diff --git a/doc/plans/2026-10-06-hermes-native-runner.md b/doc/plans/2026-10-06-hermes-native-runner.md
index 8a2ac44a9a1..547067ef689 100644
--- a/doc/plans/2026-10-06-hermes-native-runner.md
+++ b/doc/plans/2026-10-06-hermes-native-runner.md
@@ -596,3 +596,54 @@ required Gemini key. The key is now mapped only when the selected matrix cell
 requires it, and the workflow credential-boundary regression covers it. All 15
 workflow tests and actionlint pass. Another current-head review and CI run are
 required after this correction.
+
+### Managed API account qualification, 2026-10-07
+
+The Google credential mapping subsequently passed all 54 current-head CI checks
+(two intentional skips) and fresh Greptile 5/5, with no open threads, at
+`ffd6bbc1ebb10c1c8ef0db875bb88d733f64649d`. The following paid account cases
+use that controller source and the previously verified image source
+`8e31afda4ff3b4c0415bc26f8bb35cec17437741`. Independent tracked-source
+comparison proves their production runtime and dependency sources identical;
+the changes are qualification fixtures, documentation, workflow credential
+mapping, and the image's explicit candidate selection. The runtime-source
+fingerprint is `9d9e39d5390de779c619004ebbce55cc57d2c1aeefead36099903d7db1d3e26a`.
+
+| Managed API account | Exact model | Campaign suffix | Result | Duration |
+| --- | --- | --- | --- | --- |
+| Anthropic | `claude-haiku-4-5-20251001` | `2d55c9944d84` | Passed | 53.780 s |
+| OpenAI | `gpt-5.6-luna` | `4953bfa797ea` | Passed | 58.760 s |
+| xAI | `grok-4.7` | `83fd6210c933` | Passed | 69.960 s |
+| Google | `gemini-2.5-flash` | `ea98e1dd517f` | Failed | 52.511 s |
+
+Each passing case independently verifies six public native-run account/model
+checks, both scoped 200-cent budgets before task creation, one successful native
+run, one final answer, and cleanup. Each has a reviewed final browser screenshot.
+All run on the compatible EC2 Linux amd64 host with catalog environment `local`;
+they do not qualify macOS or Daytona. Token receipts are available, but monetary
+cost remains unpriced. The final screenshots show the existing budget policy
+pausing the test agents after unpriced usage; reported zeros are not free runs.
+
+xAI delivered 41 reasoning deltas and 22 assistant-text deltas before its native
+terminal event. Its final screenshot renders a thought card above the answer.
+These facts prove event delivery and final rendering, not incremental browser
+render timing while the run is active. Raw reasoning is not published.
+
+Google's authenticated catalog listed Gemini 2.5 Flash, but inference returned
+404 because the account had not previously used that model. Google's
+[access notice](https://ai.google.dev/gemini-api/docs/deprecations) confirms the
+restriction. The failed browser attempt, provider error, missing usage, and
+successful cleanup remain retained. Hermes made its native HTTP retry attempts
+inside that single controller turn; the campaign did not dispatch a new paid
+attempt. The fixture now selects `gemini-3.8-flash`, whose authenticated metadata
+supports `generateContent`. That metadata is not a live qualification pass.
+The model correction passed all 94 affected tests and Product E2E TypeScript;
+it requires a new live attempt, fresh CI, and review.
+
+Private proof and campaign archives remain encrypted in the task's restricted
+S3 prefix. Credentials are supplied only through a fresh verified job-bound
+encrypted handoff. Obsolete encrypted key transfers are removed after evidence
+collection. No local Docker is used. Subscription/credential lifecycle, custom
+protocols, Bedrock, the remaining interactions/state/routines, final macOS
+campaign, clean distribution, and actual Daytona gates remain open. Hermes
+continues to be pending qualification.
diff --git a/tests/runner-e2e/hermes-api-connections.test.ts b/tests/runner-e2e/hermes-api-connections.test.ts
index 98b85787572..2ce4254654a 100644
--- a/tests/runner-e2e/hermes-api-connections.test.ts
+++ b/tests/runner-e2e/hermes-api-connections.test.ts
@@ -15,10 +15,13 @@ describe("Hermes managed API connection qualification", () => {
     expect(suite.definitionMetadata).toMatchObject({ qualification: "pending", accountMethod: "api_key", accountMode: "responsible_user", coverage: "api-account-native-completion-only", budgetMonthlyCents: 200, maximumAttemptsPerCell: 1 });
     expect(selectRunnerExecutions(parseRunnerSelectors(["--all"])).some(e => e.suite.id === suite.id)).toBe(false);
   });
-  it("pins the exact selected candidate and model in the operator admission", () => {
-    const cell = cells.find(e => e.profile.credential === "XAI_API_KEY")!;
+  it.each([
+    ["XAI_API_KEY", "grok-4.7"],
+    ["GEMINI_API_KEY", "gemini-3.8-flash"],
+  ])("pins the %s candidate and model in the operator admission", (credential, model) => {
+    const cell = cells.find(e => e.profile.credential === credential)!;
     const env = buildRunnerE2EProcessEnvironment({ PAPERCLIP_RUNNER_ACPX_QUALIFICATION: "ambient" }, [cell]);
-    expect(JSON.parse(env.PAPERCLIP_RUNNER_ACPX_QUALIFICATION!)).toEqual([{ agent: "hermes", model: "grok-4.7" }]);
+    expect(JSON.parse(env.PAPERCLIP_RUNNER_ACPX_QUALIFICATION!)).toEqual([{ agent: "hermes", model }]);
     expect(() => buildRunnerE2EProcessEnvironment({}, [{ ...cell, suite: { ...suite, manualOnly: false } }])).toThrow("explicit");
     expect(cells.every(e => e.profile.modelQualification.source === "candidate_runner_profile")).toBe(true);
     expect(buildMatrixJobs(cells).every(job => job.qualificationCandidate === "hermes")).toBe(true);
diff --git a/tests/runner-e2e/hermes-api-connections.ts b/tests/runner-e2e/hermes-api-connections.ts
index 5a8a63cf7a7..b7b1427df64 100644
--- a/tests/runner-e2e/hermes-api-connections.ts
+++ b/tests/runner-e2e/hermes-api-connections.ts
@@ -49,7 +49,9 @@ export const hermesApiConnectionChoices = [
   { provider: "anthropic", credential: "ANTHROPIC_API_KEY", model: "claude-haiku-4-5-20251001" },
   { provider: "openai", credential: "OPENAI_API_KEY", model: "gpt-5.6-luna" },
   { provider: "xai", credential: "XAI_API_KEY", model: "grok-4.7" },
-  { provider: "google", credential: "GEMINI_API_KEY", model: "gemini-2.5-flash" },
+  // Google restricts 2.5 models to prior users even when they appear in the catalog.
+  // https://ai.google.dev/gemini-api/docs/deprecations
+  { provider: "google", credential: "GEMINI_API_KEY", model: "gemini-3.8-flash" },
 ] as const satisfies readonly { provider: string; credential: RunnerProfileFixture["credential"]; model: string }[];
 
 export const hermesApiConnectionDefinitionDigest = createHash("sha256").update(