{
  "schema": "kbatch-world-language-v1",
  "product": "ugrad.kbatch + lang.ugrad",
  "title": "Best model for all world languages",
  "mission": "Climb every language surface — geometry · orthography · writing systems · family tree · research · ranking · eval · MT arena — until one stack understands every living script and tongue.",
  "doctrine": [
    "Always climb all world axes (dictionaries · schools · museums · typing · music · translation)",
    "Path geometry is the spine; meaning + MT fidelity is the voice",
    "Alice R1 slip/understanding is the honest MT smoke — not marketing BLEU",
    "Offline (qwen3:8b) + edge (m2m100) + KBatch geometry feed one world model",
    "Honor living languages · opt-in · PD/museum-open where archival"
  ],
  "updated": "2026-08-10T12:30:00Z",
  "rung": "WL1.1-ugrad-model",
  "canonical": {
    "hub": "https://kbatch.ugrad.ai/world-language",
    "kbatch": "https://kbatch.ugrad.ai/",
    "lang": "https://lang.ugrad.ai/",
    "charts": "https://lang.ugrad.ai/charts/",
    "arena_board": "https://lang.ugrad.ai/data/language/arena-lang-board.json",
    "arena_mirror": "https://kbatch.ugrad.ai/data/world-language/arena-lang-board.json",
    "axes": "https://kbatch.ugrad.ai/data/world-ranking/axes.json",
    "manifest": "https://kbatch.ugrad.ai/data/world-language/manifest.json"
  },
  "surfaces": [
    {
      "id": "dojo",
      "label": "DOJO",
      "url": "https://kbatch.ugrad.ai/dojo/",
      "role": "Pattern matrix · letter atoms · colossus · MCP · multi-lang corpus",
      "feeds": ["geometry", "layouts", "capsules", "jax"]
    },
    {
      "id": "rosetta",
      "label": "Rosetta",
      "url": "https://kbatch.ugrad.ai/labs/rosetta",
      "role": "World writing systems · alphabets · script coverage",
      "feeds": ["scripts", "orthography"]
    },
    {
      "id": "lang-tree",
      "label": "Lang tree",
      "url": "https://kbatch.ugrad.ai/labs/lang-tree",
      "role": "Language family pack navigator · world path cost",
      "feeds": ["families", "packs", "path-cost"]
    },
    {
      "id": "research",
      "label": "Research",
      "url": "https://kbatch.ugrad.ai/research",
      "role": "Paleography · probes · history · dense lines",
      "feeds": ["probes", "history", "paleography"]
    },
    {
      "id": "world-ranking",
      "label": "World ranking",
      "url": "https://kbatch.ugrad.ai/world-ranking",
      "role": "Five(+1) axes · climb every list · collaboration slots",
      "feeds": ["axes", "partners", "product-bars"]
    },
    {
      "id": "eval",
      "label": "Eval",
      "url": "https://kbatch.ugrad.ai/eval/",
      "role": "Day-30 depth KPI · fold · public metrics",
      "feeds": ["depth", "fold", "floors"]
    },
    {
      "id": "for-ai",
      "label": "For AI",
      "url": "https://kbatch.ugrad.ai/for-ai",
      "role": "MCP tools · train pack · agent funnel · world axes ref",
      "feeds": ["mcp", "llm-train", "agents"]
    },
    {
      "id": "lang-charts",
      "label": "Lang Arena charts",
      "url": "https://lang.ugrad.ai/charts/",
      "role": "18-lang MT Arena% · complete understanding · live slip · emit",
      "feeds": ["mt-accuracy", "slip", "semantic"]
    },
    {
      "id": "lang-zen",
      "label": "Zen multi-output",
      "url": "https://lang.ugrad.ai/zen.html",
      "role": "Live fanout translate · offline qwen3 · edge m2m100",
      "feeds": ["mt-live", "offline", "edge"]
    },
