Document verified PROCEDURE assets, immutable loading, lineage and private organizer access

DATA_PROVENANCE.md

@@ -0,0 +1,50 @@
+# Data provenance and disclosure boundary
+
+This document covers the retained PROCEDURE candidate and its selected Dense48/Point320 lineage. It does not substitute the diet of another pfull, FRAME, SEGMENT, private-data continuation or later experimental checkpoint. [LINEAGE.json](LINEAGE.json) records exact evidence hashes and distinguishes recovered metadata, intended recipe, observed consumption and selected checkpoint bytes.
+
+## Selected adapter training records
+
+| Record | Dense48 | Point320 specialist |
+|---|---|---|
+| Selected checkpoint | 12,910; final-save and checkpoint tensors match | Independently accepted terminal 12,910 |
+| Initialization | Checkpoint 11,709, tensor SHA-256 `684922901db549df459e586e2d62f1535b3f6820cf6b62e7cf482a409e9ac60e` | Same exact initializer; fresh optimizer |
+| Recorded recipe | 51,635 rows, eight ranks, seed 42, learning rate 0.0002, requested 48 frames, maximum side 576 | Same recorded geometry and learning rate; exact point-target-restored diet below |
+| Proof limit | Recovery establishes selected tensor/config bytes and saved training metadata, not the exact historical corpus revision or complete per-rank input/update history | Actual terminal, consumed-input and reload evidence exists; it does not eliminate inherited exposure or annotation/decode limitations |
+
+The selected Point diet is SHA-256 `3f14484182a01b7e5e8eafad437665cc7fb7975cbeb5511296e05147412ad081`: 51,635 rows over 128 source-video identities. Its preparation restored 4,164 target answers from checked organizer sources (2,089 native views and 2,075 middle views), preserving all current windows, question strings and other columns. It did not restore original windows or guarantee that an event appears in the sampled frames.
+
+The source-lineage audit of the unchanged parent population binds 46,742 rows to organizer HeiCo/LapChole source material. A further 363 transfer rows bind to the paired parent rather than organizer faces. The 4,530 jury rows retain an unresolved original-annotation bridge. This is not a claim that every training row has a fully reconstructed original annotation source.
+
+The Point runtime recorded 103,280 consumed examples across repeated training draws, not 103,280 unique rows. Of these, 78,788 realized 48 frames and 24,492 had a requested/realized mismatch; diagnostics also record 36 substitutions, ten jittered successes and 174,825 rejected blank frames. Completion is therefore not described as perfect data/render fidelity. Later corpus filtering cannot erase possible exposure inherited through checkpoint 11,709.
+
+## Named sources and restrictions
+
+| Source | Relevant provenance | Distribution scope |
+|---|---|---|
+| [HeiCo-FOCUS-VQA](https://huggingface.co/datasets/orena-dkfz/heico-focus-vqa) | Organizer colorectal VQA training material and derived views/statistics. The audited parent source covers organizer track-specific training faces; no claim is made that every ancestor diet used the same revision. | The source records CC BY-NC-SA 4.0 plus a challenge-publication restriction. Historical campaign records preserve dataset-specific challenge-use rulings and annotation-delivery conditions; they are not a new redistribution grant or proof that an obligation has been completed. |
+| [LapChole-FOCUS-VQA](https://huggingface.co/datasets/orena-dkfz/lapchole-focus-vqa) | Organizer laparoscopic cholecystectomy VQA training material, derived point views and the selected organizer-derived v39 prior. | Access and data handling are governed by the organizer Data Usage Agreement, not a generic open-data license. The published conditions restrict data sharing, public release and reidentification. Consult the original agreement for model/data distinctions. |
+| Jury and transfer-derived rows | 4,530 jury rows retain an original-annotation gap; 363 transfer rows are traced to their paired parent. | Raw rows and source media are not distributed here. This unresolved history is not filled with an invented public-dataset attribution or a claim of complete clearance. |
+| Private V2 surgery and its derivatives | A separately identified six-predicate diagnostic used for evaluation only; no qualified matched model-accuracy comparison resulted. | Evaluation-only; excluded from adaptation and training-rule selection. No V2 media or reference labels are included. |
+
+The [PROCEDURE rules](https://procedure.orena-focus-challenge.org/rules/) and captured method-description form require actual data-source disclosure, including descriptions of private sources. This repository does not claim that the method form, architecture figure, restricted annotation handoff or publication obligations have been submitted or acknowledged. The current source pages are attribution/access references; they do not replace pinned historical training evidence.
+
+## Included derived runtime priors
+
+The package includes six small JSON assets under `image_root/app/artifacts/` in addition to learned weights:
+
+| File | Role and provenance limit |
+|---|---|
+| `train_modes.json` | Training modes by track, capability and answer format; no embedded dataset revision or complete derivation receipt. |
+| `x11_stem_table.json` | PROCEDURE question-stem answer modes; no embedded dataset revision or complete derivation receipt. |
+| `stem_table.json` | Track/capability/format/stem modes; source requires at least five training questions and 40% support, but the asset has no complete derivation receipt. |
+| `fo_quadrant_priors_colorectal.json` | HeiCo colorectal spatial priors, attributed by the exact loader source; no embedded dataset revision. |
+| `cholec_priors_v39_organizer.json` | Selected organizer LapChole prior derived using the shipped detector; embedded record describes 57 videos, 3,477 frames and 14,258 detections. This is the v39 artifact, not the historical CholecTrack20-derived v38 fallback. |
+| `duration_priors.json` | HeiCo procedure-training duration constants. The constructor reads the asset, while the retained L16 applicability flag is disabled. |
+
+ASSET_MANIFEST.json binds their bytes. These assets may contain source-derived answer modes and question stems; they are not described as purely learned parameters. Their embedded historical rulings are preserved as provenance rather than converted into a new legal conclusion.
+
+## What is withheld and what the scores mean
+
+Private surgical media and evaluation banks are not distributed here. The original inference source may retain unit-test fixtures, and the above runtime priors are explicitly included. No raw Parquet/JSONL training dataset, per-question evaluation-output bank or evaluation gold is selected for this handoff.
+
+The 1,087-question local development board was repeatedly used for model/inference selection. It contains zero challenge OOD rows and only one clinical-flagged row. Its local scores cannot establish a clinical accuracy estimate, untouched generalization, an official platform result or a finals rank. The retained failed rows and the distinction between historical and fresh native protocols are reported in README.md and EVALUATION_SUMMARY.json.

EVALUATION_SUMMARY.json

@@ -0,0 +1,107 @@
+{
+  "evaluation_gold_or_rows_included": false,
+  "fresh_native_protocol": {
+    "all_primary_outcomes_retained": 2174,
+    "external_process_pool": "120 + 30 * B seconds",
+    "failure_causes": {
+      "pfull_best_platform": {
+        "original_process_pool_timeout_rows": 124
+      },
+      "selected615": {
+        "stream_cap_rows": 28,
+        "validation_demotion_group_rows": 39
+      }
+    },
+    "profiles": {
+      "pfull_best_platform": {
+        "five_bucket_mean_exact": {
+          "denominator": 3272160,
+          "float": 0.46532229475331277,
+          "numerator": 1522609
+        },
+        "n": 1087,
+        "n_correct": 444,
+        "outcomes": {
+          "answered": 963,
+          "wrapper_failed": 124
+        },
+        "per_bucket_correct": {
+          "aggregation": 39,
+          "complex_reasoning": 13,
+          "event_understanding": 10,
+          "object_recognition": 267,
+          "temporal_grounding": 115
+        }
+      },
+      "selected615": {
+        "five_bucket_mean_exact": {
+          "denominator": 654432,
+          "float": 0.5593201432692778,
+          "numerator": 366037
+        },
+        "n": 1087,
+        "n_correct": 579,
+        "outcomes": {
+          "answered": 1020,
+          "wrapper_failed": 67
+        },
+        "per_bucket_correct": {
+          "aggregation": 53,
+          "complex_reasoning": 16,
+          "event_understanding": 10,
+          "object_recognition": 399,
+          "temporal_grounding": 101
+        }
+      }
+    },
+    "resources": {
+      "cpus": 16,
+      "gpu": "H100",
+      "host_memory_mib": 196608
+    }
+  },
+  "historical_local_protocol": {
+    "five_bucket_mean_exact": {
+      "denominator": 204510,
+      "float": 0.6191042002836047,
+      "numerator": 126613
+    },
+    "n": 1087,
+    "n_correct": 615,
+    "outcomes": {
+      "answered": 1087
+    },
+    "per_bucket_correct": {
+      "aggregation": 56,
+      "complex_reasoning": 18,
+      "event_understanding": 12,
+      "object_recognition": 416,
+      "temporal_grounding": 113
+    }
+  },
+  "independent_native_scoring_review": {
+    "bytes": 13528,
+    "sha256": "fb18091701d6dcbfc55abd1d646cd68ff29c9fad3b32a8273d2b49722e216515"
+  },
+  "joint_qualified_sensitivity": {
+    "comparator_correct": 421,
+    "n": 910,
+    "replaces_primary_population": false,
+    "selected_correct": 511
+  },
+  "limits": [
+    "Local selected development population; no untouched or challenge OOD generalization claim.",
+    "One clinical-flagged row cannot establish clinical quality.",
+    "Correlated question/video observations and repeated model selection limit rowwise comparisons.",
+    "Historical615 and fresh native579 are different protocols."
+  ],
+  "official_challenge_result": false,
+  "population": {
+    "challenge_ood_rows": 0,
+    "clinical_flagged_rows": 1,
+    "questions": 1087,
+    "used_during_development_and_selection": true,
+    "videos": 27
+  },
+  "schema": "procedure-final-local-evaluation-summary/1"
+}

