{
 "_published": {
  "source": "docs/measurements/ollama-anthropic-messages-probe-2026-11-15.json",
  "published": "2026-11-15",
  "publisher": "J4SGON S.L., vorluxai.com",
  "index": "https://vorluxai.com/data/",
  "cleaned": "Copied from our internal repository by scripts/publish_measurements.py. Every number is unchanged. Redacted: private IP addresses -> <prime> / <second-box> / <private-ip> (2). Hostnames (spark-43d5, spark-7180) and the PRIME / spark2 labels are kept: the posts use them.",
  "redactions": {
   "private_ip": 2
  }
 },
 "title": "Ollama /v1/messages (Anthropic-compatible) probed with coding-agent style tool calls",
 "date": "2026-11-15",
 "started_at": "2026-11-15T19:53:47+00:00",
 "label": "spark2 (spark-7180), Ollama 0.34.0, OLLAMA_NUM_PARALLEL=8, from spark-43d5 over the fabric",
 "command": "python3 scripts/bench/anthropic_messages_probe.py --ollama http://<second-box>:11434 --small llama3.1:8b-judge --small-ctx 6144 --resident gemma4-32k:latest --label spark2 (spark-7180), Ollama 0.34.0, OLLAMA_NUM_PARALLEL=8, from spark-43d5 over the fabric --out docs/measurements/ollama-anthropic-messages-probe-2026-11-15.json",
 "method": "every request is a POST to /v1/messages with temperature 0, the system prompt and the three tool definitions recorded here; checks are deterministic (stop_reason, block types, tool name, argument substring). One run per case, no repetitions. This is not a Claude Code session: Claude Code sends a far longer system prompt and many more tools.",
 "ollama": "http://<second-box>:11434",
 "ollama_version": "0.34.0",
 "system_prompt": "You are a coding agent working in a software repository. Use the tools when you need to read files, search or run commands. Answer directly when no tool is needed.",
 "tools": [
  {
   "name": "Read",
   "description": "Read a file from the repository and return its contents.",
   "input_schema": {
    "type": "object",
    "properties": {
     "file_path": {
      "type": "string"
     }
    },
    "required": [
     "file_path"
    ]
   }
  },
  {
   "name": "Bash",
   "description": "Run a shell command in the repository and return its output.",
   "input_schema": {
    "type": "object",
    "properties": {
     "command": {
      "type": "string"
     }
    },
    "required": [
     "command"
    ]
   }
  },
  {
   "name": "Grep",
   "description": "Search file contents in the repository for a pattern.",
   "input_schema": {
    "type": "object",
    "properties": {
     "pattern": {
      "type": "string"
     },
     "path": {
      "type": "string"
     }
    },
    "required": [
     "pattern"
    ]
   }
  }
 ],
 "resident_before": [
  {
   "name": "gemma4-32k:latest",
   "size_bytes": 22703966780,
   "size_vram_bytes": 22703966780,
   "context_length": 32768
  }
 ],
 "unknown_model": {
  "status": 404,
  "body": {
   "type": "error",
   "error": {
    "type": "not_found_error",
    "message": "model 'claude-haiku-4-5' not found"
   },
   "request_id": "req_2ea061c3072656fd2e794348"
  },
  "wall_s": 0.0
 },
 "count_tokens": {
  "status": 404,
  "body": "404 page not found",
  "wall_s": 0.0
 },
 "small_model": {
  "model": "llama3.1:8b-judge",
  "num_ctx_requested": 6144,
  "load_status": 200,
  "load_wall_s": 2.13,
  "ps_after_load": [
   {
    "name": "llama3.1:8b-judge",
    "size_bytes": 11538037473,
    "size_vram_bytes": 11538037473,
    "context_length": 6144
   }
  ],
  "ps_after_messages_requests": [
   {
    "name": "llama3.1:8b-judge",
    "size_bytes": 11538037473,
    "size_vram_bytes": 11538037473,
    "context_length": 6144
   }
  ]
 },
 "tool_cases": [
  {
   "model": "llama3.1:8b-judge",
   "passed": 4,
   "total": 6,
   "cases": [
    {
     "id": "read_file",
     "expect": "tool",
     "wall_s": 0.52,
     "block_types": [
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 297,
      "cache_read_input_tokens": 0,
      "output_tokens": 19
     },
     "tool_calls": [
      {
       "name": "Read",
