreferences: 1

package/dist/viewer/viewer.css

@@ -737,7 +737,7 @@ body {
     top: -24px;
     z-index: 20;
     display: grid;
-    grid-template-columns: minmax(180px, 1.2fr) minmax(170px, 1.2fr) repeat(4, minmax(120px, .75fr)) auto auto;
+    grid-template-columns: minmax(180px, 1.2fr) minmax(170px, 1.2fr) repeat(5, minmax(120px, .75fr)) auto auto;
     gap: 8px;
     align-items: end;
     margin: -24px -32px 18px;
@@ -902,6 +902,7 @@ body {
 .expectation-result.pass { color: var(--green); }
 .expectation-result.fail { color: var(--red); }
 .expectation-result.unknown { color: var(--yellow); }
+.grading-detail { margin-top: 3px; color: var(--text-dim); white-space: pre-wrap; }
 .expectation-failure-reason {
     margin-top: 5px;
     color: color-mix(in srgb, var(--red) 80%, var(--text-dim));

package/dist/viewer/viewer.js

@@ -1371,11 +1371,12 @@ let isLive = false;
 let liveIntervalId = null;
 let currentEval = '';
 let currentSelection = 'all';
-let sliceBy = 'run';
+let sliceBy = 'none';
 let runFilter = '';
 let agentFilter = 'all';
 let modelFilter = 'all';
 let skillFilter = 'all';
+let treatmentFilter = 'all';
 let visibleEvalKeys = [];
 let visibleSelectionKeys = [];
 
@@ -1444,6 +1445,7 @@ function attachEvalMetadata() {
             agent: firstPresent([metric && metric.agent], inferAgent(session.label, session.meta)),
             model: firstPresent([metric && metric.model_id, metric && metric.requested_model, metric && metric.model], inferModel(session.label, session.meta)),
             skill: firstPresent([metric && metric.invoke_skill], inferSkill(session)),
+            treatment: firstPresent([metric && metric.variant_id], 'Unrecorded treatment'),
         };
     });
 }
@@ -1457,7 +1459,7 @@ function sessionsForCurrentEval() {
 }
 
 function selectionDimensions() {
-    const dimensions = ['run', 'agent', 'model', 'skill'];
+    const dimensions = ['run', 'agent', 'model', 'skill', 'treatment'];
     return sliceBy === 'none' ? dimensions : dimensions.filter(dimension => dimension !== sliceBy);
 }
 
@@ -1471,6 +1473,7 @@ function selectionLabel(session) {
         agent: value => value,
         model: value => value,
         skill: value => value,
+        treatment: value => value,
     };
     return selectionDimensions().map(dimension => labels[dimension](session.dimensions[dimension])).join(' · ') || 'All transcripts';
 }
@@ -1481,7 +1484,8 @@ function matchesTopFilters(session) {
     if (agentFilter !== 'all' && d.agent !== agentFilter) return false;
     if (modelFilter !== 'all' && d.model !== modelFilter) return false;
     if (skillFilter !== 'all' && d.skill !== skillFilter) return false;
-    return !query || [session.label, d.eval, d.run, d.agent, d.model, d.skill].join(' ').toLowerCase().includes(query);
+    if (treatmentFilter !== 'all' && d.treatment !== treatmentFilter) return false;
+    return !query || [session.label, d.eval, d.run, d.agent, d.model, d.skill, d.treatment].join(' ').toLowerCase().includes(query);
 }
 
 function matchingConfigurations() {
@@ -1525,6 +1529,7 @@ function saveViewerState() {
         agentFilter: agentFilter,
         modelFilter: modelFilter,
         skillFilter: skillFilter,
+        treatmentFilter: treatmentFilter,
     }));
 }
 