EXACT_DEPENDENCIES.json

@@ -0,0 +1 @@
+{"packages": {"Pillow": "12.3.0", "accelerate": "1.14.0", "av": "16.1.0", "bitsandbytes": "0.50.1", "huggingface-hub": "1.29.0", "peft": "0.20.0", "safetensors": "0.8.0", "tokenizers": "0.22.2", "torch": "2.13.0", "transformers": "5.14.0"}, "python": "3.11.11 | packaged by conda-forge | (main, Dec  5 2024, 14:17:24) [GCC 13.3.0]"}

LICENSES.json

@@ -0,0 +1,84 @@
+{
+  "contact": "Reza Nehzati, Ph.D.",
+  "inactive_exclusions": {
+    "Qwen/Qwen3-VL-8B-Instruct": "Not in selected upload; its Apache-2.0 card was checked only to avoid confusing the inactive cached model with the selected foundation.",
+    "SAM_YOLO_ByteTrack": "No additional active selected weight artifact identified by exact source/asset map; do not import unrelated historical license inventories."
+  },
+  "overall_new_license_assigned_by_this_review": false,
+  "predecessor_license_provenance": {
+    "bytes": 4101,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/LICENSES.json",
+    "sha256": "4e0bd85f6b2344dc545a430395c512f781254e0ee10ba60501db4646fa3bf29d"
+  },
+  "private_only": true,
+  "publisher": "HeyDonto Labs",
+  "repository_mapping_addendum": {
+    "bytes": 5043,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/CACHE_MAPPING_B.json",
+    "sha256": "8dd7bdeb47c5fd2b0b001a917da42900d8f1e73f026a7f095980c0103dd7d394"
+  },
+  "retained_notices": [
+    "Retain the actual Qwen LICENSE and upstream copyright notice, model cards and the project modification description.",
+    "Model upstream licensing does not by itself establish redistribution rights for training videos, annotations, adapted weights or runtime priors.",
+    "Private access is the owner requested distribution setting, not a replacement for applicable upstream terms."
+  ],
+  "schema": "selected615-hf-upstream-license-provenance/1",
+  "training_data_scope": {
+    "dataset_redistribution_clearance_established": false,
+    "historical_owner_policy": {
+      "bytes": 3116,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/history/trees/surgfield-save-focus-regimed/training/guards/nc_license_denylist.json",
+      "sha256": "0111fb7a2ff76f1f3552020bd447402974cee82d7cf3bfdc1f9b3d5780beaf27"
+    },
+    "historical_policy_reading": "Preserved policy records contain evaluation-only restrictions and later dataset-specific owner rulings. They are not a new permission to redistribute source data or evidence that every listed dataset is present in these selected adapters. Exact included v39 organizer-derived priors replace the inactive older CholecTrack20-derived v38 artifact.",
+    "lineage": {
+      "bytes": 22451,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/LINEAGE_B.json",
+      "sha256": "4861d771843d543f75cd7a9d0440670f784f615522528b1504f17a6a3823367c"
+    },
+    "raw_data_in_upload": false,
+    "runtime_derived_statistics_in_upload": true
+  },
+  "upstream": [
+    {
+      "applies_to": "Upstream base/model metadata; selected saved int8 transformation should be disclosed as a modification.",
+      "baked_license_identical": true,
+      "declared_license": "Apache-2.0",
+      "license_text": {
+        "bytes": 11544,
+        "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/upstream/qwen38_LICENSE.raw",
+        "sha256": "bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a"
+      },
+      "model_card": {
+        "bytes": 65012,
+        "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/upstream/qwen38_README.raw",
+        "sha256": "57e4bdb258ee1a7d2635c5174ebd4e56abe392505cdb5f8bbb356b0dc4293641"
+      },
+      "repository": "Qwen/Qwen3.8-27B",
+      "revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
+      "url": "https://huggingface.co/Qwen/Qwen3.8-27B/resolve/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/LICENSE"
+    },
+    {
+      "applies_to": "Included OWL-V2 model/cache assets; model card declares Apache-2.0.",
+      "declared_license": "Apache-2.0",
+      "model_card": {
+        "bytes": 4840,
+        "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/upstream/owlv2_README.raw",
+        "sha256": "a2e10c3916166f08eaf2ab43ca1eb63c6116df228dea655c56e4c8e1607ecfe9"
+      },
+      "repository": "google/owlv2-large-patch14-ensemble",
+      "revision": "95e26936e865f87db1742128404b3c035d47d89d",
+      "url": "https://huggingface.co/google/owlv2-large-patch14-ensemble/resolve/95e26936e865f87db1742128404b3c035d47d89d/README.md"
+    }
+  ],
+  "verification": {
+    "auth_used": false,
+    "license_assessment_type": "Primary upstream license and local provenance inventory; not a new legal opinion.",
+    "retrieval": {
+      "bytes": 4473,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/upstream/RETRIEVAL.json",
+      "sha256": "27f3b5d0879dfb807b3028870d028ff214cefea008fcdc8d2a15372e4e23a919"
+    },
+    "tls_verified": true
+  }
+}

LICENSE_PROVENANCE.md

@@ -0,0 +1,17 @@
+# License provenance
+
+This document reports component terms and their evidence. It does not grant a new blanket license for the aggregate private package or assert that model, source, dataset and annotation terms are interchangeable.
+
+| Component | Recorded terms or status | Evidence and scope |
+|---|---|---|
+| Qwen/Qwen3.8-27B upstream metadata and model | Apache License 2.0 | Revision `1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0`; original [LICENSE](image_root/app/artifacts/q38_base/LICENSE) and README match independently fetched pinned upstream bytes. |
+| Prequantized Qwen base | Quantized derivative of the recorded Qwen base | Exact selected tensor/config hashes and quantization configuration are in ASSET_MANIFEST.json and LINEAGE.json. Retain the Qwen notices; no new aggregate-package license is declared here. |
+| Dense48 and Point320 LoRA adapters | Team-selected adaptations with upstream and data provenance | Exact terminal checkpoint ancestry is in LINEAGE.json. The existence of an upstream Apache license does not establish unrestricted rights to all data-derived material. |
+| google/owlv2-large-patch14-ensemble auxiliary cache | Apache License 2.0 as recorded by the model card | Pinned revision `95e26936e865f87db1742128404b3c035d47d89d`; [Google's original card](https://huggingface.co/google/owlv2-large-patch14-ensemble/blob/95e26936e865f87db1742128404b3c035d47d89d/README.md), matching cached SHA-256 `a2e10c3916166f08eaf2ab43ca1eb63c6116df228dea655c56e4c8e1607ecfe9`. |
+| Team inference source | Preserved exact source; no additional blanket grant recorded by this document | All 84 selected file hashes are in SOURCE_MANIFEST.json. Existing notices remain intact. |
+| Third-party Python/CUDA runtime | Per-package terms | Installed versions are observed in EXACT_DEPENDENCIES.json; the complete runtime is retained in the original OCI image rather than repackaged here. |
+| Organizer/private data, annotations and six derived runtime priors | Source-specific access and usage terms | DATA_PROVENANCE.md and LINEAGE.json preserve the recorded sources, restrictions and remaining evidence limits. Private media and evaluation-bank payloads are excluded. |
+
+The repository is private, and configured organizer read access is not a license grant or permission to make restricted data public. No `license: apache-2.0` label is applied to the whole model card merely because the base and auxiliary model report Apache terms. Original attribution and applicable license text must accompany components where redistribution is permitted.
+
+The exact inventory's LICENSES.json is the machine-readable provenance companion. Dataset terms and any recorded challenge-specific permissions should be read at their original source; this document does not replace those instruments or declare a separate obligation completed.

