Describe the problem/error/question

The hint text below the Threshold field in the Guardrails node contradicts the actual violation logic implemented in packages/@n8n/nodes-langchain/nodes/Guardrails/helpers/model.ts.

Current hint in UI:

Inputs scoring less than this will be treated as violations

Actual logic (helpers/model.ts line 105):

const triggered = result.flagged && result.confidenceScore >= threshold;

A violation is triggered when flagged === true AND confidenceScore >= threshold — i.e., inputs scoring at or above threshold are violations, not less than.

The tooltip on the ? icon ("Minimum confidence threshold to trigger the guardrail") is correct. Only the hint text below the input contradicts both the tooltip and the code.

Impact: users configuring the threshold read the hint literally and reason about the behavior backwards (e.g., "if I set 0.9, everything scoring below 0.9 will be blocked"). That breaks intuition for both tuning false-positive/false-negative tradeoffs and debugging why a guardrail is or isn't firing.

Suggested fix — one-line change in packages/@n8n/nodes-langchain/nodes/Guardrails/description.ts:

// Before
hint: 'Inputs scoring less than this will be treated as violations',

// After (Option A — minimal, aligned with code):
hint: 'Inputs scoring at or above this threshold (when flagged) will be treated as violations',

// After (Option B — shorter, aligned with tooltip wording):
hint: 'Flagged inputs with confidence at or above this threshold will be treated as violations',

What is the error message (if any)?

No runtime error. This is a UI documentation bug — misleading hint text in the node's Threshold field description.

Please share your workflow/screenshots/recording

{
  "nodes": [
    {
      "parameters": {
        "text": "={{ $('Code Node — Input Sanitization1').item.json.emailBody }}{{JSON.stringify($json.attachmentRows) }}",
        "guardrails": {
          "jailbreak": {
            "value": {
              "threshold": 0.7
            }
          }
        },
        "customizeSystemMessage": true
      },
      "id": "766c5336-3141-40d6-a5fd-0d41390fb1d6",
      "name": "Guardrails — Input1",
      "type": "@n8n/n8n-nodes-langchain.guardrails",
      "typeVersion": 2,
      "position": [
        3984,
        2784
      ],
      "onError": "continueErrorOutput"
    }
  ],
  "connections": {
    "Guardrails — Input1": {
      "main": [
        [],
        []
      ]
    }
  },
  "pinData": {},
  "meta": {
    "instanceId": "70f8809962a15cc6cf75168d0d844c7c78da7a81cc99b5ae5fe6741c395b0cbd"
  }
}

Reproduction steps:

1. Paste the above single-node workflow (any Guardrails node variant works — Jailbreak shown here).
2. Open the node configuration panel.
3. Locate the Threshold field.
4. Observe the hint text below the input box:

   Inputs scoring less than this will be treated as violations

5. Compare with the tooltip on the ? icon next to the label:

   Minimum confidence threshold to trigger the guardrail (0.0 to 1.0)

6. Inspect the source code of the node at packages/@n8n/nodes-langchain/nodes/Guardrails/helpers/model.ts line 105 to confirm the actual logic: triggered = flagged && confidenceScore >= threshold.

Share the output returned by the last node

Example output from the Guardrails node showing the confidenceScore and flagged fields that the threshold comparison operates on (clean input, model classifies as safe):

[
  {
    "response": {
      "generations": [
        [
          {
            "text": "```json\n{\"confidenceScore\":0.0,\"flagged\":false}\n```"
          }
        ]
      ]
    },
    "tokenUsageEstimate": {
      "completionTokens": 17,
      "promptTokens": 1071,
      "totalTokens": 1088
    }
  }
]

With this output (confidenceScore: 0.0, flagged: false) and threshold = 0.7, the code at model.ts:105 computes:

triggered = false && (0.0 >= 0.7)
triggered = false

→ safe branch. Correct behavior — clean input passes. But the hint text incorrectly claims "scoring less than 0.7 = violation", which would wrongly suggest the input should be blocked.

Debug info

core

• n8nVersion: 2.13.3
• platform: docker (cloud)
• nodeJsVersion: 24.13.1
• nodeEnv: production
• database: sqlite
• executionMode: regular
• concurrency: 5
• license: enterprise (sandbox)

storage

• success: all
• error: all
• progress: false
• manual: true
• binaryMode: filesystem

pruning

• enabled: true
• maxAge: 168 hours
• maxCount: 2500 executions

client

• userAgent: mozilla/5.0 (macintosh; intel mac os x 10_15_7) applewebkit/605.1.15 (khtml, like gecko) version/18.6 safari/605.1.15
• isTouchDevice: false

Generated at: 2026-04-23T19:06:26.994Z

Additional context

• Related commit: #25390 — fix(Guardrails Node): Improve Guardrails validation
• Source of truth: packages/@n8n/nodes-langchain/nodes/Guardrails/helpers/model.ts
• Hint location: packages/@n8n/nodes-langchain/nodes/Guardrails/description.ts
• Docs page does not carry this hint, so documentation is unaffected — only the in-UI field hint.

Happy to open a PR if maintainers prefer contribution over fix-by-maintainer.