{
  "date": "2026-10-16",
  "release_selected": true,
  "implementation": {
    "app/__init__.py": "01ba4719c80b6fe911b091a7c05124b64eeece964e09c058ef8f9805daca546b",
    "app/access.py": "727b98296f2df3651b2a2e20575faca7d1b86b96e5f6cb2e6f8dc818fd7dc1e2",
    "app/analytics.py": "4c5df3108766819606e47a3d0e278a959aeeb390426d71fbce2095b17c97924b",
    "app/associative.py": "7085dd448a772a92862d33df14d0d426115bb641103423fbdf7218b3864593db",
    "app/connections.py": "7488b3ec4b2252e0d67fbf0b788faaabdc6ff9d8591548c7b6700b7bf570e5a4",
    "app/core.py": "5e0e6b6db4575588cea41909e4f5ae910761d3302c26bc4adf700362463d4bb9",
    "app/embeddings.py": "eb0e83b72454baddacafeb1ddfc7ebaa01c572226dea26210c4456ba782cb923",
    "app/engine.py": "7dab7591225a0cce11c504cda122954065d5218b54856e67d051b1acfa9e2227",
    "app/grounding.py": "41fdc8650ed092357f3c67741a85131a1c29b40888c310c718050d0bf5f6a98d",
    "app/hybrid.py": "3d0218d2acd44edfbbdad6942417cf9226e51ecf8c3d77aaa5a9243d77778e7d",
    "app/learner.py": "d3feca8e4a04a864dcc977d4f25daab05defb0a48e13c49864e3931660e31bde",
    "app/library.py": "d54b8a7b03d8004aa9656623d350cf9a5512ba85e5cf41fa454076633a259d94",
    "app/memory.py": "958bc08cdef1ed636f8fe4d3addc34f2931e4679f009de1ff398844373dbc317",
    "app/model.py": "59324c4f235081f4ef3ded5d76adf9d2a349857c7380697dc05208c625c0371b",
    "app/narration.py": "acecb8a9cae4f3f58cfbe72b9403d0efe21243d5b9d2e5e79892da64e63b2ca1",
    "app/nlu.py": "df5ede17026dc78cf7132033fa59641e50c38e505a868425775f2dc476e84887",
    "app/parse_document.py": "79847968885c902bf90e1b48ee53067afc6e7d6f38ef53a0ba7cdfd6c3bca0cd",
    "app/plan_transport.py": "7ab1d7e5fb7c50be6b878704dbae8f8636f1979f2583d157998b980de1884ba1",
    "app/planning.py": "f69b533c9776e51be5b9aa3d46be57ae085b0379f88ff77e3bbc1cbb5dc1d3d2",
    "app/privacy.py": "37f746d1194c1691b42bf344f0a5a99c6d635c842a268a93610e5ca44a421c7f",
    "app/questions.py": "978d616a50097c6ab4ce4711524d0bfc12ad323e5030309b98490ded45b57e4f",
    "app/research.py": "d4135a6e732f1accbfa113a19f74711452e82ca6a72f5f5d6a16ad6d3060968d",
    "app/research_broker.py": "6e2fd28d9200aa9f81b3f3e273e0861d9fe4756861b6d182f11e9020959024d5",
    "app/rules.py": "52d4979e81bac19274f0bbf9c54d34f683b7a6d7be45e474c3a80f30298b332e",
    "app/sections.py": "d5f71e2da59890a044ab1a12ca99cf5dbaf3846d3818d353e304584f95a37393",
    "app/server.py": "9dd01258ec448da41d1d4747e50b5cd7b14dd48ebe2f350387102408744fc5d1",
    "app/sourcepacks.py": "bca98af60f95e70d49ced8ba89e6250c8530e9560fcc5d2d7a6e2c8d4b35f330",
    "app/structured.py": "b002d38113b1645c7f7601541e5bf12206ef5abf9bd9c8ec41039b0e62f80066",
    "scripts/evaluate_dialogue_http.py": "2eca9555112118878b7124283713da631bff9dd56f6dcbe6546d8d5739d9d599"
  },
  "model": {
    "name": "Qwen3 4B Instruct 2507",
    "quantization": "Q4_K_M",
    "source": "https://huggingface.co/bartowski/Qwen_Qwen3-4B-Instruct-2507-GGUF/resolve/ae44f08e1392f39c0e474af10c3ff8355c8b6688/Qwen_Qwen3-4B-Instruct-2507-Q4_K_M.gguf",
    "original_model": "https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507",
    "revision": "ae44f08e1392f39c0e474af10c3ff8355c8b6688",
    "sha256": "2fde00ce69dd4899c70d020845e2638353015bba0fdf161b3eb965f2bca4464e",
    "bytes": 2497280736,
    "license": "Apache-2.0",
    "filename": "Qwen_Qwen3-4B-Instruct-2507-Q4_K_M.gguf",