LINEAGE.json

@@ -0,0 +1,406 @@
+{
+  "asset_manifest": {
+    "bytes": 71837,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/ASSET_MANIFEST_B.json",
+    "sha256": "bf9706c9c565dae44f5c3285c565490c89c84c44194d48130d39cfcbd238824d"
+  },
+  "auxiliary_detector": {
+    "both_weight_formats_retained": true,
+    "cache_root": "/app/.cache/huggingface/hub/models--google--owlv2-large-patch14-ensemble",
+    "initialization": {
+      "bytes": 21477,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/container.py",
+      "sha256": "df39335c4843e814b354b2e587bf48a77aebbe88325be99a6f7e2f9e136474c7"
+    },
+    "load": "Owlv2Processor and Owlv2ForObjectDetection.from_pretrained(...,local_files_only=True), using exact cache refs/main.",
+    "only_one_loaded_format_proven": false,
+    "revision": "95e26936e865f87db1742128404b3c035d47d89d",
+    "separate_clip_download_required": false,
+    "source": {
+      "bytes": 8568,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/detection.py",
+      "sha256": "9c8bfec8df2effb3b642c2bbd092dcecf837b851e81c45008c531e5dcc0a1a58"
+    },
+    "upstream_repository": "google/owlv2-large-patch14-ensemble"
+  },
+  "base": {
+    "actual_architectures": [
+      "Qwen3_5ForConditionalGeneration"
+    ],
+    "actual_model_type": "qwen3_5",
+    "evidence": "Exact revision config, README, and LICENSE are byte-identical to baked q38_base metadata; upstream README explains the Qwen3.5 architectural foundation. Repository and architecture names are distinct verified fields.",
+    "limitations": [
+      "PREQUANT_MANIFEST does not contain base_model_rev; matching metadata and training snapshot do not independently rederive every prequantized tensor from original upstream weights.",
+      "No bf16 base weight shards are in the selected q38_base metadata directory.",
+      "Extracted assets/source alone do not reproduce all OCI dependencies, filesystem ownership, and runtime environment."
+    ],
+    "prequant_manifest": {
+      "bytes": 1335,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/metadata/artifacts/q38_base_int8_prequant/PREQUANT_MANIFEST.json",
+      "sha256": "7207316b0b252d832a685bc3345f589ed809013680279e8c2f601a660a924271"
+    },
+    "prequant_source_image": "us-central1-docker.pkg.dev/heydonto-425716/surgfield/surgfield-proc-finals-q38@sha256:2f27588bdde87ab9ed168f82daaef41007fbf9306c78da5eb2f10c1c7616621f",
+    "processor_path": "/app/artifacts/q38_base",
+    "quantization": "Saved bitsandbytes int8; local-only from_pretrained consumes saved quantization config, not a freshly created quantization recipe.",
+    "quantized_config": {
+      "bytes": 4169,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/metadata/artifacts/q38_base_int8_prequant/config.json",
+      "sha256": "e3943d663db8686a81a987ac31f6edcaf940cdf622c7df57ba0c1795f76df55c"
+    },
+    "revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
+    "upstream_metadata_retrieval": {
+      "bytes": 4473,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/upstream/RETRIEVAL.json",
+      "sha256": "27f3b5d0879dfb807b3028870d028ff214cefea008fcdc8d2a15372e4e23a919"
+    },
+    "upstream_repository": "Qwen/Qwen3.8-27B",
+    "weight_path": "/app/artifacts/q38_base_int8_prequant/model.safetensors"
+  },
+  "contact": "Reza Nehzati, Ph.D.",
+  "dense48": {
+    "configuration": {
+      "bytes": 1135,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/metadata/artifacts/proc_q38_dense48_ck12910/adapter_config.json",
+      "sha256": "52e9ce4678fc4bcf4e717245952665c921c13a28247c67b679eeafdb25c4f6f8"
+    },
+    "download_receipt": {
+      "bytes": 8411,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/evidence/arm_runtime_2026-09-07/recovery/DOWNLOAD_RECEIPT.json",
+      "sha256": "9288bc2db9e16e6ea00dc0a314e37d681e54a4750632e4b39022f1ea7b56d2c3"
+    },
+    "image_directory": "/app/artifacts/proc_q38_dense48_ck12910",
+    "independent_recovery": {
+      "bytes": 11671,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/evidence/arm_runtime_2026-09-07/recovery/INDEPENDENT_REVIEW.json",
+      "sha256": "0d8ce2d5f459a6e5d14663731db3176e33faa3c8cdcaddd169dec9ea7e89ea15"
+    },
+    "initializer": "/work/proc/main/checkpoint-11709",
+    "initializer_model_sha256": "684922901db549df459e586e2d62f1535b3f6820cf6b62e7cf482a409e9ac60e",
+    "limits": [
+      "Historical recovered train metadata and saved/final byte equality are available; do not upgrade that recovery to complete historical per-rank optimizer/input proof.",
+      "Exact historical Dense training-corpus revision is not established by this metadata record alone; do not substitute the later quality-original corpus.",
+      "Writer decode_fallback_stats=0 does not prove no fallback on every worker."
+    ],
+    "lora_alpha": 64,
+    "lora_rank": 32,
+    "lr": 0.0002,
+    "max_side": 576,
+    "model_sha256": "500126c5a3a7e813483929225bf2321c82f8a5b59c32ed8aac268db34ab7c4fe",
+    "n_train": 51635,
+    "optimizer_resume_claimed": false,
+    "requested_frames": 48,
+    "role": "Default adapter attached to selected prequantized base.",
+    "saved_epoch": 2,
+    "seed": 42,
+    "terminal_step": 12910,
+    "training_metadata": {
+      "bytes": 936,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/evidence/arm_runtime_2026-09-07/recovery/train_meta_dense48.json",
+      "sha256": "b27e4bf8f3620f053f7ef783c0903305b63003f1e90830e0d62ce89a0c710d1c"
+    },
+    "world_size": 8
+  },
+  "disclosures": [
+    "The upload includes learned model weights AND training/annotation-derived inference priors/stem statistics; it is not only weights.",
+    "No raw training videos, raw Parquet/JSONL question datasets, per-question evaluation outputs or evaluation gold are selected for upload. Derived runtime priors may contain source-derived answer modes and question stems.",
+    "Historical cached Qwen3-VL-8B, cholec_adapter, checkpoint13246 and optimizer/RNG states are excluded; their names in stale baked environment metadata do not make them selected active weights.",
+    "SURGFIELD_NATIVE_VIDEO=0 disables the historical second temporal slot; do not invent a required temporal_adapter directory.",
+    "This provenance review is not a training-data redistribution license or blanket legal clearance."
+  ],
+  "exact_dependencies": {
+    "bytes": 333,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/EXACT_DEPENDENCIES.json",
+    "sha256": "e9345708c455fd7b275781736f6aac2f23621e0251bab8484a8840dacf2bf5ca"
+  },
+  "hf_asset_revision": "6a7fc05f196b98d751e3a14775d60f1161166d2a",
+  "image_config_digest": "sha256:8d3bbd92c7ef8f73ff058382bef1b8523ec55961b831422c1fb04226b31b35d5",
+  "image_identity": {
+    "bytes": 5975,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/IMAGE_IDENTITY.json",
+    "sha256": "913403bda9aea20157c859fa92566e20185a45a18a82ec83ebfb67f4e36dabf7"
+  },
+  "metadata_derivation": {
+    "asset_commit_creation_claimed": true,
+    "change": "Add actual immutable HF asset commit and explicit selected image config; original lineage/evidence fields preserved.",
+    "cpu_qualification": {
+      "bytes": 183011,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/procedure/review/hf_final_20260911/readback_cpu_actual_b/RESULT.json",
+      "sha256": "c8440ff025d0ea7135066db41d681b2acba5c8c72e849ffd9bf6a4a303b89f45"
+    },
+    "cpu_qualification_scope": "Actual readback files/aliases, local configurations/processors, and all640 adapter headers; no whole-model/GPU/quality acceptance.",
