Release visual answerability benchmark v1.0.0

12,000 evaluated views; 9,000 embedded licensed images; GQA references and reconstruction; bounded chart proofs and numerical replay. Frozen manifest verified locally.

.gitattributes

@@ -1,60 +1,2 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.avro filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.lz4 filter=lfs diff=lfs merge=lfs -text
-*.mds filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
 *.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
-# Audio files - uncompressed
-*.pcm filter=lfs diff=lfs merge=lfs -text
-*.sam filter=lfs diff=lfs merge=lfs -text
-*.raw filter=lfs diff=lfs merge=lfs -text
-# Audio files - compressed
-*.aac filter=lfs diff=lfs merge=lfs -text
-*.flac filter=lfs diff=lfs merge=lfs -text
-*.mp3 filter=lfs diff=lfs merge=lfs -text
-*.ogg filter=lfs diff=lfs merge=lfs -text
-*.wav filter=lfs diff=lfs merge=lfs -text
-# Image files - uncompressed
-*.bmp filter=lfs diff=lfs merge=lfs -text
-*.gif filter=lfs diff=lfs merge=lfs -text
-*.png filter=lfs diff=lfs merge=lfs -text
-*.tiff filter=lfs diff=lfs merge=lfs -text
-# Image files - compressed
-*.jpg filter=lfs diff=lfs merge=lfs -text
-*.jpeg filter=lfs diff=lfs merge=lfs -text
-*.webp filter=lfs diff=lfs merge=lfs -text
-# Video files - compressed
-*.mp4 filter=lfs diff=lfs merge=lfs -text
-*.webm filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text

ATTRIBUTION.md

@@ -0,0 +1,53 @@
+# Upstream attribution and modifications
+
+License sources checked on **2026-09-28**. License texts and download URLs are
+recorded in `licenses/`; pinned construction resources are recorded in
+`protocol/source-resources.json`.
+
+**PlotQA: Reasoning over Scientific Plots** — Nitesh Methani, Pritha Ganguly,
+Mitesh M. Khapra, and Pratyush Kumar, WACV 2020.
+[Official dataset/code license statement](https://github.com/NiteshMethani/PlotQA#license)
+and [pinned source instructions](https://github.com/NiteshMethani/PlotQA/blob/e0f5c34acbe92753f70798aafda99184e3e0cd8c/PlotQA_Dataset.md).
+Dataset: [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
+The construction used the pinned
+[Dodon mirror](https://huggingface.co/datasets/Dodon/plotqa-dataset/tree/6e97b674c6afcc28f79b90108e4d736b08bb2966)
+as a transport source, retaining official upstream terms. Charts are re-rendered
+from source tables; target values, visibility and referents are changed under
+explicit programs while question text is preserved. The official PlotQA paper
+describes source data collected from real-world organizations; this package
+does not assert ownership of those underlying facts or additional third-party
+rights. Numeric table provenance remains in the per-world metadata.
+
+**CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual
+Reasoning** — Justin Johnson, Bharath Hariharan, Laurens van der Maaten,
+Li Fei-Fei, C. Lawrence Zitnick, and Ross Girshick, CVPR 2017.
+[Official data and license statement](https://cs.stanford.edu/people/jcjohns/clevr/).
+Dataset: [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
+[Official generator](https://github.com/facebookresearch/clevr-dataset-gen/tree/f0ce2c81750bfae09b5bf94d009f42e055f2cb3a).
+Scenes are freshly rendered with the official Blender 2.79b/Cycles assets, with
+source-program attribute changes and object visibility interventions. New label
+and edit evidence is supplied separately from the upstream data.
+
+**GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question
+Answering** — Drew A. Hudson and Christopher D. Manning, CVPR 2019.
+[Official source downloads](https://cs.stanford.edu/people/dorarad/gqa/download.html)
+and [official question-generator repository](https://github.com/dorarad/gqa).
+No explicit annotation redistribution grant was located on these pages or in
+the downloaded `questions1.2.zip` and `sceneGraphs.zip` README files. The public
+candidate uses source references and local reconstruction. The reconstruction
+normalizes source JPEGs and applies constant gray patches at source object
+boxes; symbolic world edits and source-program alternatives support its labels.
+
+**Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image
+Annotations** — Ranjay Krishna and collaborators, IJCV 2017.
+[Official data page](https://homes.cs.washington.edu/~ranjay/visualgenome/api.html)
+links [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
+This is the source of GQA photographs and the basis of its scene annotations.
+Visual Genome's notice is retained as provenance, not used to infer a license
+for all GQA-authored annotations. No photographs are bundled in this candidate.
+
+The project software is the Explicit Learning project (2026), Apache 2.0 as
+declared by its package metadata. The dataset release and documentation are
+prepared by Sungguk Cha, Mintae Kim, Youngsub Han, Byoung-Ki Jeon, and Sangyeob
+Lee, using the manuscript author list. No upstream author endorses these edits
+or the reported model results by virtue of this attribution.

