refactor(onboard): return ordered provider FSM results

src/lib/onboard.ts

@@ -6411,7 +6411,7 @@ async function onboard(opts: OnboardOptions = {}): Promise<void> {
         },
       },
     });
-    session = (await onboardRuntimeBoundary.recordStateResultsWithStepCompatibility([...providerInferenceResult.retryStateResults, providerInferenceResult.stateResult]), providerInferenceResult.session);
+    await onboardRuntimeBoundary.recordStateResultsWithStepCompatibility(providerInferenceResult.stateResults);
     sandboxName = providerInferenceResult.sandboxName;
     const {
       model,
@@ -6429,7 +6429,7 @@ async function onboard(opts: OnboardOptions = {}): Promise<void> {
       resume,
       fresh,
       resumeAgentChanged,
-      session,
+      session: providerInferenceResult.session,
       sandboxName,
       model,
       provider,

src/lib/onboard/machine/handlers/provider-inference.test.ts

@@ -172,6 +172,16 @@ describe("handleProviderInferenceState", () => {
       metadata: { state: "inference", provider: "nvidia-prod", model: "nvidia/test" },
     });
     expect(result.retryStateResults).toEqual([]);
+    expect(result.stateResults).toEqual([
+      {
+        type: "transition",
+        next: "inference",
+        transitionKind: "advance",
+        updates: undefined,
+        metadata: { state: "provider_selection", provider: "nvidia-prod", model: "nvidia/test" },
+      },
+      result.stateResult,
+    ]);
   });
 
   it("clears non-NVIDIA provider credentials when inference setup fails", async () => {
@@ -435,6 +445,12 @@ describe("handleProviderInferenceState", () => {
       },
     ]);
     expect(result.stateResult).toMatchObject({ next: "sandbox", transitionKind: "advance" });
+    expect(result.stateResults.map((stateResult) => [stateResult.next, stateResult.transitionKind])).toEqual([
+      ["inference", "advance"],
+      ["provider_selection", "retry"],
+      ["inference", "advance"],
+      ["sandbox", "advance"],
+    ]);
   });
 
   it("aborts before inference setup when the configuration summary is rejected", async () => {

src/lib/onboard/machine/handlers/provider-inference.ts

@@ -4,7 +4,7 @@
 import type { WebSearchConfig } from "../../../inference/web-search";
 import type { Session, SessionUpdates } from "../../../state/onboard-session";
 import { withInferenceTrace, withProviderSelectionTrace } from "../../tracing";
-import { advanceTo, retryTo, type OnboardStateTransitionResult } from "../result";
+import { advanceTo, type OnboardStateTransitionResult, retryTo } from "../result";
 
 export type ProviderInferenceRetry = { retry: "selection" } | { ok: true; retry?: undefined };
 
@@ -127,6 +127,7 @@ export interface ProviderInferenceStateResult {
   webSearchConfig: WebSearchConfig | null;
   session: Session | null;
   stateResult: OnboardStateTransitionResult;
+  stateResults: OnboardStateTransitionResult[];
   retryStateResults: OnboardStateTransitionResult[];
 }
 
@@ -177,6 +178,7 @@ export async function handleProviderInferenceState<Gpu, Agent, Host>({
   const webSearchConfig = initial.webSearchConfig;
   let forceProviderSelection = initialForceProviderSelection;
   let allowToolsIncompatible = false;
+  const stateResults: OnboardStateTransitionResult[] = [];
   const retryStateResults: OnboardStateTransitionResult[] = [];
 
   while (true) {
@@ -261,6 +263,11 @@ export async function handleProviderInferenceState<Gpu, Agent, Host>({
         }),
       );
     }
+    stateResults.push(
+      advanceTo("inference", {
+        metadata: { state: "provider_selection", provider, model },
+      }),
+    );
     env.NEMOCLAW_OPENSHELL_BIN = deps.getOpenshellBinary();
     const needsBedrockRuntimeAdapter = deps.needsBedrockRuntimeAdapter(provider, endpointUrl);
     const resumeInference =
@@ -297,11 +304,11 @@ export async function handleProviderInferenceState<Gpu, Agent, Host>({
           clearStagedCredentialEnv(deps, credentialEnv);
         }
         if (inferenceResult?.retry === "selection") {
-          retryStateResults.push(
-            retryTo("provider_selection", {
-              metadata: { state: "inference", provider, model, reason: "selection_retry" },
-            }),
-          );
+          const retryStateResult = retryTo("provider_selection", {
+            metadata: { state: "inference", provider, model, reason: "selection_retry" },
+          });
+          retryStateResults.push(retryStateResult);
+          stateResults.push(retryStateResult);
           forceProviderSelection = true;
           continue;
         }
@@ -395,11 +402,11 @@ export async function handleProviderInferenceState<Gpu, Agent, Host>({
       clearStagedCredentialEnv(deps, credentialEnv);
     }
     if (inferenceResult?.retry === "selection") {
-      retryStateResults.push(
-        retryTo("provider_selection", {
-          metadata: { state: "inference", provider, model, reason: "selection_retry" },
-        }),
-      );
+      const retryStateResult = retryTo("provider_selection", {
+        metadata: { state: "inference", provider, model, reason: "selection_retry" },
+      });
+      retryStateResults.push(retryStateResult);
+      stateResults.push(retryStateResult);
       forceProviderSelection = true;
       continue;
     }
@@ -411,6 +418,11 @@ export async function handleProviderInferenceState<Gpu, Agent, Host>({
     break;
   }
 
+  const stateResult = advanceTo("sandbox", {
+    metadata: { state: "inference", provider, model },
+  });
+  stateResults.push(stateResult);
+
   return {
     sandboxName,
     model,
@@ -423,9 +435,8 @@ export async function handleProviderInferenceState<Gpu, Agent, Host>({
     nimContainer,
     webSearchConfig,
     session,
-    stateResult: advanceTo("sandbox", {
-      metadata: { state: "inference", provider, model },
-    }),
+    stateResult,
+    stateResults,
     retryStateResults,
   };
 }