+    "independent_remote_readback_claimed": true,
+    "remote_readback": {
+      "bytes": 24367,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/procedure/review/hf_final_20260911/cloud_upload_a/HF_FULL_READBACK.json",
+      "sha256": "14810b6e1436b868c130d8d0e150cb3afceb35685e8b5cbdf0b63e65f04730fb"
+    },
+    "remote_readback_scope": "Fresh HF download and full SHA-256 for all 136 uploaded files; no component or whole-model load claim.",
+    "source_lineage": {
+      "bytes": 22451,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/LINEAGE_B.json",
+      "sha256": "4861d771843d543f75cd7a9d0440670f784f615522528b1504f17a6a3823367c"
+    }
+  },
+  "point_specialist": {
+    "acceptance_summary": {
+      "bytes": 1071,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/review/quality_terminal_custody_20260909/point_acceptance_a/SUMMARY.json",
+      "sha256": "7e3b1eaaa48d358793f47db2f7a15629acc83e9f60f9ba653597da5cf146f2ac"
+    },
+    "actual_consumed_rows": 103280,
+    "actual_elements": 61865984,
+    "actual_tensors": 320,
+    "actual_updates_per_rank": 12910,
+    "configuration": {
+      "bytes": 1135,
+      "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/metadata/artifacts/proc_q38_point_ck12910/adapter_config.json",
+      "sha256": "52e9ce4678fc4bcf4e717245952665c921c13a28247c67b679eeafdb25c4f6f8"
+    },
+    "data": {
+      "comparison_contract": {
+        "bytes": 2167,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/runtime/q38/quality_data_20260909/COMPARISON_CONTRACT.json",
+        "sha256": "4763a0c5c07723eeca39a949120ccea3e44fdae3b48eaf6bd56129ea299bdfd9"
+      },
+      "current_windows_preserved": true,
+      "derivation_description": {
+        "bytes": 2203,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/runtime/q38/quality_data_20260909/README.md",
+        "sha256": "eff7e81e7c0c4f81fac97ceccf85c5fef024de93b17cf337f5981b12fe6c32bf"
+      },
+      "diet_sha256": "3f14484182a01b7e5e8eafad437665cc7fb7975cbeb5511296e05147412ad081",
+      "original_windows_restored": false,
+      "raw_dataset_uploaded": false,
+      "restored_A_mid": 2075,
+      "restored_A_native": 2089,
+      "rows": 51635,
+      "source_proposal": {
+        "bytes": 2680,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/runtime/q38/quality_data_20260909/INPUTS.PROPOSED.json",
+        "sha256": "c2e6ddc44a527d123486908d6c9557af800ee5e3a254a828145cadacfe4eb3d3"
+      },
+      "source_videos": 128,
+      "target_answers_restored": 4164,
+      "unresolved_jury_annotation_bridge_rows": 4530
+    },
+    "image_directory": "/app/artifacts/proc_q38_point_ck12910",
+    "independent_acceptance": {
+      "bytes": 141515,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/review/quality_terminal_custody_20260909/point_acceptance_a/REVIEW.json",
+      "sha256": "6c7308b48e28fd01aa5d7a395ba6b99701d0a0af27000b6f6c0f45b51e5e52d0"
+    },
+    "initializer": "/work/proc/main/checkpoint-11709",
+    "initializer_model_sha256": "684922901db549df459e586e2d62f1535b3f6820cf6b62e7cf482a409e9ac60e",
+    "limits": [
+      "Preserve the 4530-row original-annotation bridge gap; target restoration did not restore original windows or prove event visibility.",
+      "Later corpus decontamination does not erase possible exposure in inherited checkpoint11709.",
+      "Training closure/fresh reload does not establish untouched out-of-distribution or clinical efficacy."
+    ],
+    "lora_alpha": 64,
+    "lora_rank": 32,
+    "lr": 0.0002,
+    "max_side": 576,
+    "model_sha256": "24762d93abdd35704e0087feefb5a9e4d0794d6b7cea39d0df195f93829481a7",
+    "n_train": 51635,
+    "optimizer_resume_claimed": false,
+    "point320_means": "320 adapter tensors; not 320 training steps.",
+    "producer_evidence": {
+      "bytes": 74338,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/review/quality_terminal_custody_20260909/point_actual_a/capture/raw/evidence/ARM_Q38_EVIDENCE.json",
+      "sha256": "8f7265921316784b760f8638babe318b417bbf2ef9a0f9cfd65c54fa7ede41db"
+    },
+    "requested_frames": 48,
+    "role": "Second point_object adapter on same base; source runtime controller selects object capabilities and restores default adapter.",
+    "seed": 42,
+    "tensor_fingerprint": "813714323f3ad235669e2872f693c5a0f2b57faed544f824a3641cdd3592460c",
+    "terminal_step": 12910,
+    "training_decode_diagnostics": {
+      "blank_frames_rejected": 174825,
+      "consumed_at48frames": 78788,
+      "jittered_successes": 10,
+      "not_all_examples_realized48frames": true,
+      "requested_realized_mismatch": 24492,
+      "source": {
+        "bytes": 2355,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/review/quality_terminal_custody_20260909/point_actual_a/TRAINING_DIAGNOSTICS.json",
+        "sha256": "fbebbe51a5629aad16053fe587f59303dadbb16c18595c4f5e7278a00580255d"
+      },
+      "substitutions": 36
+    },
+    "training_metadata": {
+      "bytes": 1566,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/training/review/quality_terminal_custody_20260909/point_actual_a/capture/raw/checkpoints/train_meta.json",
+      "sha256": "b5fb7c0bee3a8cf472b3ebec24206c4bd9c91f8c0817e20505720003f9e790af"
+    },
+    "world_size": 8
+  },
+  "predecessor_lineage": {
+    "bytes": 21723,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/LINEAGE.json",
+    "sha256": "0f188e13b76d4f618978e751e1a282db53142b6557abf3083dd63dcf7d52ab21"
+  },
+  "private_only": true,
+  "publisher": "HeyDonto Labs",
+  "repository": "HeyDonto/SURGFIELD-ORena-2026-PROCEDURE-FINAL",
+  "repository_mapping_addendum": {
+    "bytes": 5043,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/CACHE_MAPPING_B.json",
+    "sha256": "8dd7bdeb47c5fd2b0b001a917da42900d8f1e73f026a7f095980c0103dd7d394"
+  },
+  "repository_mapping_note": "HF forbids .cache path components: OWL destination is image_root/app/hf_cache. Original /app/.cache paths, aliases and selected image identity remain unchanged. Mapping is storage-only; reconstruct original runtime paths explicitly.",
+  "runtime_priors": {
+    "/app/artifacts/cholec_priors_v39_organizer.json": {
+      "classification": "training_or_annotation_derived_runtime_statistics",
+      "embedded_provenance": {
+        "n_videos": 57,
+        "provenance": {
+          "behavior_delta_named": "cap1d_oe_default changes '1. Clip: top/right' (v38) -> '1. Specimen: top/left' (organizer-derived) \u2014 priced by the paired no-harm read banked beside this artifact",
+          "derivation_run": "modal cholec-priors-regen v3, 2026-08-24 (~15:2xZ), L40S, d8e1e188 image + committed solution overlay",
+          "dua_reading": "lapchole-train processed on OUR OWN Modal compute is lawful \u2014 the DUA fence forbids external-API sends, and this never leaves our tenancy (coordinator-countersigned 2026-08-24)",
+          "instrument": "THE SHIP'S OWN detector (solution/detection.py Owlv2Detector + FOCUS_PROMPTS + its quadrant binning) \u2014 priors derived by the exact instrument that consumes them",
+          "recipe_fidelity": "v38 params held: min_confidence 0.2, 61 frames/video uniform; the ONE change is the ruled source swap",
+          "ruling": "Reza 2026-08-24 (hybrid): keep/disclose v38 in old finished models; REPLACE in all new models",
+          "run_postmortems": [
+            "v1: upload dep missing in-image (return-first cure)",
+            "v2: detection schema guessed not read \u2014 14,258 dets aggregated to zero (read-the-docstring cure)"