@@ -1563,14 +1568,15 @@ function init() {
     currentEval = saved && visibleEvalKeys.includes(saved.currentEval)
         ? saved.currentEval
         : ((SESSIONS[Math.max(0, initialSession)] || {}).dimensions || {}).eval || visibleEvalKeys[0] || '';
-    sliceBy = saved && ['none', 'run', 'agent', 'model', 'skill'].includes(saved.sliceBy)
+    sliceBy = saved && ['none', 'run', 'agent', 'model', 'skill', 'treatment'].includes(saved.sliceBy)
         ? saved.sliceBy
-        : (SESSIONS.length > 1 ? 'run' : 'none');
+        : 'none';
     currentSelection = (saved && (saved.currentSelection || saved.currentRun)) || 'all';
     runFilter = (saved && saved.runFilter) || '';
     agentFilter = (saved && saved.agentFilter) || 'all';
     modelFilter = (saved && saved.modelFilter) || 'all';
     skillFilter = (saved && saved.skillFilter) || 'all';
+    treatmentFilter = (saved && saved.treatmentFilter) || 'all';
 
     if (SESSIONS.length > 1) {
         document.getElementById('sidebar').classList.remove('hidden');
@@ -1720,6 +1726,7 @@ function switchEval(index) {
     agentFilter = 'all';
     modelFilter = 'all';
     skillFilter = 'all';
+    treatmentFilter = 'all';
     renderSidebar();
     renderRunSidebar();
     renderWorkspace();
@@ -1744,7 +1751,7 @@ function switchSession(i) {
 
 function switchToCompare() {
     currentSelection = 'all';
-    sliceBy = 'run';
+    sliceBy = 'none';
     renderRunSidebar();
     renderWorkspace();
 }
@@ -1760,14 +1767,17 @@ function renderWorkspaceControls() {
     const agents = uniqueValues(sessions.map(s => s.dimensions.agent));
     const models = uniqueValues(sessions.map(s => s.dimensions.model));
     const skills = uniqueValues(sessions.map(s => s.dimensions.skill));
+    const treatments = uniqueValues(sessions.map(s => s.dimensions.treatment));
     return `<section class="workspace-controls" aria-label="Transcript filters">
         <div class="workspace-context"><span class="eyebrow">Evaluation</span><strong>${esc(currentEval)}</strong></div>
         <label class="control-field control-search"><span>Filter runs</span><input id="run-filter" value="${esc(runFilter)}" placeholder="model, skill, label…" oninput="runFilter=this.value" onkeydown="if(event.key==='Enter'){applyWorkspaceFilters()}"></label>
         <label class="control-field"><span>Agent</span><select id="agent-filter" onchange="agentFilter=this.value;applyWorkspaceFilters()">${optionList(agents, agentFilter, 'All agents')}</select></label>
         <label class="control-field"><span>Model</span><select id="model-filter" onchange="modelFilter=this.value;applyWorkspaceFilters()">${optionList(models, modelFilter, 'All models')}</select></label>
         <label class="control-field"><span>Skill</span><select id="skill-filter" onchange="skillFilter=this.value;applyWorkspaceFilters()">${optionList(skills, skillFilter, 'All skills')}</select></label>
+        <label class="control-field"><span>Treatment</span><select id="treatment-filter" onchange="treatmentFilter=this.value;applyWorkspaceFilters()">${optionList(treatments, treatmentFilter, 'All treatments')}</select></label>
         <label class="control-field slice-control"><span>Slice by</span><select id="slice-by" onchange="changeSlice(this.value)">
             <option value="none" ${sliceBy === 'none' ? 'selected' : ''}>No slicing</option>
+            <option value="treatment" ${sliceBy === 'treatment' ? 'selected' : ''}>Treatment</option>
             <option value="run" ${sliceBy === 'run' ? 'selected' : ''}>Run number</option>
             <option value="agent" ${sliceBy === 'agent' ? 'selected' : ''}>Agent</option>
             <option value="model" ${sliceBy === 'model' ? 'selected' : ''}>Model</option>
@@ -1790,6 +1800,9 @@ function resetWorkspaceFilters() {
     agentFilter = 'all';
     modelFilter = 'all';
     skillFilter = 'all';
+    treatmentFilter = 'all';
+    sliceBy = 'none';
+    currentSelection = 'all';
     renderRunSidebar();
     renderWorkspace();
 }
@@ -1801,12 +1814,43 @@ function changeSlice(value) {
     renderWorkspace();
 }
 