    "downloaded_at": "2026-10-16T09:15:02.244383+00:00"
  },
  "source_packages": {
    "data/medical/digestive-de.json": "62f28f1cb006ac72ddefad3add4292fb824e884b2e1e681ba9cf8b2aca6feb1b",
    "data/medical/digestive-originals.json": "2631225435df1f06a0b139e8ecf3b6fcf1cddce3776c97d295521a1e4b579238",
    "data/medical/foundations.json": "a91604397a524a6c08f78ab885d6d9b27207f5d76e72e3c7305f4b24b1c82f36",
    "data/medical/metabolic-de.json": "f5d39f1c016d5b4bb3889511c52d1fccc23980b2b6281e65cba39d3f12912e08",
    "data/medical/metabolic-originals.json": "2ae14985fd85ef7cfde194168f23753f89b3af89b0175ff68fe854ab808c5ed4",
    "data/medical/niddk-de.json": "08ff3d7551ac2f035bec4c779c2a479d7f6a3437c98e927922ec6269978d884c",
    "data/medical/niddk-originals.json": "a1ab3365c98fc6588c4bc78fb8143602b7bbe491ff13565469a3da547a8a9c3b",
    "data/medical/pharmacology-expanded.json": "535b1f2795a22556627e52eb7e74fc9597eee20326fa4ca7ff05bceb2d16542c",
    "data/medical/wikipedia.json": "9067108be2db656bc337780f55c30f5d5219a21e324f28562fd536a516b44fc4"
  },
  "plan_encoding": "object",
  "speculation": "ngram-simple",
  "contract": "sacretus.request.v3",
  "transport": "sacretus.request.v3",
  "contract_sha256": "14672e954cd0335a2fa16cfed3b0fadbb0a140dfa3bd2b0cf82a6ab90dd01f06",
  "environment": {
    "host": "Sacretus VPS, Ubuntu, 4 vCPU / 8 GB RAM; CPU-only",
    "test_app": "Isolated loopback-only app, temporary API key and separate public/synthetic database",
    "load": "3 vCPU / Nice=10, production remained available; no parallel software suite during timing runs"
  },
  "automated_checks": {
    "passed": 318,
    "seconds": 326.79,
    "scope": "Full software suite on VPS; model doubles are not real NLU quality evidence.",
    "report": "software-suite-vps-2026-10-16.txt",
    "sha256": "9f94edd8385a3e5b5c7d3d3fdf9e17d14b5688b88e962b76cde1cf8ca4f35fcd",
    "command": "python -m pytest -q",
    "implementation_matches": {
      "app/__init__.py": "01ba4719c80b6fe911b091a7c05124b64eeece964e09c058ef8f9805daca546b",
      "app/access.py": "727b98296f2df3651b2a2e20575faca7d1b86b96e5f6cb2e6f8dc818fd7dc1e2",
      "app/analytics.py": "4c5df3108766819606e47a3d0e278a959aeeb390426d71fbce2095b17c97924b",
      "app/associative.py": "7085dd448a772a92862d33df14d0d426115bb641103423fbdf7218b3864593db",
      "app/connections.py": "7488b3ec4b2252e0d67fbf0b788faaabdc6ff9d8591548c7b6700b7bf570e5a4",
      "app/core.py": "5e0e6b6db4575588cea41909e4f5ae910761d3302c26bc4adf700362463d4bb9",
      "app/embeddings.py": "eb0e83b72454baddacafeb1ddfc7ebaa01c572226dea26210c4456ba782cb923",
      "app/engine.py": "7dab7591225a0cce11c504cda122954065d5218b54856e67d051b1acfa9e2227",
      "app/grounding.py": "41fdc8650ed092357f3c67741a85131a1c29b40888c310c718050d0bf5f6a98d",
      "app/hybrid.py": "3d0218d2acd44edfbbdad6942417cf9226e51ecf8c3d77aaa5a9243d77778e7d",
      "app/learner.py": "d3feca8e4a04a864dcc977d4f25daab05defb0a48e13c49864e3931660e31bde",
      "app/library.py": "d54b8a7b03d8004aa9656623d350cf9a5512ba85e5cf41fa454076633a259d94",
      "app/memory.py": "958bc08cdef1ed636f8fe4d3addc34f2931e4679f009de1ff398844373dbc317",
      "app/model.py": "59324c4f235081f4ef3ded5d76adf9d2a349857c7380697dc05208c625c0371b",