+          ]
+        },
+        "source": "ORGANIZER lapchole train videos (m5_phaseB_train77 subset), OWL-V2-large zero-shot via the SHIP detector (solution/detection.py), recipe-faithful to v38 params; replaces CholecTrack20-sourced v38 per the 2026-08-24 ruling"
+      },
+      "file": {
+        "bytes": 3920,
+        "image_path": "/app/artifacts/cholec_priors_v39_organizer.json",
+        "repo_path": "image_root/app/artifacts/cholec_priors_v39_organizer.json",
+        "sha256": "9780988bdba5fbb69fda26bc11efba00bf5bda4bb8981ef53c46b85fc8cbc721"
+      },
+      "role": "OWL-V2-derived quadrant/presence/density statistics on organizer lapchole training videos.",
+      "scope_limits": "Actual artifact names m5_phaseB_train77 subset:57 videos,3477 frames,14258 detections,confidence0.2. Its embedded historical owner ruling is provenance, not a new independent legal conclusion.",
+      "source_loader": {
+        "bytes": 9607,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/priors.py",
+        "sha256": "b35cb57c886bfdb033b92f3ab51f9c07952e60ff5682c9b0eaea9f95859c29a1"
+      }
+    },
+    "/app/artifacts/duration_priors.json": {
+      "classification": "training_or_annotation_derived_runtime_statistics",
+      "embedded_provenance": {
+        "_doc": "PROC 2b duration constant per procedure_type, fitted to maximize train hits under orena Time \u00b15s. Consumed by L16 (pipeline/layers/duration_constant.py). Re-fit on cholec batch-2 via eval/scripts/fit_duration_prior.py <cholec-train.parquet>.",
+        "_fitted_from": "heico-focus-vqa procedure/train.parquet (HF orena-dkfz/heico-focus-vqa; see MODELS.md)"
+      },
+      "file": {
+        "bytes": 508,
+        "image_path": "/app/artifacts/duration_priors.json",
+        "repo_path": "image_root/app/artifacts/duration_priors.json",
+        "sha256": "d59b2a1226f03c114976bca7a3c4ecf7dab4e93c739a1f9c9e313ca0e897f312"
+      },
+      "role": "Procedure duration constants fitted to heico-focus-vqa procedure/train.parquet under +/-5s tolerance.",
+      "scope_limits": "Constructor reads this file; selected L16 applicability is disabled. Not a newly enabled inference intervention.",
+      "source_loader": {
+        "bytes": 9607,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/priors.py",
+        "sha256": "b35cb57c886bfdb033b92f3ab51f9c07952e60ff5682c9b0eaea9f95859c29a1"
+      }
+    },
+    "/app/artifacts/fo_quadrant_priors_colorectal.json": {
+      "classification": "training_or_annotation_derived_runtime_statistics",
+      "embedded_provenance": {},
+      "file": {
+        "bytes": 849,
+        "image_path": "/app/artifacts/fo_quadrant_priors_colorectal.json",
+        "repo_path": "image_root/app/artifacts/fo_quadrant_priors_colorectal.json",
+        "sha256": "0299f175494ce7255bb6488093781f0a25d3e44c5221545bd0e623830bb72fdd"
+      },
+      "role": "Foreign-object quadrant probabilities from HeiCo colorectal training data.",
+      "scope_limits": "This file has no embedded dataset revision; attribution is from the exact source loader docstring.",
+      "source_loader": {
+        "bytes": 9607,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/priors.py",
+        "sha256": "b35cb57c886bfdb033b92f3ab51f9c07952e60ff5682c9b0eaea9f95859c29a1"
+      }
+    },
+    "/app/artifacts/stem_table.json": {
+      "classification": "training_or_annotation_derived_runtime_statistics",
+      "embedded_provenance": {},
+      "file": {
+        "bytes": 6342,
+        "image_path": "/app/artifacts/stem_table.json",
+        "repo_path": "image_root/app/artifacts/stem_table.json",
+        "sha256": "3cd4120e0a637293ce10f6d46224fa7bb894325b7f965c385009aa02dee90acc"
+      },
+      "role": "All-track question-stem answer modes; source says at least five training questions and >=40% support.",
+      "scope_limits": "Source describes training-derived constants; this file has no embedded dataset revision or full derivation receipt.",
+      "source_loader": {
+        "bytes": 9607,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/priors.py",
+        "sha256": "b35cb57c886bfdb033b92f3ab51f9c07952e60ff5682c9b0eaea9f95859c29a1"
+      }
+    },
+    "/app/artifacts/train_modes.json": {
+      "classification": "training_or_annotation_derived_runtime_statistics",
+      "embedded_provenance": {},
+      "file": {
+        "bytes": 2127,
+        "image_path": "/app/artifacts/train_modes.json",
+        "repo_path": "image_root/app/artifacts/train_modes.json",
+        "sha256": "34aede7983742bea062ea1557dc91998fd03b93cb04311540079c10045e92d57"
+      },
+      "role": "Per track/capability/format training truth modes.",
+      "scope_limits": "Source describes training-derived constants; this file has no embedded dataset revision or full derivation receipt.",
+      "source_loader": {
+        "bytes": 9607,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/priors.py",
+        "sha256": "b35cb57c886bfdb033b92f3ab51f9c07952e60ff5682c9b0eaea9f95859c29a1"
+      }
+    },
+    "/app/artifacts/x11_stem_table.json": {
+      "classification": "training_or_annotation_derived_runtime_statistics",
+      "embedded_provenance": {},
+      "file": {
+        "bytes": 8309,
+        "image_path": "/app/artifacts/x11_stem_table.json",
+        "repo_path": "image_root/app/artifacts/x11_stem_table.json",
+        "sha256": "0be76a1c0ea4530414c33cf91aa1707d76a91a6c1fb8c3bf4df7ccf4d5add823"
+      },
+      "role": "Procedure question-stem answer modes.",
+      "scope_limits": "Source describes training-derived constants; this file has no embedded dataset revision or full derivation receipt.",
+      "source_loader": {
+        "bytes": 9607,
+        "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/priors.py",
+        "sha256": "b35cb57c886bfdb033b92f3ab51f9c07952e60ff5682c9b0eaea9f95859c29a1"
+      }
+    }
+  },
+  "runtime_sources": {
+    "runtime_zoom_entry.py": {
+      "bytes": 4645,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/runtime_zoom_entry.py",
+      "sha256": "ac44daa78b946456a755b44c40c06ea07ea8708206d64b98fd2bcb25039f43e5"
+    },
+    "solution/layers/duration_constant.py": {
+      "bytes": 3862,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/layers/duration_constant.py",
+      "sha256": "8c9bdd828635b87fd56e54b5afffd616d0deff3ea284336e36328c209ae4293f"
+    },
+    "solution/layers/vlm_cholec.py": {
+      "bytes": 64217,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/layers/vlm_cholec.py",
+      "sha256": "44df3d7ca03f6e96460334ef59ffb988538b6be0d926cfe359e67728c79a9f7f"
+    },
+    "solution/object_specialist.py": {
+      "bytes": 10032,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/solution/object_specialist.py",
+      "sha256": "3b684cbf7e3581bf63b559cb393294a6c088d84e851d783f676d419d0be196ad"
+    },
+    "verify_vehicle.py": {
+      "bytes": 13796,
+      "path": "/Users/rezanehzati/Projects/orena-finals/packages/foundation-q38/review/exact615_platform_compatibility_20260910/cpu_source_a/source/app/verify_vehicle.py",
+      "sha256": "225f06c54f4fbff02cfa88391aefe04a6449c35c9a43bcb86d33a5969175d9d0"
+    }
+  },
+  "schema": "selected615-hf-lineage/1",
+  "source_manifest": {
+    "bytes": 52053,
+    "path": "/private/tmp/orena-selected615-hf-inventory-20260911-decode-a/SOURCE_MANIFEST.json",
+    "sha256": "df1cf7f38c87eae4719b721a7f4a6e7c1d19d9c59d33d065808f9976d937afd5"
+  }
+}