+function renderSavedGrading(grading) {
+    if (!grading || typeof grading !== 'object') return '';
+    const outcome = grading.result || {};
+    const graders = Array.isArray(grading.graders) ? grading.graders : [];
+    if (!grading.result && !graders.length) return '';
+    // Quality is an outcome, not a pass/fail check. In particular, zero is a
+    // real Low score; missing evidence and judge errors must remain ungraded.
+    const quality = outcome.score != null
+        ? `${outcome.score}/${outcome.max_score ?? 2} (${outcome.grade || 'scored'})`
+        : `Not scored (${outcome.status || 'unknown'})`;
+    const checks = outcome.checks_status || 'unknown';
+    const rows = graders.map(item => {
+        const status = item.status || 'unknown';
+        const result = item.score != null
+            ? `${item.score} (${item.grade || status})`
+            : status;
+        const resultClass = item.type === 'deterministic' && status === 'passed'
+            ? 'pass' : (status === 'failed' || status === 'error' ? 'fail' : (item.score != null ? '' : 'unknown'));
+        const errors = Array.isArray(item.evidence_errors) ? item.evidence_errors : [];
+        const details = [item.reason, item.error_kind, ...errors.map(error => [error.code, error.detail].filter(Boolean).join(': '))].filter(Boolean);
+        const reason = details.length ? `<div class="${status === 'failed' || status === 'error' ? 'expectation-failure-reason' : 'grading-detail'}">${esc(details.join(' · '))}</div>` : '';
+        return `<tr><td>${esc(item.id || 'Unnamed grader')}${reason}</td><td>${esc(item.type || 'unknown')}</td><td><span class="expectation-result ${resultClass}">${esc(result)}</span></td></tr>`;
+    }).join('');
+    return `<div class="expectations-mini${checks === 'failed' || checks === 'error' ? ' has-failures' : ''}">
+        <div class="expectations-mini-heading"><span>Outcome quality</span><strong>${esc(quality)}</strong></div>
+        <div class="expectations-mini-heading"><span>Required checks: ${esc(checks)}</span><span>Grading: ${esc(outcome.status || 'unknown')}</span></div>
+        ${rows ? `<div class="expectations-table-wrap"><table><thead><tr><th>Grader</th><th>Type</th><th>Result</th></tr></thead><tbody>${rows}</tbody></table></div>` : ''}
+    </div>`;
+}
+
 function renderExpectationMiniTable(session) {
     const metric = session.evalMetric;
+    const grading = renderSavedGrading(metric && metric.grading);
     const expectations = metric && Array.isArray(metric.expectations) ? metric.expectations : [];
     const summary = metric && metric.expectation_summary;
     if (!expectations.length) {
-        return `<div class="expectations-empty"><span>Eval results</span><strong>Pending or unavailable</strong></div>`;
+        return grading || `<div class="expectations-empty"><span>Eval results</span><strong>Pending or unavailable</strong></div>`;
     }
     const passed = summary ? summary.passed : expectations.filter(item => String(item.result || '').toUpperCase() === 'PASS').length;
     const total = summary ? summary.total : expectations.length;
@@ -1820,7 +1864,7 @@ function renderExpectationMiniTable(session) {
             : '';
         return `<tr><td>${esc(item.expectation || '—')}${reason}</td><td>${esc(item.category || 'other')}</td><td><span class="expectation-result ${resultClass}">${esc(result)}</span></td></tr>`;
     }).join('');
-    return `<div class="expectations-mini ${statusClass}">
+    return grading + `<div class="expectations-mini ${statusClass}">
         <div class="expectations-mini-heading"><span>Eval results</span><strong>${passed}/${total} passed</strong></div>
         <div class="expectations-table-wrap"><table><thead><tr><th>Expectation</th><th>Category</th><th>Result</th></tr></thead><tbody>${rows}</tbody></table></div>
     </div>`;
@@ -1887,6 +1931,14 @@ function renderFinalArtifacts(session) {
     </section>`;
 }
 