      "app/narration.py": "acecb8a9cae4f3f58cfbe72b9403d0efe21243d5b9d2e5e79892da64e63b2ca1",
      "app/nlu.py": "df5ede17026dc78cf7132033fa59641e50c38e505a868425775f2dc476e84887",
      "app/parse_document.py": "79847968885c902bf90e1b48ee53067afc6e7d6f38ef53a0ba7cdfd6c3bca0cd",
      "app/plan_transport.py": "7ab1d7e5fb7c50be6b878704dbae8f8636f1979f2583d157998b980de1884ba1",
      "app/planning.py": "f69b533c9776e51be5b9aa3d46be57ae085b0379f88ff77e3bbc1cbb5dc1d3d2",
      "app/privacy.py": "37f746d1194c1691b42bf344f0a5a99c6d635c842a268a93610e5ca44a421c7f",
      "app/questions.py": "978d616a50097c6ab4ce4711524d0bfc12ad323e5030309b98490ded45b57e4f",
      "app/research.py": "d4135a6e732f1accbfa113a19f74711452e82ca6a72f5f5d6a16ad6d3060968d",
      "app/research_broker.py": "6e2fd28d9200aa9f81b3f3e273e0861d9fe4756861b6d182f11e9020959024d5",
      "app/rules.py": "52d4979e81bac19274f0bbf9c54d34f683b7a6d7be45e474c3a80f30298b332e",
      "app/sections.py": "d5f71e2da59890a044ab1a12ca99cf5dbaf3846d3818d353e304584f95a37393",
      "app/server.py": "9dd01258ec448da41d1d4747e50b5cd7b14dd48ebe2f350387102408744fc5d1",
      "app/sourcepacks.py": "bca98af60f95e70d49ced8ba89e6250c8530e9560fcc5d2d7a6e2c8d4b35f330",
      "app/structured.py": "b002d38113b1645c7f7601541e5bf12206ef5abf9bd9c8ec41039b0e62f80066",
      "scripts/evaluate_dialogue_http.py": "2eca9555112118878b7124283713da631bff9dd56f6dcbe6546d8d5739d9d599"
    }
  },
  "comparison": {
    "cases": 20,
    "baseline_passed": 18,
    "candidate_passed": 19,
    "baseline_median_seconds": 36.71,
    "candidate_median_seconds": 34.45,
    "baseline": {
      "total": 20,
      "strict_assignment_passed": 18,
      "median_seconds": 36.71,
      "maximum_seconds": 90.168,
      "mean_seconds": 41.26,
      "summed_request_seconds": 825.201,
      "completion_tokens": 704,
      "completion_tokens_counted": 19,
      "cases_with_claims": 14,
      "modes": {
        "sections": 11,
        "composite": 3,
        "unavailable": 1,
        "clarify": 2,
        "boundary": 2,
        "conversation": 1
      }
    },
    "candidate": {
      "total": 20,
      "strict_assignment_passed": 19,
      "median_seconds": 34.45,
      "maximum_seconds": 69.745,
      "mean_seconds": 33.514,
      "summed_request_seconds": 670.288,
      "completion_tokens": 745,
      "completion_tokens_counted": 20,
      "cases_with_claims": 15,
      "modes": {
        "sections": 12,
        "composite": 3,
        "clarify": 2,
        "boundary": 2,
        "conversation": 1
      }
    },
    "before_report": "speed-baseline-vps-2026-10-16.json",
    "after_report": "speed-speculative-vps-2026-10-16.json",
    "fixture_sha256": "49372972cba5cfc8b47b34d9ee1d1595d2dfa19af55c03be86f92bd13533b11c",
    "same_corpus_hash": "a237f61becd7a4d26f683f5e1401f4e4636832a43390be7d04a9fa755c9e3c58",
    "development_note": "The 20 cases were frozen before the first experiment. Compact tuple transport was rejected after context regressions (9/20 vs 16/20). The selected candidate uses the original named object contract, prompt and examples, with bounded target-verified local n-gram drafting. All questions are now known developer regressions, not an independent holdout. No runtime question patterns or fixture changes were added.",
    "matched_token_usage": {
      "cases": 19,