LOAD.md

@@ -0,0 +1,105 @@
+# Loading the retained PROCEDURE assets
+
+Repository: `HeyDonto/SURGFIELD-ORena-2026-PROCEDURE-FINAL`.
+
+## Pin the asset revision
+
+The private asset commit is pinned below. A fresh Hugging Face download verified all 136 uploaded files (33,969,040,571 bytes) against their full SHA-256 values at 2026-09-10 22:20 UTC. The readback receipt SHA-256 is `14810b6e1436b868c130d8d0e150cb3afceb35685e8b5cbdf0b63e65f04730fb`. A subsequent CPU qualification on that downloaded snapshot passed in 37.203 seconds: all 130 asset/source files and 11 aliases reverified, Qwen3_5Config/Qwen3VLProcessor and Owlv2Config/Owlv2Processor loaded locally, and all 640 adapter tensor headers checked. Its result SHA-256 is `c8440ff025d0ea7135066db41d681b2acba5c8c72e849ffd9bf6a4a303b89f45`. These checks did not load the whole model or perform GPU generation.
+
+```text
+ASSET_REVISION=6a7fc05f196b98d751e3a14775d60f1161166d2a
+```
+
+A later metadata/documentation commit may contain this guide while referring back to this asset commit. Record both commits when reproducing an experiment; do not substitute `main` or a movable tag. The initial empty repository commit is not the asset release.
+
+For an account granted read access to the private repository, use locally configured Hugging Face authentication. Do not paste a token into this guide or into a command that is stored in logs.
+
+```python
+import re
+from huggingface_hub import snapshot_download
+
+ASSET_REVISION = "6a7fc05f196b98d751e3a14775d60f1161166d2a"
+if not re.fullmatch(r"[0-9a-f]{40}", ASSET_REVISION):
+    raise SystemExit("Set a full immutable asset commit before downloading")
+
+snapshot = snapshot_download(
+    repo_id="HeyDonto/SURGFIELD-ORena-2026-PROCEDURE-FINAL",
+    repo_type="model",
+    revision=ASSET_REVISION,
+    local_dir="procedure-final-assets",
+)
+print(snapshot)
+```
+
+The official [Hugging Face download guide](https://huggingface.co/docs/huggingface_hub/guides/download) documents downloading a snapshot at a specific revision. No network access is required during the retained image's inference; repository download is a separate acquisition step.
+
+## Preserve the image paths
+
+Most files are stored under `image_root/` with their original absolute image path made relative: `/app/artifacts/...` becomes `image_root/app/artifacts/...`, and `/app/runtime_zoom_entry.py` becomes `image_root/app/runtime_zoom_entry.py`. The [ASSET_MANIFEST.json](ASSET_MANIFEST.json) lists 46 physical asset files totaling 33,967,875,210 bytes and 11 relative OWLv2 snapshot aliases. [SOURCE_MANIFEST.json](SOURCE_MANIFEST.json) lists the 84 exact selected source files. [IMAGE_IDENTITY.json](IMAGE_IDENTITY.json) binds the native environment and original archive. Both OWLv2 weight formats are retained by the asset inventory; this is a conservative complete cache set, not a minimum-file claim. Hugging Face disallows a `.cache` repository path segment, so the OWLv2 cache alone is stored under `image_root/app/hf_cache/` instead of its original `/app/.cache/`. The manifests preserve both `image_path` and `repo_path`, and aliases explicitly record `resolved_repo_path`. These manifests supply every selected path, byte size and SHA-256. Verify those values before using any component; directory names and adapter configuration alone do not establish tensor identity. Download canonical physical files once, then recreate the eleven relative aliases exactly from their repository mappings when reconstructing the cache. To restore the original image filesystem semantics in a separately assembled root, map the contents of `image_root/app/hf_cache/` back to `/app/.cache/`, preserving the relative links. Do not edit the baked `HF_HOME=/app/.cache/huggingface` configuration and call that an unchanged native run; this package does not overwrite a running container automatically. Never turn an alias into an unrelated model download.
+
+Run the bundled verifier from the trusted documentation checkout against the downloaded snapshot:
+
+```bash
+python3 -B verify_files.py procedure-final-assets --restore-aliases
+```
+
+It authenticates both manifests, streams the SHA-256 of all 130 selected physical asset/source files, and verifies or creates the eleven prescribed relative aliases. It refuses a mismatched existing alias rather than overwriting it. Without `--restore-aliases` it performs read-only verification. The pass result is file verification, not a model-load or GPU-runtime result.
+
+The Dense48 and Point320 adapters are distinct 247,513,400-byte files even though their 1,135-byte configuration files match. Both use the selected checkpoint 12,910. Do not merge, rename or exchange them, substitute another base revision, regenerate int8 weights, or install a later research adapter when reproducing this candidate.
+
+## Load the model components for inspection or custom research
+
+The exact image reports Python **3.11.11** and the following installed versions: Transformers **5.14.0**, PEFT **0.20.0**, Torch **2.13.0**, bitsandbytes **0.50.1**, Accelerate **1.14.0**, safetensors **0.8.0**, tokenizers **0.22.2**, huggingface-hub **1.29.0**, Pillow **12.3.0**, and PyAV **16.1.0**. These are observed installed versions, not the older `PYTORCH_VERSION` environment label. Matching these numbers alone does not reconstruct the CUDA libraries or full image.
+
+[load_components.py](load_components.py) is a concrete CUDA component-loading example derived from the shipped `vlm_cholec.py` and `object_specialist.py` calls. It verifies the extracted files, loads the processor from `image_root/app/artifacts/q38_base`, loads the existing prequantized base from `q38_base_int8_prequant`, attaches Dense48 as `default`, and loads Point320 as `point_object` while preserving RNG state for that second attachment. It does not supply a new quantization configuration or merge the adapters.
+
+```python
+from load_components import load
+
+processor, model = load("procedure-final-assets")
+# Dense48 is active initially. For a custom component-level experiment:
+try:
+    model.set_adapter("point_object", inference_mode=True)
+    model.eval()
+    # Prepare authorized inputs with processor, then call model.generate(...).
+finally:
+    model.set_adapter("default", inference_mode=True)
+    model.eval()
+```
+
+For a separate auxiliary-detector component experiment, use the relocated, exact OWLv2 snapshot rather than a model name resolved through the network:
+
+```python
+from pathlib import Path
+from transformers import Owlv2Processor, Owlv2ForObjectDetection
+
+owl_path = (
+    Path("procedure-final-assets").resolve()
+    / "image_root/app/hf_cache/huggingface/hub"
+    / "models--google--owlv2-large-patch14-ensemble/snapshots"
+    / "95e26936e865f87db1742128404b3c035d47d89d"
+)
+owl_processor = Owlv2Processor.from_pretrained(str(owl_path), local_files_only=True)
+owl_model = Owlv2ForObjectDetection.from_pretrained(
+    str(owl_path), local_files_only=True
+).to("cuda")
+owl_model.eval()
+```
+
+This optional snippet follows the original detector's processor/model API and assumes the manifest verifier has restored the cache aliases. It is not a second whole-pipeline qualification.
+
+The example checks exact package versions and CUDA availability. Its API sequence was checked with explicit CPU doubles and source-call comparison. Actual downloaded-asset CPU configuration, processor and header checks passed separately; the CUDA component-loading function itself has not been executed as a new GPU load. It intentionally does not recreate the native question router, warmups, per-request controller witnesses, input shell or temporal budget policy. Use the unchanged image below for the demonstrated end-to-end behavior.
+
+## Reproduce the demonstrated runtime
+
+The extracted files are not a standalone operating-system or Python environment. Loading them with an arbitrary current Transformers/PEFT installation is not the qualified native runtime. The reference is the retained OCI image:
+
+```text
+us-central1-docker.pkg.dev/heydonto-425716/surgfield/surgfield-proc-q38-validation@sha256:32d4f04c32fc304888b0571ce76a332c8d2fe8d2e4aac262de9709dca07f4291
+```
+
+Its image configuration is `sha256:8d3bbd92c7ef8f73ff058382bef1b8523ec55961b831422c1fb04226b31b35d5`. The retained archive has SHA-256 `d633c21f620c4e4afbdd6dadb181dfc4d41a4fc446a6dbede1718cd949ac6acc` and size 52,710,560,644 bytes. Registry/GCS access is managed separately from private Hugging Face access; the full archive is not duplicated in this asset repository.
+
+The original default command is `/opt/conda/bin/python -B /app/runtime_zoom_entry.py --submission`, working directory `/app`, UID/GID 1000. The shell reads `/input/request.json`, organizer `/input/FO_definitions.json` and the supplied video layout, and writes `/output/answer.json` containing `{qID, content, latency}` records. Preserve the original input contract and use only data you are authorized to process. Private evaluation fixtures and reference answers are not supplied here.
+
+The published native checks used an H100 and, for the fresh full-board comparison, 16 CPUs with 196,608 MiB host memory. Those resources describe that experiment; they are not a proven minimum or a claim that every deployment meets the original process budget. Preserve the image's default entrypoint, flags and environment for an exact comparison.

MODIFICATIONS.md

@@ -0,0 +1,33 @@
+# Modifications and selected-runtime scope
+
+Prepared by **HeyDonto Labs** for `HeyDonto/SURGFIELD-ORena-2026-PROCEDURE-FINAL`. Contact: **Reza Nehzati, Ph.D.**
+
+The handoff extracts selected files from image configuration `8d3bbd92c7ef8f73ff058382bef1b8523ec55961b831422c1fb04226b31b35d5`. Extraction preserves file bytes. It does not merge adapters, quantize a new base, edit the inference source, or turn a later research branch into the retained candidate.
+
+## Model adaptation
+
+The selected runtime records upstream `Qwen/Qwen3.8-27B` with actual configuration architecture `Qwen3_5ForConditionalGeneration`. It uses a prequantized bitsandbytes int8 vision-language base and two separate LoRA adapters, Dense48 and Point320, selected at checkpoint 12,910. Each selected adapter has rank 32, alpha 64 and dropout 0.05. “Point320” names its 320-tensor adapter payload, not 320 training updates. The selected Point adapter completed 12,910 optimizer updates; it is the point-target-restored quality arm, distinct from later private-data continuation experiments. The base's recorded model identifier and its actual architecture field must both be retained in the inventory; the family name is not a substitute for configuration and tensor hashes.
+
+Dense48 is the default selected adapter; Point320 is separately loaded for the configured specialist route. The combined selected adapter payload and runtime lock are checked as distinct parts of the release. Exact model and source inventories define the release; files with similar directory or checkpoint names are not equivalent.
+
+## Selected tensor files
+
+| Original image path | Bytes | SHA-256 |
+|---|---:|---|
+| `/app/artifacts/proc_q38_dense48_ck12910/adapter_model.safetensors` | 247,513,400 | `500126c5a3a7e813483929225bf2321c82f8a5b59c32ed8aac268db34ab7c4fe` |
+| `/app/artifacts/proc_q38_point_ck12910/adapter_model.safetensors` | 247,513,400 | `24762d93abdd35704e0087feefb5a9e4d0794d6b7cea39d0df195f93829481a7` |
+| `/app/artifacts/q38_base_int8_prequant/model.safetensors` | 29,924,045,158 | `238b0622e3b2446e71daeaef8289f9a81560ed26cc280f47182a0475917b6f04` |
+
+The two adapter configuration files share SHA-256 `52e9ce4678fc4bcf4e717245952665c921c13a28247c67b679eeafdb25c4f6f8`. Their matching configuration does not make the adapter tensors identical.
+
+## Inference composition
+
+The retained source composes question routing, sampled video frames, model generation, format validation, rule-based answers and output hygiene. Six training/annotation-derived JSON prior and stem-statistic assets are included alongside learned weights, as listed in DATA_PROVENANCE.md. A source-bound 120-second temporal refinement policy uses actual decoded timestamps and retains the verified coarse answer when the original budget guard declines a refinement. Runtime observations distinguish loader return, per-question generation and rules/fallback behavior.
+
+Private cache preparation, run-as-application setup, input staging and the original default launcher are part of the demonstrated behavior. Merely invoking the base model with a prompt does not reproduce the full PROCEDURE pipeline.
+
+## Excluded successors and legacy material
+
+The shipping-guard `441` image, 240-second policy experiment, private-data training successors, Gate512/LR1920 or recovered1440 branches, V2 observer derivatives and synthetic warmup experiments are not this candidate. Older weights and optimizer state retained in the original image are separately classified by the inventory and excluded from the selected Hugging Face asset set. Their presence in the archive does not make them active selected components.
+
+The original archive remains the lineage reference. The Hugging Face package is an explicitly selected extraction and is not claimed to reproduce every incidental cache file or the entire container filesystem.