    {
      "id": "lang-alice",
      "label": "Alice MT pack",
      "url": "https://lang.ugrad.ai/data/language/alice-mt-analysis/",
      "role": "Literary pairs · semantic bands · truncation ablation",
      "feeds": ["train-pairs", "T0-T3", "proxy-bleu"]
    },
    {
      "id": "ugrad-model",
      "label": "ugrad MODEL",
      "url": "https://kbatch.ugrad.ai/ugrad-model",
      "role": "Compiled capsule · Alice cal · device-scale zen/watch routing · word metrics",
      "feeds": ["routing", "seats", "alice-cal", "word-metrics"]
    }
  ],
  "model_stack": {
    "goal": "Best all-world-language model — geometry + meaning + MT",
    "layers": [
      {
        "layer": "geometry",
        "engine": "KBatch path / steno / 15 layouts / colossus",
        "host": "kbatch.ugrad.ai"
      },
      {
        "layer": "lexicon",
        "engine": "54M+ spellings · Wiktionary glosses · capsules · concept mesh",
        "host": "kbatch.ugrad.ai"
      },
      {
        "layer": "scripts",
        "engine": "Rosetta writing systems · alphabet matrix · 88+ langs",
        "host": "kbatch.ugrad.ai/labs/rosetta"
      },
      {
        "layer": "families",
        "engine": "Lang tree · world-path · Rubik language stair",
        "host": "kbatch.ugrad.ai/labs/lang-tree"
      },
      {
        "layer": "mt-offline",
        "engine": "Ollama qwen3:8b",
        "host": "local · lang.ugrad zen offline"
      },
      {
        "layer": "mt-edge",
        "engine": "Workers AI m2m100-1.2b",
        "host": "lang.ugrad.ai"
      },
      {
        "layer": "mt-arena",
        "engine": "Alice R1 multi-signal Arena% board",
        "host": "lang.ugrad.ai/charts"
      },
      {
        "layer": "ugrad-model",
        "engine": "Compiled capsule · device routing · watch20 / zen-core seats",
        "host": "kbatch.ugrad.ai/ugrad-model"
      },
      {
        "layer": "eval",
        "engine": "KBatch eval floors + lang scorecard + COMET thresholds",
        "host": "kbatch.ugrad.ai/eval · lang.ugrad scorecard"
      }
    ]
  },
  "arena_summary": {
    "source": "Alice R1 smoke · 18 langs",
    "system": {
      "semantic_fidelity": 88,
      "fluency": 86,
      "completeness_live_trunc": 62,
      "overall_usable_trunc": 78
    },
    "top": ["JA", "KO", "ZH", "EN"],
    "formula": "0.50*complete_understanding + 0.30*live_slip + 0.20*emit_completeness",
    "board": "/data/world-language/arena-lang-board.json"
  },
  "read_order_for_ai": [
    "https://kbatch.ugrad.ai/data/world-language/manifest.json",
    "https://kbatch.ugrad.ai/ugrad-model",
    "https://kbatch.ugrad.ai/data/world-ranking/axes.json",
    "https://lang.ugrad.ai/data/language/arena-lang-board.json",
    "https://kbatch.ugrad.ai/for-ai",
    "https://lang.ugrad.ai/charts/",
    "https://kbatch.ugrad.ai/dojo/",
    "https://kbatch.ugrad.ai/data/ugrad-model/manifest.json",
    "https://lang.ugrad.ai/data/language/alice-mt-analysis/ANALYSIS.md"
  ],
  "next": [
    "Wire zen.html + fcs #/watch to GET /data/ugrad-model/routing.json",
    "Grow word-metrics from analyzed slivers for every dict spelling",
    "Expand Arena beyond Alice 18 → full zen 23+ targets with COMET-QE",
    "Train positives only from pairs_complete · density rebalance CJK",
    "Wire kbatch_concept_solve ↔ lang fanout for meaning-checked MT",
    "Pull NLLB/MADLAD offline weights into 07-lang-ugrad",
    "Optional Kimi teacher path for hard pairs (cloud only)"
  ]
}