      "baseline_completion_tokens": 704,
      "candidate_completion_tokens": 704
    }
  },
  "regressions": [
    {
      "label": "context",
      "total": 20,
      "strict_assignment_passed": 17,
      "median_seconds": 40.05,
      "maximum_seconds": 56.598,
      "mean_seconds": 39.196,
      "summed_request_seconds": 783.922,
      "completion_tokens": 730,
      "completion_tokens_counted": 20,
      "cases_with_claims": 15,
      "modes": {
        "composite": 2,
        "sections": 9,
        "definition": 3,
        "conversation": 1,
        "clarify": 4,
        "comparison": 1
      }
    },
    {
      "label": "followup",
      "total": 8,
      "strict_assignment_passed": 7,
      "median_seconds": 39.0,
      "maximum_seconds": 70.093,
      "mean_seconds": 40.695,
      "summed_request_seconds": 325.561,
      "completion_tokens": 331,
      "completion_tokens_counted": 8,
      "cases_with_claims": 7,
      "modes": {
        "sections": 4,
        "composite": 2,
        "comparison": 1,
        "boundary": 1
      }
    }
  ],
  "knowledge": {
    "documents": 107,
    "passages": 338,
    "records": 0,
    "corpus_hash": "a237f61becd7a4d26f683f5e1401f4e4636832a43390be7d04a9fa755c9e3c58"
  },
  "known_limitations": [
    "Die Testfälle stammen aus der Entwicklung und sind keine unabhängige oder klinische Validierung.",
    "Strikte Auftragszuordnung und wörtliche Quellenbelege prüfen keine vollständige medizinische Richtigkeit. Quellen können lückenhaft oder veraltet sein.",
    "Die Messreihen liefen nacheinander auf einem geteilten VPS mit weiter erreichbarer Produktion. Während der Profilierung und des Vergleichs wurden produktive Modellanfragen beobachtet; unterschiedliche Last kann die Laufzeiten beeinflussen.",
    "Auch nach der Optimierung dauern einzelne Antworten auf diesem kleinen CPU-VPS deutlich länger. Es gibt keine zugesicherte Antwortzeit.",
    "NIDDK-Bearbeitungen sind KI-unterstützte deutsche Auszüge mit Originalstellen; die unabhängige fachliche und sprachliche Prüfung steht aus.",
    "Die Fragen wurden sequenziell gestellt. Ein systematischer Mehrnutzer-Lasttest und ein zugesichertes Antwortzeitniveau stehen aus.",
    "20 neue Fälle: „Erkläre den Wirkmechanismus von Salbutamol.“ erfüllt die strikte Zuordnung nicht (sections). Details stehen im Rohbericht.",
    "Kontextregression: „Und wodurch entsteht Laktoseintoleranz?“ erfüllt die strikte Zuordnung nicht (sections). Details stehen im Rohbericht.",
    "Kontextregression: „Jetzt zu Pantoprazol: Wie funktioniert es?“ erfüllt die strikte Zuordnung nicht (sections). Details stehen im Rohbericht.",
    "Kontextregression: „Was bedeutet das?“ erfüllt die strikte Zuordnung nicht (clarify). Details stehen im Rohbericht.",
    "Zusatzregression: „Anderes Thema: Was versteht man unter Sucralfat?“ erfüllt die strikte Zuordnung nicht (sections). Details stehen im Rohbericht.",
    "Im neuen Salbutamol-Fall fehlte der erfragte Wirkmechanismus als Eigenschaft im Sprachauftrag; ausgegeben wurde ein allgemeiner Überblick. Der Fall bleibt als Fehler gewertet, auch wenn der Artikelanfang mechanistische Informationen enthält.",
    "Der Geschwindigkeitsgewinn ist begrenzt und über die Prüfsätze nicht einheitlich. Das Ziel deutlich schnellerer Antworten ist noch nicht erreicht."