CITATION.bib

@@ -0,0 +1,32 @@
+@misc{cha2026visualanswerability,
+  title = {When Does an Image Determine the Answer? Benchmarking Visual Answerability across Charts and Scenes},
+  author = {Cha, Sungguk and Kim, Mintae and Han, Youngsub and Jeon, Byoung-Ki and Lee, Sangyeob},
+  year = {2026},
+  url = {https://huggingface.co/datasets/sungguk/visual-answerability},
+  version = {1.0.0},
+  note = {Dataset release answerability-v1.0.0-20260928; Hub tag v1.0.0}
+}
+
+@inproceedings{methani2020plotqa,
+  title = {PlotQA: Reasoning over Scientific Plots},
+  author = {Methani, Nitesh and Ganguly, Pritha and Khapra, Mitesh M. and Kumar, Pratyush},
+  booktitle = {WACV}, year = {2020}
+}
+
+@inproceedings{johnson2017clevr,
+  title = {CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning},
+  author = {Johnson, Justin and Hariharan, Bharath and van der Maaten, Laurens and Li, Fei-Fei and Zitnick, C. Lawrence and Girshick, Ross},
+  booktitle = {CVPR}, year = {2017}
+}
+
+@inproceedings{hudson2019gqa,
+  title = {GQA: A New Dataset for Real-World Visual Reasoning and Compositional Question Answering},
+  author = {Hudson, Drew A. and Manning, Christopher D.},
+  booktitle = {CVPR}, year = {2019}
+}
+
+@article{krishna2017visualgenome,
+  title = {Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations},
+  author = {Krishna, Ranjay and Zhu, Yuke and Groth, Oliver and Johnson, Justin and Hata, Kenji and Kravitz, Joshua and Chen, Stephanie and Kalantidis, Yannis and Li, Li-Jia and Shamma, David A. and Bernstein, Michael S. and Fei-Fei, Li},
+  journal = {International Journal of Computer Vision}, year = {2017}
+}

CITATION.cff

@@ -0,0 +1,20 @@
+cff-version: 1.2.0
+message: "Cite this frozen dataset and the upstream datasets in ATTRIBUTION.md."
+type: dataset
+title: "Visual Answerability across Charts and Scenes"
+version: "answerability-v1.0.0-20260928"
+authors:
+  - family-names: Cha
+    given-names: Sungguk
+  - family-names: Kim
+    given-names: Mintae
+  - family-names: Han
+    given-names: Youngsub
+  - family-names: Jeon
+    given-names: Byoung-Ki
+  - family-names: Lee
+    given-names: Sangyeob
+date-released: 2026-09-28
+url: "https://huggingface.co/datasets/sungguk/visual-answerability"
+repository-code: "https://huggingface.co/datasets/sungguk/visual-answerability/tree/v1.0.0"
+abstract: "Frozen evaluated cohorts, bounded chart witnesses, scene edit evidence, and scoring replay for visual answerability. Hub release v1.0.0."