ORGANIZER_ACCESS.md

@@ -0,0 +1,13 @@
+# Organizer access
+
+Repository and clean submission-form URL: [HeyDonto/SURGFIELD-ORena-2026-PROCEDURE-FINAL](https://huggingface.co/HeyDonto/SURGFIELD-ORena-2026-PROCEDURE-FINAL).
+
+Maintainer: **HeyDonto Labs**. Corresponding contact: **Reza Nehzati, Ph.D.**
+
+The repository is private. The initial access setup records the dedicated resource group **ORena 2026 PROCEDURE organizer review** (id `6aa329480ec8205ac77ce835`) for this repository and read access for `orena-dkfz` through that group. Organization-wide access remains `no_access`, automatic joining is disabled, and unrelated FRAME, SEGMENT and annotation groups were unchanged in the setup comparison.
+
+This is a configured-access observation. It does not establish that an organizer has accepted an invitation, acknowledged receipt, downloaded the assets, verified their hashes or successfully loaded the model. Organizer acknowledgement has not been received at this preparation stage.
+
+The private asset commit `6a7fc05f196b98d751e3a14775d60f1161166d2a` has been created and passed a fresh full-file download/hash comparison. The initial empty repository commit is not an asset release. LOAD.md pins the actual asset commit, and later documentation updates must continue to point to it. This does not claim a recipient has downloaded or loaded the model.
+
+Access is intended for authorized challenge review. This repository excludes private surgical media, annotation banks and evaluation-bank payloads. The preserved source may retain original unit-test fixtures. Any separately required restricted-data or annotation handoff must use its own authorized channel and disclosure; this model repository does not imply that such a handoff has occurred.

README.md

@@ -1,10 +1,59 @@
 ---
 pretty_name: SURGFIELD ORena 2026 PROCEDURE FINAL
+language:
+- en
+tags:
+- surgical-video-question-answering
+- procedure
+- lora
+- bitsandbytes
+- research
 ---
 # SURGFIELD ORena 2026 PROCEDURE FINAL
 
-Attribution: HeyDonto Labs. Responsible contact: Reza Nehzati, Ph.D.
+**HeyDonto Labs** · Corresponding contact: **Reza Nehzati, Ph.D.**
 
-Private organizer-review repository. The exact selected PROCEDURE model payload and verified loading instructions are being prepared. This initial repository commit is not a completed model release.
+This private repository records the retained PROCEDURE candidate from the ORena–SAVE FOCUS challenge campaign. It contains selected model assets and inference source extracted from the exact image identified below, with component provenance and loading instructions. It does not distribute private surgical media or evaluation banks. The exact inference source may retain its original unit-test fixtures; those fixtures are distinct from private evaluation-bank payloads.
 
-Selected image configuration: `8d3bbd92c7ef8f73ff058382bef1b8523ec55961b831422c1fb04226b31b35d5`.
+The final candidate is the retained `8d3` image. The repository is a component and source handoff; the original image remains the reference environment for the demonstrated end-to-end behavior. The private asset commit is `6a7fc05f196b98d751e3a14775d60f1161166d2a`, as pinned in [LOAD.md](LOAD.md). A fresh Hugging Face download verified all 136 uploaded files against their full SHA-256 values. A separate CPU check also reverified all 130 asset/source files and 11 aliases, loaded the actual processor/configuration objects, and checked all 640 adapter tensor headers. It did not load the whole model or perform GPU generation. A later documentation commit does not change the asset revision.
+
+## Exact candidate
+
+| Identity | Value |
+|---|---|
+| OCI image configuration | `sha256:8d3bbd92c7ef8f73ff058382bef1b8523ec55961b831422c1fb04226b31b35d5` |
+| Registry manifest | `sha256:32d4f04c32fc304888b0571ce76a332c8d2fe8d2e4aac262de9709dca07f4291` |
+| Retained Docker archive SHA-256 | `d633c21f620c4e4afbdd6dadb181dfc4d41a4fc446a6dbede1718cd949ac6acc` |
+| Retained archive size | 52,710,560,644 bytes |
+| Native entrypoint | `/opt/conda/bin/python -B /app/runtime_zoom_entry.py --submission` |
+| Native identity | UID/GID 1000; Python 3.11.11 |
+
+## Method
+
+The serving path combines an int8 vision-language base with separate Dense48 and Point320 LoRA adapters at checkpoint 12,910, a question router, answer-format validation and a budget-aware temporal refinement policy. The retained temporal policy uses a 120-second refinement window and 48 sampled frames. A refinement may be declined by the original remaining-budget guard; completion does not mean every question used a second model pass. Rule answers and degraded fallbacks are distinct from observed model generation.
+
+The recorded upstream base is `Qwen/Qwen3.8-27B`; the shipped configuration architecture is `Qwen3_5ForConditionalGeneration`. Point320 denotes its 320-tensor adapter payload, not 320 optimizer steps. Exact configuration, adapter hyperparameters, auxiliary assets and upstream revisions are disclosed in the asset inventory and [MODIFICATIONS.md](MODIFICATIONS.md). Older weights retained inside the original image are listed separately from the active selected payload. They are not silently substituted into this handoff.
+
+## Evaluation and limits
+
+These are local research results, not official challenge standings. [EVALUATION_SUMMARY.json](EVALUATION_SUMMARY.json) provides aggregate counts and the independent review hash without releasing per-question data.
+
+| Evaluation protocol | Selected candidate | Comparator | Processing failures retained |
+|---|---:|---:|---|
+| Historical local 1,087-question protocol | 615/1,087 correct; five-bucket mean 0.6191042 | Not a fresh native comparison | Historical protocol recorded 1,087 answered |
+| Fresh native default-entrypoint comparison, 1,087 questions per image | 579/1,087 correct; five-bucket mean 0.5593201 | Exact pfull best-platform variant: 444/1,087; mean 0.4653223 | Selected 67; comparator 124 |
+| Jointly native- and pool-qualified sensitivity population | 511/910 correct | 421/910 correct | Subset only; primary denominators remain 1,087 each |
+
+The fresh comparison used the unchanged images on H100 with 16 CPUs and 196,608 MiB host memory, and the original external process allowances of `120 + 30 × number_of_questions` seconds. The selected candidate's 67 failures comprise 28 questions in two stream-cap terminations and 39 questions in three groups excluded by the frozen validation-demotion rule. Comparator failures include nine original process-pool timeouts. Failed rows were not removed from the primary scores.
+
+The 1,087-question development population contains no challenge OOD rows and only one clinical-flagged row. It was used during development and selection; repeated use, correlated questions within videos and selection bias limit transfer claims. The results establish neither a clinical accuracy estimate nor a finals Copeland rank, and do not establish official platform superiority.
+
+The exact archive also completed the organizer's ten-question canonical compatibility fixture in two local input layouts. Owner-supplied platform try-out output matched those ten answer strings. Each layout showed nine question-bound VLM answers and one rule answer for the original invalid clip. This is an execution-compatibility observation, not an accuracy benchmark or independent organizer approval of this repository.
+
+A separate private, one-surgery, six-predicate diagnostic retained all 12 selected-image observations as processing/observation failures; it yielded no qualified matched accuracy comparison. Its media, labels and derivatives remain evaluation-only and are not included here.
+
+## Data and intended use
+
+This release supports authorized challenge review and surgical-video VQA research. Dataset provenance, annotation sources and restrictions are recorded in [DATA_PROVENANCE.md](DATA_PROVENANCE.md). Private surgical media, annotation banks and evaluation-bank payloads are not distributed with the model. This research system has no established clinical safety or diagnostic performance and is not validated for patient-care decisions.
+
+Component licenses and notices are documented in [UPSTREAM_NOTICES.md](UPSTREAM_NOTICES.md) and [LICENSE_PROVENANCE.md](LICENSE_PROVENANCE.md). [ORGANIZER_ACCESS.md](ORGANIZER_ACCESS.md) separates configured repository access from organizer acknowledgement.

UPSTREAM_NOTICES.md

@@ -0,0 +1,23 @@
+# Upstream notices
+
+This private handoff is attributed to **HeyDonto Labs**; contact **Reza Nehzati, Ph.D.** Upstream authors retain their respective copyrights and licenses.
+
+## Qwen base and selected adaptations
+
+The recorded upstream is **Qwen/Qwen3.8-27B**, revision `1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0`. Its configuration declares `Qwen3_5ForConditionalGeneration`. The pinned upstream configuration, README and LICENSE were fetched independently and matched the exact retained image's metadata bytes. The original model card identifies Apache License 2.0. See the [pinned upstream model card](https://huggingface.co/Qwen/Qwen3.8-27B/blob/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/README.md) and [pinned license](https://huggingface.co/Qwen/Qwen3.8-27B/blob/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/LICENSE).
+
+The original license is included at [image_root/app/artifacts/q38_base/LICENSE](image_root/app/artifacts/q38_base/LICENSE), SHA-256 `bbedc3fda3305820b977265f01b8619d87570a6739de3a5582c3464840f1e57a`. The model card is retained at [image_root/app/artifacts/q38_base/README.md](image_root/app/artifacts/q38_base/README.md). Team quantization and the selected Dense48/Point320 adapters are described in MODIFICATIONS.md; these changes do not replace the upstream attribution.
+
+## OWLv2 auxiliary assets
+
+The auxiliary cache records **google/owlv2-large-patch14-ensemble**, revision `95e26936e865f87db1742128404b3c035d47d89d`. Its [pinned Google model card](https://huggingface.co/google/owlv2-large-patch14-ensemble/blob/95e26936e865f87db1742128404b3c035d47d89d/README.md) identifies Apache License 2.0 and cites the OWL-ViT/OWLv2 work. The independently retrieved card matches the cached card, SHA-256 `a2e10c3916166f08eaf2ab43ca1eb63c6116df228dea655c56e4c8e1607ecfe9`. Both authenticated cached weight formats and their metadata are retained; the alias map avoids duplicate physical uploads of a logical snapshot file.
+
+## Runtime software and excluded cache material
+
+The reference image uses third-party libraries including PyTorch, Transformers, PEFT, bitsandbytes, Accelerate, safetensors, tokenizers, Hugging Face Hub, Pillow and PyAV. EXACT_DEPENDENCIES.json records the installed versions observed inside the image. Their applicable licenses and notices are separate from the model license; this extracted asset handoff does not redistribute the complete Python/CUDA environment or relicense those packages.
+
+The original image also contains inactive Qwen3-VL-8B, older adapter/checkpoint and optimizer payloads. They are classified in ASSET_MANIFEST.json and excluded from this selected component handoff. They are not additional active bases for the published loader.
+
+## Data-derived assets
+
+This repository does not distribute private surgical media or evaluation banks. Its exact source may contain original unit-test fixtures, and its selected runtime assets include six disclosed training/annotation-derived prior and statistic JSON files. These carry their source provenance and restrictions; the base-model license is not a grant of rights to private data. DATA_PROVENANCE.md and LICENSE_PROVENANCE.md record that separate scope.