       "input": {
        "file_path": "src/parser.py"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": true,
     "why": "tool_use with the expected tool and argument"
    },
    {
     "id": "run_tests",
     "expect": "tool",
     "wall_s": 0.5,
     "block_types": [
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 18,
      "cache_read_input_tokens": 281,
      "output_tokens": 21
     },
     "tool_calls": [
      {
       "name": "Bash",
       "input": {
        "command": "pytest && pytest --version"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": true,
     "why": "tool_use with the expected tool and argument"
    },
    {
     "id": "search_usage",
     "expect": "tool",
     "wall_s": 0.57,
     "block_types": [
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 19,
      "cache_read_input_tokens": 281,
      "output_tokens": 24
     },
     "tool_calls": [
      {
       "name": "Grep",
       "input": {
        "path": "repository",
        "pattern": "load_config"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": true,
     "why": "tool_use with the expected tool and argument"
    },
    {
     "id": "three_words",
     "expect": "text",
     "wall_s": 0.52,
     "block_types": [
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 15,
      "cache_read_input_tokens": 281,
      "output_tokens": 22
     },
     "tool_calls": [
      {
       "name": "Bash",
       "input": {
        "command": "echo 'local model ready'"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": false,
     "why": "called a tool where plain text was expected"
    },
    {
     "id": "plain_question",
     "expect": "text",
     "wall_s": 0.61,
     "block_types": [
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 16,
      "cache_read_input_tokens": 281,
      "output_tokens": 26
     },
     "tool_calls": [
      {
       "name": "Bash",
       "input": {
        "command": "echo 'JSON stands for JavaScript Object Notation'"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": false,
     "why": "called a tool where plain text was expected"
    },
    {
     "id": "tool_result_round_trip",
     "expect": "text",
     "wall_s": 0.08,
     "block_types": [
      "text"
     ],
     "stop_reason": "end_turn",
     "usage": {
      "input_tokens": 68,
      "cache_read_input_tokens": 86,
      "output_tokens": 2
     },
     "tool_calls": [],
     "text": "2",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": false,
     "ok": true,
     "why": "plain text with the expected content"
    }
   ]
  },
  {
   "model": "gemma4-32k:latest",
   "passed": 5,
   "total": 6,
   "cases": [
    {
     "id": "read_file",
     "expect": "tool",
     "wall_s": 0.47,
     "block_types": [
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 214,
      "cache_read_input_tokens": 0,
      "output_tokens": 23
     },
     "tool_calls": [
      {
       "name": "Read",
       "input": {
        "file_path": "src/parser.py"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": true,
     "why": "tool_use with the expected tool and argument"
    },
    {
     "id": "run_tests",
     "expect": "tool",
     "wall_s": 0.99,
     "block_types": [
      "thinking",
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 214,
      "cache_read_input_tokens": 0,
      "output_tokens": 57
     },
     "tool_calls": [
      {
       "name": "Bash",
       "input": {
        "command": "ls -R | grep test"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": false,
     "why": "no tool_use block with the expected tool and argument"
    },
    {
     "id": "search_usage",
     "expect": "tool",
     "wall_s": 0.94,
     "block_types": [
      "thinking",
      "tool_use"
     ],
     "stop_reason": "tool_use",
     "usage": {
      "input_tokens": 216,
      "cache_read_input_tokens": 0,
      "output_tokens": 54
     },
     "tool_calls": [
      {
       "name": "Grep",
       "input": {