EVALUATION.md

@@ -0,0 +1,95 @@
+# Evaluate new predictions and replay the paper
+
+Install `requirements.txt` in a Python 3.12+ environment. The verified runtime
+and exact package versions are in `protocol/verified-runtime.json`. All commands
+below run on CPU, use local files, and make no model/API calls. Run from this
+dataset folder, with outputs outside it so the frozen manifest stays valid.
+
+## New model predictions
+
+Use only `image` and `question` as model inputs. Read the exact prompt and schema
+from `protocol/prompt.json`. Obtain GQA inputs using `RECONSTRUCTION.md` first.
+Write one JSON object per input in `predictions.jsonl`:
+
+```json
+{"item_id":"item-0001","answerable":false,"answer":null,"reason":"The needed value is not visible."}
+```
+
+That is a format example, not a prescribed answer for other inputs. Retain each
+stable item ID. For a terminal malformed model output use:
+
+```json
+{"item_id":"item-0001","status":"invalid_schema","terminal_invalid":true,"parsed":null}
+```
+
+One row is required for every selected input. Do not submit both examples for
+the same ID. Missing/duplicate/unexpected IDs or incomplete state groups cause
+the scorer to fail rather than silently shrink the denominator. Invalid final
+outputs count as failures. A valid answerable decision with a null answer can
+pass decision scoring but fails answer correctness. Unanswerable decisions
+must have a null answer and all valid outputs need a nonempty reason.
+
+```bash
+python scripts/score_hf_dataset.py --predictions ../predictions.jsonl \
+  --sources plotqa clevr gqa --output ../scores.json
+```
+
+For a single subset, supply only its predictions and e.g. `--sources plotqa`.
+Output includes exact counts, per-state failures, per-view decision accuracy,
+all-decisions-correct groups, supported-answer accuracy, class-balanced failure,
+and complete joint answer/abstention success. Chart answers use the paper's
+conservative exact-rational parser (no numeric tolerance). Scene scoring first
+checks exact rationals, then Unicode NFKC, case and whitespace normalization;
+it does not infer synonyms. A percentage marker follows the historical parser
+and is not automatically divided by 100.
+
+The GQA flag in the metadata defines the same 376-group location sensitivity
+stratum as the paper. The supplied saved-response replay reports both strata.
+
+## Reproduce reported statistics
+
+```bash
+python scripts/replay_hf_statistics.py --output ../replay
+```
+
+This checks all 72,000 scoring projections, recomputes all 18 model/domain cells
+and the paper's 10,000-draw paired group bootstrap intervals, compares the full
+analysis JSON to the frozen expected result, regenerates tables and chart
+diagnostics, verifies the 5,000 chart view-to-proof identity joins, and runs the
+new-prediction scorer against all six saved configurations. It also checks the
+finite-study counts and enumeration/SMT agreement from saved per-case records.
+It does not regenerate the finite solver-study witnesses, raw generations or
+historical serving traces. GQA question-ID tokens replace upstream question
+text consistently in numerical replay; reconstruction checks the actual text.
+
+The supplementary shortcut and answer-error counts can also be checked directly:
+
+```bash
+python scripts/verify_argument_evidence.py --inputs inputs \
+  --finite-results evidence/finite/original-results.jsonl.gz \
+  --solver-results evidence/finite/solver-results.jsonl.gz \
+  --output ../supplementary-counts.json
+```
+
+```bash
+python scripts/verify_hf_dataset.py --workers 8 --output ../verification.json
+```
+
+This verifies the frozen file manifest, loads all three configurations through
+Hugging Face Datasets, decodes and hashes all 9,000 bundled images, checks every
+state group and unique source-image identity, verifies all 1,000 complete chart
+proofs against their evaluated PNGs, and replays chart/CLEVR source programs.
+The GQA source-program and image check is run by the separate reconstruction
+command. A small machine may reduce `--workers`.
+
+To run only the chart proof checker:
+
+```bash
+python scripts/verify_bounded_witness_bundle.py \
+  --proofs evidence/plotqa/bounded-witness-proofs.jsonl.gz \
+  --workers 8 --output ../chart-proof-verification.json
+```
+
+The prepared package includes historical counts plus current validation reports.
+Reports are evidence of the checks they describe, not a guarantee about every
+upstream annotation or a substitute for independent scientific review.