load_components.py

@@ -0,0 +1,71 @@
+"""Load the extracted selected components with the recorded native library versions.
+
+This example does not install the native router, warmups, input shell, controller
+witnesses or temporal budget policy, and is not a full-pipeline runtime claim.
+"""
+import argparse
+from importlib.metadata import version
+import os
+from pathlib import Path
+import platform
+
+EXPECTED_VERSIONS = {
+    'Pillow': '12.3.0', 'accelerate': '1.14.0', 'av': '16.1.0',
+    'bitsandbytes': '0.50.1', 'huggingface-hub': '1.29.0', 'peft': '0.20.0',
+    'safetensors': '0.8.0', 'tokenizers': '0.22.2', 'torch': '2.13.0',
+    'transformers': '5.14.0',
+}
+
+
+def load(snapshot):
+    if platform.python_version() != '3.11.11':
+        raise RuntimeError('Use the recorded Python 3.11.11 environment')
+    actual = {name: version(name) for name in EXPECTED_VERSIONS}
+    if actual != EXPECTED_VERSIONS:
+        raise RuntimeError('Installed library versions differ from the retained image')
+    for name in ('HF_HUB_OFFLINE', 'TRANSFORMERS_OFFLINE', 'HF_DATASETS_OFFLINE'):
+        os.environ[name] = '1'
+    from verify_files import verify
+    verify(snapshot, restore_aliases=True)
+    import torch
+    from transformers import AutoModelForImageTextToText, AutoProcessor
+    from peft import PeftModel
+    if not torch.cuda.is_available():
+        raise RuntimeError('The selected int8 component-loading example requires CUDA')
+    artifacts = Path(snapshot).resolve(strict=True) / 'image_root' / 'app' / 'artifacts'
+    processor = AutoProcessor.from_pretrained(
+        str(artifacts / 'q38_base'), trust_remote_code=True, local_files_only=True)
+    # The saved prequant config already declares bitsandbytes int8. Do not
+    # supply a new BitsAndBytesConfig or re-quantize the base at load time.
+    base = AutoModelForImageTextToText.from_pretrained(
+        str(artifacts / 'q38_base_int8_prequant'),
+        torch_dtype=torch.bfloat16, device_map='cuda', attn_implementation='sdpa',
+        trust_remote_code=True, local_files_only=True)
+    model = PeftModel.from_pretrained(
+        base, str(artifacts / 'proc_q38_dense48_ck12910'),
+        adapter_name='default', is_trainable=False, local_files_only=True)
+    devices = sorted({p.device.index for p in model.parameters() if p.device.type == 'cuda'})
+    with torch.random.fork_rng(devices=devices):
+        result = model.load_adapter(
+            str(artifacts / 'proc_q38_point_ck12910'), 'point_object',
+            is_trainable=False, local_files_only=True, autocast_adapter_dtype=True)
+    if result.missing_keys or result.unexpected_keys:
+        raise RuntimeError('Point adapter load has missing or unexpected keys')
+    if set(model.peft_config) != {'default', 'point_object'}:
+        raise RuntimeError('The selected component pair was not loaded')
+    model.set_adapter('default', inference_mode=True)
+    model.eval()
+    return processor, model
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('snapshot', type=Path)
+    args = parser.parse_args()
+    processor, model = load(args.snapshot)
+    print({'processor_type': type(processor).__name__, 'model_type': type(model).__name__,
+           'active_adapter': model.active_adapter, 'full_pipeline_qualified': False})
+
+
+if __name__ == '__main__':
+    main()

verify_files.py

@@ -0,0 +1,98 @@
+"""Verify the selected extracted files; optionally restore exact OWLv2 aliases.
+
+This does not load a model or reproduce the original OCI environment.
+"""
+import argparse
+import hashlib
+import json
+import os
+from pathlib import Path, PurePosixPath
+import stat
+
+MANIFEST_SHA256 = {
+    'ASSET_MANIFEST.json': 'bf9706c9c565dae44f5c3285c565490c89c84c44194d48130d39cfcbd238824d',
+    'SOURCE_MANIFEST.json': 'df1cf7f38c87eae4719b721a7f4a6e7c1d19d9c59d33d065808f9976d937afd5',
+}
+
+
+def digest(path):
+    h = hashlib.sha256()
+    with path.open('rb') as stream:
+        for chunk in iter(lambda: stream.read(8 * 1024 * 1024), b''):
+            h.update(chunk)
+    return h.hexdigest()
+
+
+def member(root, relative):
+    p = PurePosixPath(relative)
+    if p.is_absolute() or not p.parts or '..' in p.parts or p.parts[0] != 'image_root':
+        raise ValueError('invalid manifest repository path')
+    path = root.joinpath(*p.parts)
+    if not path.parent.resolve().is_relative_to(root):
+        raise ValueError('path leaves snapshot')
+    return path
+
+
+def checked_manifest(root, name):
+    p = root / name
+    raw = p.read_bytes()
+    if hashlib.sha256(raw).hexdigest() != MANIFEST_SHA256[name]:
+        raise ValueError('manifest identity mismatch: ' + name)
+    return json.loads(raw)
+
+
+def verify(root, restore_aliases=False):
+    root = Path(root).resolve(strict=True)
+    assets = checked_manifest(root, 'ASSET_MANIFEST.json')
+    sources = checked_manifest(root, 'SOURCE_MANIFEST.json')
+    if (assets['schema'] != 'selected615-hf-asset-manifest/1' or
+            sources['schema'] != 'selected615-hf-source-manifest/1'):
+        raise ValueError('manifest schema mismatch')
+    files = list(assets['regular_files'].values()) + list(sources['files'].values())
+    checked = {}
+    for entry in files:
+        p = member(root, entry['repo_path'])
+        before = p.lstat()
+        if not stat.S_ISREG(before.st_mode) or before.st_size != entry['bytes']:
+            raise ValueError('file kind or size mismatch: ' + entry['repo_path'])
+        actual = digest(p)
+        after = p.lstat()
+        if actual != entry['sha256'] or (before.st_dev, before.st_ino, before.st_size, before.st_mtime_ns) != (after.st_dev, after.st_ino, after.st_size, after.st_mtime_ns):
+            raise ValueError('file bytes changed or mismatched: ' + entry['repo_path'])
+        checked[entry['repo_path']] = entry
+    restored = 0
+    for entry in assets['aliases'].values():
+        alias = member(root, entry['repo_path'])
+        target = member(root, entry['resolved_repo_path'])
+        target_entry = checked.get(entry['resolved_repo_path'])
+        if target_entry is None or any(target_entry[k] != entry['target_pin'][k] for k in ('bytes', 'sha256')):
+            raise ValueError('alias target is not an authenticated physical file')
+        link = entry['relative_link_target']
+        if PurePosixPath(link).is_absolute() or (alias.parent / link).resolve() != target.resolve():
+            raise ValueError('alias target mapping mismatch')
+        if os.path.lexists(alias):
+            if not alias.is_symlink() or os.readlink(alias) != link:
+                raise ValueError('existing alias differs: ' + entry['repo_path'])
+        elif restore_aliases:
+            alias.parent.mkdir(parents=True, exist_ok=True)
+            if not alias.parent.resolve().is_relative_to(root):
+                raise ValueError('alias parent leaves snapshot')
+            alias.symlink_to(link)
+            restored += 1
+        else:
+            raise ValueError('alias missing; rerun with --restore-aliases: ' + entry['repo_path'])
+    for name in MANIFEST_SHA256:
+        checked_manifest(root, name)
+    return {'physical_files_verified': len(checked), 'aliases_verified': len(assets['aliases']), 'aliases_restored': restored, 'model_loaded': False}
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('snapshot', type=Path)
+    parser.add_argument('--restore-aliases', action='store_true')
+    args = parser.parse_args()
+    print(json.dumps(verify(args.snapshot, args.restore_aliases), sort_keys=True))
+
+
+if __name__ == '__main__':
+    main()