{
  "description": "Messgrößen des optimierten Entwicklungsbuilds vom 27. September 2026, direkt mit dem gesicherten Ausgangsbuild verglichen. Keine pauschale Aussage über das veröffentlichte Orbit 0.8.5.",
  "date": "2026-09-27",
  "build_status": "development",
  "cli_package_version": "0.8.5",
  "base_commit": "54f3284c5dc4d815d46cb35618bd2b29fb280fab",
  "binaries": {
    "before": "aa620573fadb748378ebb6e3246209b4791feb8e24d311666ebc7f28d3b72aca",
    "after": "74d09184747e2a42c155aa98c3e3ec088bc2f057e3e916f8eafc55976660b8d4"
  },
  "hardware": {
    "cpu": "Apple M4",
    "memory_bytes": 17179869184,
    "macOS": "26.2"
  },
  "markdown_files": 20831,
  "command": "context --agent",
  "budget": {
    "max_tokens": 4000,
    "max_bytes": 65536
  },
  "limit": 5,
  "conditions": {
    "primary_questions": 24,
    "control_questions": 32,
    "repetitions_per_question_per_build_per_transport": 5,
    "transports": [
      "cli",
      "ipc"
    ],
    "randomly_interleaved": true,
    "warmup_requests_per_service": 100,
    "timing_includes": [
      "CLI process start for cli transport",
      "IPC",
      "request processing",
      "JSON output"
    ],
    "timing_excludes": [
      "index build",
      "LLM answer time"
    ],
    "median_definition": "statistics.median; average of middle two observations for an even sample size",
    "graph_cold_start": "Warmup requests without hits do not build the graph. First graph-bearing control request: before IPC 187.738542 ms, after CLI 171.122333 ms. The randomized transport order allocates this one-time cost differently; CLI maxima in this panel are not an isolated paired cold-start comparison."
  },
  "cli": {
    "primary": {
      "before": {
        "ms": {
          "n": 120,
          "p50": 14.1051045,
          "p95": 26.382583,
          "p99": 33.98875,
          "max": 34.422083
        },
        "positive": 22,
        "hit": 19,
        "all_gold": 11
      },
      "after": {
        "ms": {
          "n": 120,
          "p50": 12.897583000000001,
          "p95": 21.345167,
          "p99": 27.665792,
          "max": 30.025292
        },
        "positive": 22,
        "hit": 19,
        "all_gold": 11
      }
    },
    "control": {
      "before": {
        "ms": {
          "n": 160,
          "p50": 15.429853999999999,
          "p95": 25.155583,
          "p99": 28.314708,
          "max": 28.842625
        },
        "positive": 31,
        "hit": 25,
        "all_gold": 17
      },
      "after": {
        "ms": {
          "n": 160,
          "p50": 14.162082999999999,
          "p95": 21.010917,
          "p99": 25.31525,
          "max": 171.122333
        },
        "positive": 31,
        "hit": 25,
        "all_gold": 17
      }
    }
  },
  "maintenance": {
    "duration_s": 75.1028682920005,
    "pairs": 559,
    "same_evidence": true,
    "stats": {
      "before": {
        "n": 559,
        "p50": 23.806875,
        "p95": 38.490958,
        "p99": 45.747208,
        "max": 748.785709
      },
      "after": {
        "n": 559,
        "p50": 21.277625,
        "p95": 32.632042,
        "p99": 37.7815,
        "max": 39.961084
      }
    }
  },
  "maintenance_conditions": "Separate 75-second observation; graph warmed explicitly; 80 ms pause between randomly ordered CLI pairs; crosses periodic 60-second validation. Observed maximum is not a latency guarantee.",
  "quality": {
    "unique_questions": 1000,
    "same_evidence": 1000,
    "same_scores": 1000,
    "compared_evidence": [
      "path",
      "text",
      "order",
      "lines",
      "byte_range"
    ],
    "existing_dataset": true,
    "includes_the_56_timing_questions": true,
    "independent_holdout": false,
    "evaluates_llm_answers": false
  },
  "validation": {
    "rust_tests_passed": 184,
    "rust_tests_failed": 0,
    "existing_manual_performance_test_ignored": 1,
    "cli_tests_passed": 34,
    "cli_tests_failed": 0,
    "final_cache_repair_change": "Two affected background-validation tests rerun after the final workspace-cache repair addition; final binary used for all final benchmarks and the 1000-case evaluation."
  },
  "limits": [
    "Local benchmark on one machine and an existing corpus.",
    "Development build; installed or later versions may differ.",
    "Identical retrieved evidence is regression evidence, not proof of perfect answers.",
    "First graph construction, error repair and large updates may take longer.",
    "Historical marketing value 13.88 ms comes from a different run and is not the paired baseline here."
  ],
  "sources": [
    {
      "path": "docs/qa/context-latency-implementation-20260927/REPORT.md",
      "sha256": "b1e1b29bacd63d4caa02eb37620373deedfad08fe305efdd82ba468efade49e3"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/manifest.json",
      "sha256": "25dad9fe43893964a227f731ac0a731e9e69d7dc76396e56fbab76af1cd62cbc"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/measurement.json",
      "sha256": "48c733bb6fd2747915bef97356720a7d1c070412c87234a9362d8485c3b596b6"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/quality.json",
      "sha256": "c9da2731b0e6ebf556aadd2d1156063d8ffcd0c7b2b3bfe889bc469f07827abf"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/maintenance.json",
      "sha256": "9505f2f448c0af50c2177b3ceec6832197ee3ce493ab0b6969296a72f5efef70"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/tests.log",
      "sha256": "d377fe4b619ab2b62bf2d0ccc07c53898eb4130d781fc36ffcc52b05948c1839"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/cli-tests.log",
      "sha256": "fc1b1d35ed44148d61dea23a92b17d3a8a1a24f6399c3770bc6ccb3da5d3a309"
    },
    {
      "path": "docs/qa/context-latency-implementation-20260927/read-scan-final-tests.log",
      "sha256": "9cba7f7a6df54be2e2c6c30e3e15f9b98316870c17aedd42ce2c2af4b750b410"
    }
  ]
}