{
  "generator": {
    "matra": "0.3.0",
    "udpipe_model": {
      "name": "english-ewt-ud-2.5-191206",
      "sha256": "784bd0fa85e3d831fd02a55290d0acfd05c953159dc38cc33d52e1b28add9957"
    }
  },
  "page": "guides/tls-proxy.md",
  "paragraphs": [
    {
      "compression_ratio": null,
      "kind": "paragraph",
      "lexical_density": 0.5882,
      "readability_grade": 7.6988,
      "sentences": 1,
      "text": "Place the parsing model by hand when a proxy that re-signs TLS stops matra from downloading it.",
      "words": 17
    },
    {
      "compression_ratio": 0.5831,
      "kind": "paragraph",
      "lexical_density": 0.569,
      "readability_grade": 14.2338,
      "sentences": 3,
      "text": "matra verifies TLS against root certificates compiled into the binary and never reads the system trust store. That is why it needs no ca-certificates package, on any platform, and it is also why a proxy that re-signs TLS cannot be trusted by installing its CA anywhere on the machine. The first run that needs the model fails with:",
      "words": 60
    },
    {
      "compression_ratio": null,
      "kind": "paragraph",
      "lexical_density": 0.5588,
      "readability_grade": 6.1191,
      "sentences": 3,
      "text": "In Python the same failure is a matra.IoError, an OSError, from matra.Engine(), with the URL in details[\"url\"] and details[\"retryable\"] false, since retrying meets the same proxy. Errors lists every way a download can fail.",
      "words": 36
    },
    {
      "compression_ratio": 0.5333,
      "kind": "paragraph",
      "lexical_density": 0.4878,
      "readability_grade": 11.1871,
      "sentences": 3,
      "text": "The artifact is pinned by name, size and SHA-256, so a hand-placed file is exactly as trustworthy as a fetched one: it goes through the same verification on load. A file that is not the pinned model is never used and never deleted. matra tries to download the pinned model in its place, and where that download cannot get through, which is the reason to place the file by hand, the call fails and the file stays exactly as you left it.",
      "words": 85
    },
    {
      "compression_ratio": null,
      "kind": "list_item",
      "lexical_density": 0.6471,
      "readability_grade": 7.1603,
      "sentences": 2,
      "text": "Find the model directory this machine resolves, and create it. matra config show prints it as model_dir.",
      "words": 17
    },
    {
      "compression_ratio": null,
      "kind": "list_item",
      "lexical_density": 0.4706,
      "readability_grade": 5.6165,
      "sentences": 1,
      "text": "Download the model with a client that trusts the proxy, such as curl with the system store.",
      "words": 17
    },
    {
      "compression_ratio": null,
      "kind": "list_item",
      "lexical_density": 0.5769,
      "readability_grade": 8.0877,
      "sentences": 2,
      "text": "Check its SHA-256 before moving it in, so a truncated or proxy-rewritten download fails here rather than on the next run. The digest must read 784bd0fa85e3d831fd02a55290d0acfd05c953159dc38cc33d52e1b28add9957.",
      "words": 28
    },
    {
      "compression_ratio": null,
      "kind": "list_item",
      "lexical_density": 0.5,
      "readability_grade": null,
      "sentences": 1,
      "text": "Move it into the model directory.",
      "words": 6
    },
    {
      "compression_ratio": null,
      "kind": "paragraph",
      "lexical_density": 0.4,
      "readability_grade": 9.9267,
      "sentences": 1,
      "text": "MATRA_MODEL_DIR points at a directory of your own if the resolved one is not writable.",
      "words": 15
    },
    {
      "compression_ratio": null,
      "kind": "paragraph",
      "lexical_density": 0.5366,
      "readability_grade": 3.7251,
      "sentences": 2,
      "text": "Run any command that parses. The command line writes matra: downloading ... to standard error before a fetch and nothing when the model is already in place, so a run that prints no such line has loaded the file you placed.",
      "words": 40
    },
    {
      "compression_ratio": null,
      "kind": "paragraph",
      "lexical_density": 0.6,
      "readability_grade": 16.4,
      "sentences": 1,
      "text": "The reference embedding model has the same route: three artifacts into a directory loaded with Model2Vec::from_dir, described in semantic clusters.",
      "words": 20
    }
  ]
}