  ],
  "failed_cases": [
    {
      "suite": "20 neue Fälle",
      "question": "Erkläre den Wirkmechanismus von Salbutamol.",
      "mode": "sections",
      "plan": {
        "operation": "define",
        "subjects": [
          "Salbutamol"
        ],
        "aspect": "",
        "conditions": [],
        "intent": "knowledge",
        "requests": [
          {
            "operation": "explain",
            "subjects": [
              "Salbutamol"
            ],
            "aspects": [],
            "conditions": [],
            "reference": 0
          }
        ],
        "unhandled": [],
        "schema": "sacretus.request.v3",
        "parser": "local_llm",
        "model_used": true,
        "model_sha256": "2fde00ce69dd4899c70d020845e2638353015bba0fdf161b3eb965f2bca4464e",
        "context_subjects": [],
        "context_turns": []
      },
      "checks": {
        "intent_correct": true,
        "assignment_correct": false,
        "reference_correct": true,
        "quotes_exact": true
      },
      "planner_error": null
    },
    {
      "suite": "Kontextregression",
      "question": "Und wodurch entsteht Laktoseintoleranz?",
      "mode": "sections",
      "plan": {
        "operation": "lookup",
        "subjects": [
          "Laktoseintoleranz"
        ],
        "aspect": "Ursachen",
        "conditions": [],
        "intent": "knowledge",
        "requests": [
          {
            "operation": "explain",
            "subjects": [
              "Laktoseintoleranz"
            ],
            "aspects": [
              "Ursachen"
            ],
            "conditions": [],
            "reference": 1
          }
        ],
        "unhandled": [],
        "schema": "sacretus.request.v3",
        "parser": "local_llm",
        "model_used": true,
        "model_sha256": "2fde00ce69dd4899c70d020845e2638353015bba0fdf161b3eb965f2bca4464e",
        "context_subjects": [
          "Bioverfügbarkeit",
          "Laktoseintoleranz"
        ],
        "context_turns": [
          {
            "turn": 1,
            "question": "Nenne die Symptome von Laktoseintoleranz und erkläre, was Bioverfügbarkeit bedeutet.",
            "requests": [
              {
                "operation": "explain",
                "subjects": [
                  "Bioverfügbarkeit"
                ],
                "aspects": [],
                "conditions": [],
                "reference": 0
              },
              {
                "operation": "explain",
                "subjects": [
                  "Laktoseintoleranz"
                ],
                "aspects": [
                  "Symptome"
                ],
                "conditions": [],
                "reference": 0
              }
            ],
            "outcome": "composite"
          }
        ]
      },
      "checks": {
        "intent_correct": true,
        "assignment_correct": true,
        "reference_correct": false,
        "quotes_exact": true
      },
      "planner_error": null
    },
    {
      "suite": "Kontextregression",
      "question": "Jetzt zu Pantoprazol: Wie funktioniert es?",
      "mode": "sections",
      "plan": {
        "operation": "lookup",
        "subjects": [
          "Pantoprazol"
        ],
        "aspect": "Funktion",
        "conditions": [],
        "intent": "knowledge",
        "requests": [
          {
            "operation": "explain",
            "subjects": [
              "Pantoprazol"
            ],
            "aspects": [
              "Funktion"
            ],
            "conditions": [],
            "reference": 0
          }
        ],
        "unhandled": [],
        "schema": "sacretus.request.v3",
        "parser": "local_llm",
        "model_used": true,
        "model_sha256": "2fde00ce69dd4899c70d020845e2638353015bba0fdf161b3eb965f2bca4464e",
        "context_subjects": [
          "Laktoseintoleranz"
        ],
        "context_turns": [
          {