+function renderSourceProvenance(metric) {
+    if (!metric.bundle_digest) return '';
+    const fields = { 'Source run': metric.source_run, 'Bundle': metric.bundle_digest, 'Assignment': metric.assignment_id };
+    return '<details class="lane-prompt"><summary>Source provenance</summary><div class="prompt-text">'
+        + Object.entries(fields).filter(([, value]) => value).map(([key, value]) => esc(key + ': ' + value)).join('<br>')
+        + '</div></details>';
+}
+
 function renderLaneSession(session) {
     const i = session.sourceIndex;
     const d = session.dimensions;
@@ -1900,9 +1952,10 @@ function renderLaneSession(session) {
     let html = `<article class="lane-session">
         <header class="lane-session-header">
             <div class="lane-title-row"><div><span class="run-kicker">Run ${esc(d.run)}</span><h3>${esc(d.agent)} / ${esc(d.model)}</h3></div><button class="download-btn" onclick="downloadJsonl(${i})">JSONL</button></div>
-            <div class="dimension-chips"><span>${esc(d.skill)}</span><span>${formatDuration(session.stats.duration_s)}</span><span>${session.stats.tool_calls} calls</span><span class="${session.stats.tool_errors ? 'chip-error' : ''}">${session.stats.tool_errors} errors${errorRate ? ' · ' + errorRate + '%' : ''}</span>${embeddedFailures ? `<span class="chip-error">${embeddedFailures} embedded failures</span>` : ''}${infraAffected ? `<span class="chip-error">infra affected${infraCodes ? ' · ' + esc(infraCodes) : ''}</span>` : ''}</div>
+            <div class="dimension-chips"><span>${esc(d.treatment)}</span><span>${esc(d.skill)}</span><span>${formatDuration(session.stats.duration_s)}</span><span>${session.stats.tool_calls} calls</span><span class="${session.stats.tool_errors ? 'chip-error' : ''}">${session.stats.tool_errors} errors${errorRate ? ' · ' + errorRate + '%' : ''}</span>${embeddedFailures ? `<span class="chip-error">${embeddedFailures} embedded failures</span>` : ''}${infraAffected ? `<span class="chip-error">infra affected${infraCodes ? ' · ' + esc(infraCodes) : ''}</span>` : ''}</div>
             ${renderExpectationMiniTable(session)}
             ${renderInfraSignals(metric)}
+            ${renderSourceProvenance(metric)}
         </header>`;
     if (!session.transcript_available) {
         html += '<div class="transcript-unavailable">No transcript was retained for this completed eval. Its scored expectations remain included in this comparison.</div>';
@@ -1985,6 +2038,7 @@ function renderSession(i) {
     html += item('Agent', s.dimensions.agent);
     html += item('Model', s.dimensions.model || s.meta.model || '\u2014');
     html += item('Skill', s.dimensions.skill);
+    html += item('Treatment', s.dimensions.treatment);
     html += item('Run', s.dimensions.run);
     html += item('Session', (s.meta.session_id || '\u2014').slice(0, 12));
     if (s.meta.uploaded_by) html += item('Uploaded by', s.meta.uploaded_by);
@@ -2013,6 +2067,7 @@ function renderSession(i) {
     if (s.stats.loop_groups > 0) html += item('Retry Loops', s.stats.loop_groups);
     if (s.stats.error_streak_groups > 0) html += item('Error Streaks', s.stats.error_streak_groups + ' (max ' + s.stats.max_error_streak + ')');
     html += '</div>';
+    html += renderSourceProvenance(metric);
     html += renderExpectationMiniTable(s);
     html += renderInfraSignals(metric);
     if (!s.transcript_available) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "deepline",
-  "version": "0.3.234",
+  "version": "0.3.236",
   "description": "GTM data CLI and TypeScript SDK for coding agents",
   "license": "MIT",
   "homepage": "https://code.deepline.com",