        "pattern": "load_config"
       }
      }
     ],
     "text": "",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": true,
     "ok": true,
     "why": "tool_use with the expected tool and argument"
    },
    {
     "id": "three_words",
     "expect": "text",
     "wall_s": 0.25,
     "block_types": [
      "text"
     ],
     "stop_reason": "end_turn",
     "usage": {
      "input_tokens": 211,
      "cache_read_input_tokens": 0,
      "output_tokens": 8
     },
     "tool_calls": [],
     "text": "local model ready",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": false,
     "ok": true,
     "why": "plain text with the expected content"
    },
    {
     "id": "plain_question",
     "expect": "text",
     "wall_s": 1.01,
     "block_types": [
      "thinking",
      "text"
     ],
     "stop_reason": "end_turn",
     "usage": {
      "input_tokens": 212,
      "cache_read_input_tokens": 0,
      "output_tokens": 58
     },
     "tool_calls": [],
     "text": "JSON stands for JavaScript Object Notation.",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": false,
     "ok": true,
     "why": "plain text with the expected content"
    },
    {
     "id": "tool_result_round_trip",
     "expect": "text",
     "wall_s": 1.06,
     "block_types": [
      "thinking",
      "text"
     ],
     "stop_reason": "end_turn",
     "usage": {
      "input_tokens": 279,
      "cache_read_input_tokens": 0,
      "output_tokens": 60
     },
     "tool_calls": [],
     "text": "2",
     "tool_call_written_as_text": false,
     "well_formed_tool_use": false,
     "ok": true,
     "why": "plain text with the expected content"
    }
   ]
  }
 ],
 "tool_choice": {
  "note": "same prompt three ways; tool_choice only shows an effect if the answers differ. Ollama's documentation lists tool_choice as unsupported.",
  "no_tool_choice": {
   "status": 200,
   "stop_reason": "tool_use",
   "block_types": [
    "tool_use"
   ],
   "tool_calls": [
    {
     "name": "Bash",
     "input": {
      "command": "echo Hello"
     }
    }
   ],
   "text": "",
   "error_body": null
  },
  "forced_bash": {
   "status": 200,
   "stop_reason": "tool_use",
   "block_types": [
    "tool_use"
   ],
   "tool_calls": [
    {
     "name": "Bash",
     "input": {
      "command": "echo Hello"
     }
    }
   ],
   "text": "",
   "error_body": null
  },
  "forced_read": {
   "status": 200,
   "stop_reason": "tool_use",
   "block_types": [
    "tool_use"
   ],
   "tool_calls": [
    {
     "name": "Bash",
     "input": {
      "command": "echo Hello"
     }
    }
   ],
   "text": "",
   "error_body": null
  },
  "none": {
   "status": 200,
   "stop_reason": "tool_use",
   "block_types": [
    "tool_use"
   ],
   "tool_calls": [
    {
     "name": "Bash",
     "input": {
      "command": "echo Hello"
     }
    }
   ],
   "text": "",
   "error_body": null
  }
 },
 "context_overflow": {
  "fits": {
   "filler_lines": 30,
   "prompt_chars": 2585,
   "status": 200,
   "usage": {
    "input_tokens": 601,
    "cache_read_input_tokens": 5,
    "output_tokens": 8
   },
   "stop_reason": "end_turn",
   "text": "MAGNOLIA-47",
   "answered_code_word": true,
   "error_body": null,
   "wall_s": 0.38
  },
  "longer_than_context": {
   "filler_lines": 768,
   "prompt_chars": 64507,
   "status": 200,
   "usage": {
    "input_tokens": 3070,
    "cache_read_input_tokens": 4,
    "output_tokens": 40
   },
   "stop_reason": "end_turn",
   "text": "I don't see any deployment code word in the provided text. The text appears to be a repetitive log entry from a nightly job that copies the reports folder and writes an entry to the log.",
   "answered_code_word": false,
   "error_body": null,
   "wall_s": 1.96
  }
 },
 "unload_status": 200,
 "resident_after": [
  {
   "name": "gemma4-32k:latest",
   "size_bytes": 22703966780,
   "size_vram_bytes": 22703966780,
   "context_length": 32768
  }
 ],
 "finished_at": "2026-11-15T19:54:04+00:00"
}