            "turn": 1,
            "question": "Und wodurch entsteht Laktoseintoleranz?",
            "requests": [
              {
                "operation": "explain",
                "subjects": [
                  "Laktoseintoleranz"
                ],
                "aspects": [
                  "Ursachen"
                ],
                "conditions": [],
                "reference": 1
              }
            ],
            "outcome": "sections"
          }
        ]
      },
      "checks": {
        "intent_correct": true,
        "assignment_correct": false,
        "reference_correct": true,
        "quotes_exact": true
      },
      "planner_error": null
    },
    {
      "suite": "Kontextregression",
      "question": "Was bedeutet das?",
      "mode": "clarify",
      "plan": {
        "operation": "clarify",
        "subjects": [],
        "aspect": "",
        "conditions": [],
        "intent": "clarify",
        "requests": [],
        "unhandled": [],
        "schema": "sacretus.request.v3",
        "parser": "local_llm",
        "model_used": true,
        "model_sha256": "2fde00ce69dd4899c70d020845e2638353015bba0fdf161b3eb965f2bca4464e",
        "context_subjects": [
          "Bioverfügbarkeit"
        ],
        "context_turns": [
          {
            "turn": 1,
            "question": "Was ist Bioverfügbarkeit?",
            "requests": [
              {
                "operation": "explain",
                "subjects": [
                  "Bioverfügbarkeit"
                ],
                "aspects": [],
                "conditions": [],
                "reference": 0
              }
            ],
            "outcome": "definition"
          }
        ]
      },
      "checks": {
        "intent_correct": false,
        "assignment_correct": false,
        "reference_correct": true,
        "quotes_exact": true
      },
      "planner_error": null
    },
    {
      "suite": "Zusatzregression",
      "question": "Anderes Thema: Was versteht man unter Sucralfat?",
      "mode": "sections",
      "plan": {
        "operation": "lookup",
        "subjects": [
          "Sucralfat"
        ],
        "aspect": "Wirkung",
        "conditions": [],
        "intent": "knowledge",
        "requests": [
          {
            "operation": "explain",
            "subjects": [
              "Sucralfat"
            ],
            "aspects": [
              "Wirkung"
            ],
            "conditions": [],
            "reference": 0
          }
        ],
        "unhandled": [],
        "schema": "sacretus.request.v3",
        "parser": "local_llm",
        "model_used": true,
        "model_sha256": "2fde00ce69dd4899c70d020845e2638353015bba0fdf161b3eb965f2bca4464e",
        "context_subjects": [
          "Famotidin"
        ],
        "context_turns": [
          {
            "turn": 1,
            "question": "Und wie wirkt es?",
            "requests": [
              {
                "operation": "explain",
                "subjects": [
                  "Famotidin"
                ],
                "aspects": [
                  "Wirkung"
                ],
                "conditions": [],
                "reference": 1
              }
            ],
            "outcome": "sections"
          }
        ]
      },
      "checks": {
        "intent_correct": true,
        "assignment_correct": false,
        "reference_correct": true,
        "quotes_exact": true
      },
      "planner_error": null
    }
  ],
  "showcases": [
    {
      "case_index": 12,
      "label": "MEHRTEILIGE FRAGE",
      "title": "Zwei Themen. Zwei belegte Teilantworten.",
      "image": "assets/testing/multipart.jpg",
      "alt": "Sacretus zeigt die Frage zu Hypothyreose und Hyperthyreose, zwei belegte Teilfragen und den Beginn der ersten Quellenantwort.",
      "observation": "Der Sprachauftrag trennt die beiden Erkrankungen und ordnet beiden die Eigenschaft Symptome zu. Die Engine liefert zwei getrennte Auszüge aus dem lokalen Bestand.",
