{
  "dateUtc": "2026-10-22",
  "design": "One brand, no competing offers, same SaaS buyer question. Existing public offer supplied in W0-W3; hypothetical extra source in W1-W3. No live search or organic discovery.",
  "method": "Separate one-off OpenRouter chat completions harness, not production GEO engine inference. Temperature 0.3, JSON-object response format, 1200 max completion tokens. Two repeats per world/model. No tools or web retrieval.",
  "sourceTextChangeInFollowup": {
    "old": "Founding Launch is €1,499 once for the first five brands;",
    "new": "Founding Launch costs €1,499 per brand for each of the first five customers; the launch fee applies to every founding customer and is not waived;",
    "verifiedOnlyDifferenceInPairedUserPrompts": true,
    "systemPromptIdentical": true
  },
  "status": {
    "original_batch:openai/gpt-4.1-nano": {
      "recorded": 8,
      "completeValidJson": 8,
      "recommendedSuppliedOffer": 8,
      "malformedOrProviderError": 0,
      "recordedListPriceEstimateUsd": 0.0008032,
      "providerReportedCostUsd": 0.0008032
    },
    "original_batch:qwen/qwen3.5-flash-02-23": {
      "recorded": 8,
      "completeValidJson": 0,
      "recommendedSuppliedOffer": 0,
      "malformedOrProviderError": 8,
      "recordedListPriceEstimateUsd": 0.000387985,
      "providerReportedCostUsd": 0.00018681
    },
    "replacement_mistral:mistralai/mistral-small-3.2-24b-instruct": {
      "recorded": 8,
      "completeValidJson": 8,
      "recommendedSuppliedOffer": 6,
      "malformedOrProviderError": 0,
      "recordedListPriceEstimateUsd": 0.000632188,
      "providerReportedCostUsd": 0.000632188
    },
    "clarified_W1_followup:openai/gpt-4.1-nano": {
      "recorded": 2,
      "completeValidJson": 2,
      "recommendedSuppliedOffer": 2,
      "malformedOrProviderError": 0,
      "recordedListPriceEstimateUsd": 0.00022,
      "providerReportedCostUsd": 0.00022
    },
    "clarified_W1_followup:mistralai/mistral-small-3.2-24b-instruct": {
      "recorded": 2,
      "completeValidJson": 2,
      "recommendedSuppliedOffer": 2,
      "malformedOrProviderError": 0,
      "recordedListPriceEstimateUsd": 0.0001485,
      "providerReportedCostUsd": 0.0001485
    }
  },
  "balancedComparison": {
    "openai/gpt-4.1-nano": {
      "W0": {
        "completeAnswers": 2,
        "recommendedOffer": 2,
        "sourceIdsByRepeat": [
          [
            "public_offer"
          ],
          [
            "public_offer"
          ]
        ],
        "materialErrorFlags": []
      },
      "W1": {
        "completeAnswers": 2,
        "recommendedOffer": 2,
        "sourceIdsByRepeat": [
          [
            "public_offer"
          ],
          [
            "public_offer"
          ]
        ],
        "materialErrorFlags": [
          "incorrectly says €1,499 covers first five brands; package is one brand"
        ]
      },
      "W2": {
        "completeAnswers": 2,
        "recommendedOffer": 2,
        "sourceIdsByRepeat": [
          [
            "public_offer",
            "buyer_guide"
          ],
          [
            "public_offer",
            "buyer_guide"
          ]
        ],
        "materialErrorFlags": [
          "incorrectly says €1,499 covers first five brands; package is one brand"
        ]
      },
      "W3": {
        "completeAnswers": 2,
        "recommendedOffer": 2,
        "sourceIdsByRepeat": [
          [
            "public_offer",
            "independent_comparison"
          ],
          [
            "public_offer",
            "independent_comparison"
          ]
        ],
        "materialErrorFlags": []
      }
    },
    "mistralai/mistral-small-3.2-24b-instruct": {
      "W0": {
        "completeAnswers": 2,
        "recommendedOffer": 2,
        "sourceIdsByRepeat": [
          [
            "public_offer"
          ],
          [
            "public_offer"
          ]
        ],
        "materialErrorFlags": []
      },
      "W1": {
        "completeAnswers": 2,
        "recommendedOffer": 1,
        "sourceIdsByRepeat": [
          [
            "public_offer"
          ],
          [
            "public_offer"
          ]
        ],
        "materialErrorFlags": [
          "incorrectly says €499 recurring plus €1,499 launch exceed €500 recurring budget"
        ]
      },
      "W2": {
        "completeAnswers": 2,
        "recommendedOffer": 2,
        "sourceIdsByRepeat": [
          [
            "public_offer",
            "buyer_guide"
          ],
          [
            "public_offer",
            "buyer_guide"
          ]
        ],
        "materialErrorFlags": []
      },
      "W3": {
        "completeAnswers": 2,
        "recommendedOffer": 1,
        "sourceIdsByRepeat": [
          [
            "public_offer",
            "independent_comparison"
          ],
          [
            "public_offer",
            "independent_comparison"
          ]
        ],
        "materialErrorFlags": [
          "incorrectly says €499 recurring exceeds €500 recurring budget",
          "unsupported statement that no competitor fits; no competitor was supplied"
        ]
      }
    }
  },
  "observations": [
    "GPT-4.1 Nano recommended the supplied offer 8/8; Mistral Small 3.2 recommended it 6/8. Baseline already produced a positive recommendation in both models' repeats, so no incremental recommendation lift was established.",
    "GPT made first-five-brands scope errors in W1 and W2 once each. Mistral wrongly treated €499 recurring as exceeding the €500 recurring cap in W1 and W3 once each.",
    "In the separately labeled W1 pricing-clarification follow-up, GPT and Mistral had 4/4 complete positive recommendations with no flagged first-five or recurring-budget errors, versus 2 flagged material errors in their four original W1 answers. This tiny, unseeded comparison does not establish a general causal improvement.",
    "Qwen 3.5 Flash returned malformed numeric text on all eight requests, including three provider error finish reasons. It supplies no usable answer comparison.",
    "Source IDs are model references to supplied prompt material, not live retrieved citations. W2/W3 extra sources were hypothetical and unpublished."
  ],
  "manualFlagScope": "Clear false price/package/budget statements in original balanced GPT and Mistral answers; omissions and softer framing issues were not counted as correct or incorrect in this count.",
  "openRouterRecordedListPriceEstimateUsd": 0.002191873,
  "openRouterProviderReportedCostUsd": 0.001990698,
  "pricingMetadataSource": "https://openrouter.ai/api/v1/models",
  "limits": "Small supplied-context experiment. No organic visibility, search citation rate, competitive ranking, ROI, or general error rate can be inferred.",
  "harnessSha256": {
    "openrouter-run.py": "92b0a4cab3b484360a85aef6cdf0d50740095ab56f5d9cc6a976fe98b1208f0e",
    "openrouter-mistral-run.py": "c3e321f2b30c2e1eb6a1b1409b3cba43bffbd19051753bcd7361c0d7c610b122",
    "openrouter-pricing-followup.py": "81ea0be950795023f15e8bf8f3e43d3df8d6186bb4bd2951882dd06d7546f76d"
  }
}