      "limitation": "Die Aufnahme zeigt einen sichtbaren Ausschnitt des Gesprächs. Die deutschen NIDDK-Bearbeitungen sind noch nicht unabhängig fachlich geprüft.",
      "question": "Welche Symptome haben Hypothyreose und Hyperthyreose?",
      "report": "speed-speculative-vps-2026-10-16.json",
      "run_id": "626523a3b8"
    },
    {
      "case_index": 15,
      "label": "WISSENSLÜCKE",
      "title": "Eine Teilfrage bleibt ausdrücklich offen.",
      "image": "assets/testing/knowledge-gap.jpg",
      "alt": "Die App zeigt einen belegten und einen offenen Frageteil; für den künstlichen Prüfstoff Zeta fehlt eine eindeutige Definition.",
      "observation": "Zum erfundenen Prüfstoff Zeta meldet die Engine eine Lücke. Der getrennte Pantoprazol-Auftrag findet einen vorhandenen Nebenwirkungsabschnitt.",
      "limitation": "Der enzyklopädische Arzneimitteltext ist keine vollständige Fachinformation. Er enthält unter anderem ältere Beobachtungsstudien; der Beleg beweist keine aktuelle medizinische Vollständigkeit.",
      "question": "Nenne Nebenwirkungen von Pantoprazol und erkläre Prüfstoff Zeta.",
      "report": "speed-speculative-vps-2026-10-16.json",
      "run_id": "626523a3b8"
    },
    {
      "case_index": 17,
      "label": "ANWENDUNGSGRENZE",
      "title": "Die persönliche Entscheidung bleibt offen.",
      "image": "assets/testing/clinical-boundary.jpg",
      "alt": "Künstliche Prüffrage zur persönlichen Einnahme von Apixaban und Ibuprofen mit sichtbarer Anwendungsgrenze.",
      "observation": "Das Modell ordnet die künstliche persönliche Einnahmefrage als klinische Entscheidung ein. Die Engine gibt dazu keine Therapieempfehlung aus.",
      "limitation": "Dies ist eine synthetische Prüffrage, kein Patientenfall. Die richtige Abgrenzung in diesem Beispiel belegt nicht, dass alle ähnlichen Fragen zuverlässig erkannt werden.",
      "question": "Ich nehme Apixaban. Soll ich heute zusätzlich Ibuprofen nehmen?",
      "report": "speed-speculative-vps-2026-10-16.json",
      "run_id": "626523a3b8"
    },
    {
      "case_index": 12,
      "label": "BELEGANSICHT",
      "title": "Eine Antwort behält ihren Quellenstand.",
      "image": "assets/testing/source-version.jpg",
      "alt": "Belegansicht mit Originalzitat, Quellenversion und der Kennzeichnung Quellenstand unverändert.",
      "observation": "Die Belegansicht verbindet die Aussage mit ihrem gespeicherten Originaltext und der konkreten Quellenversion. Der genaue Stand lässt sich aus der Antwort öffnen.",
      "limitation": "Unverändert beschreibt die gespeicherte Version. Es bedeutet weder aktuell nach medizinischem Konsens noch unabhängig fachlich freigegeben.",
      "question": "Welche Symptome haben Hypothyreose und Hyperthyreose?",
      "report": "speed-speculative-vps-2026-10-16.json",
      "run_id": "626523a3b8"
    }
  ],
  "runtime_parameters": {
    "threads": 3,
    "spec_type": "ngram-simple",
    "ngram_n": 3,
    "ngram_m": 8,
    "draft_max": 8,
    "target_verification": true,
    "additional_model": false
  },
  "ui_implementation": {
    "app/static/app.js": "c1a84b23230c37ffaf8736ffe3f1ec3e9959175441f8e9956574821dae462fc2",
    "app/static/connections.css": "af1a824866b15a53c2255163feaf8549c26d2604452c1bfcf51bc023b3362e8b",
    "app/static/connections.js": "f34a95f1702bb7cb4db60c3863d0778b49d00922fe207809145cea53d7ac9a26",
    "app/static/deployment.css": "4940fb1cf9c03ff6789e90f59184531cfd982311044039392bd3f57e75e94150",
    "app/static/deployment.js": "bc8d19623638f31b4ec4f79a8d9c3be2528feb32d5855592e044eacd594b008e",
    "app/static/dialogue.css": "5e799f13277efc9d657f2ba394d60a42256de3d064ba82962ffb3b2672f07722",
    "app/static/dialogue.js": "e42681dae7711acdbcb1b918a4a0f1c77cc75a69774ecd35429402efc7b64cdd",
    "app/static/experience.css": "5dda4a02678944def4aae3dec194e3ef33e56a4c47a5338fb257ddec8d8a5059",
    "app/static/experience.js": "c6a20b7549c18aad64ad2c640929ad13393cb827a9bacbcdc49aa8eb4a372c2d",
    "app/static/guide.css": "87114bd04b5a7c482af12e8d75c55e7a7edfbb974a31d204cd9ef5463f921bb2",
    "app/static/guide.js": "b1e24c77224a6b260d0f9e7146c61323375957e128b03606c0ab17a3202ab74e",
    "app/static/index.html": "319a4ad77edee54c060c6eb6218f3dd4215b019e24f7f1d3e1b17392f6fa5b04",
    "app/static/kernel.css": "e8eddea2fa5cb7c94e7dd3ab91ccd5a0edd711c02ff85a22621cbef76ca9dc4c",
    "app/static/kernel.js": "1c43772ff000fd017301311d63276795851ccbec75595155048ea220ae162256",
    "app/static/library.js": "ac59843ba33619620f29f9e3a720e5a0367a9361e9d1e47fa92fe9d611ad1059",
    "app/static/login.css": "b65827288ac9f5fd011a81c2a0fcd4649eb061d066a7877d2d3ffe92a2362ed5",
    "app/static/login.html": "79e201d3df0789827564886bf12a4597690d8a45c30bb65a008e8c91d12cc1e6",
    "app/static/login.js": "1ac1a0421f5c51e4fe33f37b774d89e8a6588f952b8b9d623d46442059c69dd2",
    "app/static/memory.js": "5bed577aba82cf1d3a50fd34368dae86b794a1736d6349776e941c9869d12408",
    "app/static/refinement.css": "c5dc76c7f8a6e02b1054a7fa3bea406510ff426bea9cc699b648b6600b882b12",
    "app/static/research.js": "e2840f4e192883eb0a90aaff3d85d5d1533a86a42c0c02ab8285adb2f925193b",
    "app/static/sacretus.css": "09613edf19f71d0b080321de9eb2247722f9e82d2a67c96bcac1aeafe5e48eff",
    "app/static/studio.css": "b96bef9dc8c420198b0b1b02a3d4b715cd9ff6af814b4cb39a7b4d66d42f8758",
    "app/static/style.css": "74b3f29ea39794dff82695da5bd66c58ad8bae6a030e14987704fd1a8878d8ce",
    "app/static/theme.css": "91b6a56d6de7567d277e1a9977a15c3066b2077337c9cc65d9db482545c5f660",
    "app/static/theme.js": "0bb64cf8aef49cea6d6a8cd61bb0827af7e27b3945c429d324a9b1121e26818d",
    "app/static/workbench.css": "c59a9cc17efbf7549ecbd81f174c68015a4a8d65dd6eaba99258226642b82a3c",
    "app/static/workbench.js": "461bab9eef58bf109447fd44e0b6f6c14eaf143554fe63e18dda318a34a86fe2"
  },
  "preserves_previously_passing_regressions": true,
  "selection_rationale": "Bounded target-verified drafting preserves the named NLU contract and all previously passing cases in the completed regressions. The paired 20-step run is modestly faster, but the wider timings are mixed. It is selected as a limited runtime improvement, not evidence that the goal of substantially faster answers has been reached. Both semantic-shortening candidates were rejected.",
  "significant_speed_goal_achieved": false,
  "rejected_optimization": {
    "plan_encoding": "compact",
    "paired_score": "19/20",
    "context_score": "9/20",
    "previous_context_score": "16/20",
    "followup_score": "8/8",
    "reason": "New scope omissions and extra contextual subjects despite lower latency; not selected for production."
  },
  "abandoned_runs": [
    {
      "report": "speed-compact-initial-context-vps-2026-10-16.json",
      "reason": "Stopped after discovering a new semantic error in the preceding completed compact trial. Incomplete run excluded from aggregate request/time counts."
    },
    {
      "report": "speed-speculative-initial-followup-vps-2026-10-16.json",
      "reason": "Paused after three requests to investigate a shorter named representation. After that representation regressed, the unchanged named target-verified implementation was resumed and the full followup set rerun. This partial run is excluded from aggregate totals."
    },
    {
      "report": "speed-sparse-initial-vps-2026-10-16.json",
      "reason": "Stopped after semantic regressions; partial run excluded from aggregate totals."
    }
  ],
  "rejected_sparse_transport": {
    "report": "speed-sparse-initial-vps-2026-10-16.json",
    "completed": false,
    "observed_requests": 9,
    "source_commit": "1a2b8c917cd81c689f79977883cae04c7082a359",
    "reason": "Stopped after properties such as Gegenanzeigen, Wechselwirkungen and Ursachen disappeared. The original named full contract was restored; no reduced-field transport is shipped."
  }
}