INFERENCE_ENVIRONMENTS.json

@@ -0,0 +1,85 @@
+{
+  "scope": "Historical inference evidence; separate from the pinned current CPU reproduction environment",
+  "models": {
+    "qwen38-flash-next": {
+      "model": "Qwen/Qwen3.8-Flash-Next",
+      "temperature": 0,
+      "max_tokens": 512,
+      "serving_engine": "vLLM (historical manuscript record)",
+      "runtime_versions": null,
+      "checkpoint_revision": null,
+      "provenance_limit": "Per-run raw/native contracts are not in the portable numerical archive; unknown versions are not inferred.",
+      "chart_checkpoint_revision": "de4b8e4d43b917e7706784d8bb445c9af86a3540",
+      "chart_decoding": {
+        "temperature": 0,
+        "top_p": 1,
+        "max_tokens": 512,
+        "structured_output": "json_schema",
+        "thinking": false,
+        "seed": 20260914
+      },
+      "chart_runtime": {
+        "vllm": "0.1.dev20073+g8e685d198",
+        "transformers": "5.15.1",
+        "dtype": "bfloat16",
+        "batch": 32,
+        "tensor_parallel_size": 2,
+        "max_model_len": 16384,
+        "enforce_eager": false,
+        "gpu_memory_utilization": 0.9,
+        "ple_cpu_offload": true
+      },
+      "chart_source_record": "results/visual_benchmark_20260915/qwen-next-full/summary.json"
+    },
+    "gemma4-31b-it": {
+      "model": "google/gemma-4-31B-it",
+      "temperature": 0,
+      "max_tokens": 512,
+      "serving_engine": "vLLM (historical manuscript record)",
+      "runtime_versions": null,
+      "checkpoint_revision": null,
+      "provenance_limit": "Per-run raw/native contracts are not in the portable numerical archive; unknown versions are not inferred."
+    },
+    "qwen": {
+      "model": "Qwen/Qwen3.8-27B-FP8",
+      "temperature": 0,
+      "max_tokens": 512,
+      "serving_engine": "vLLM (historical manuscript record)",
+      "runtime_versions": null,
+      "checkpoint_revision": null,
+      "provenance_limit": "Per-run raw/native contracts are not in the portable numerical archive; unknown versions are not inferred.",
+      "scene_native_checkpoint_revision": "017b9c7af6b5689d5dd426a76e0bc077eb5ca20a",
+      "bridge_checkpoint_revision": null,
+      "scene_native_settings": "inputs/crossdomain/qwen-backend-provenance.json"
+    },
+    "gemma4-26b-a4b-it": {
+      "model": "google/gemma-4-26B-A4B-it",
+      "temperature": 0,
+      "max_tokens": 512,
+      "serving_engine": "vLLM (historical manuscript record)",
+      "runtime_versions": null,
+      "checkpoint_revision": null,
+      "provenance_limit": "Per-run raw/native contracts are not in the portable numerical archive; unknown versions are not inferred."
+    },
+    "glm": {
+      "model": "GLM-5.3-Flash",
+      "temperature": 0,
+      "max_tokens": 512,
+      "serving_engine": "vLLM (historical manuscript record)",
+      "runtime_versions": null,
+      "checkpoint_revision": null,
+      "provenance_limit": "Per-run raw/native contracts are not in the portable numerical archive; unknown versions are not inferred."
+    },
+    "molmo2-8b": {
+      "model": "allenai/Molmo2-8B",
+      "temperature": 0,
+      "max_tokens": 512,
+      "serving_engine": "vLLM (historical manuscript record)",
+      "runtime_versions": null,
+      "checkpoint_revision": null,
+      "provenance_limit": "Per-run raw/native contracts are not in the portable numerical archive; unknown versions are not inferred."
+    }
+  },
+  "raw_token_archives_bundled": false,
+  "new_inference_performed": false
+}

LICENSE.md

@@ -0,0 +1,41 @@
+# Component licenses
+
+This is an aggregation with distinct component terms, not a blanket relicensing
+of upstream content. The card's `other` license points here.
+
+- **PlotQA-derived images, questions, tables and evidence:** Creative Commons
+  Attribution 4.0 International. The official PlotQA README explicitly applies
+  CC BY 4.0 to its datasets; its MIT license applies to code/models. Preserve
+  the PlotQA attribution and the description of our re-rendering and edits.
+- **CLEVR-derived images, questions, scenes, programs and evidence:** Creative
+  Commons Attribution 4.0 International, as stated on the official CLEVR page.
+  Preserve the CLEVR attribution and identify these as modified/re-rendered
+  evaluation views. CLEVR generator code/assets have their own BSD license;
+  the upstream generator/assets are obtained separately, not bundled here.
+- **Release-authored labels, edit recipes, documentation, benchmark metadata
+  and scoring projections:** CC BY 4.0, to the extent the project holds rights.
+  This does not grant additional rights over source datasets or model weights.
+- **Project software under `src/` and `scripts/`:** Apache License 2.0,
+  following the existing project's `pyproject.toml` declaration. Upstream
+  attribution must remain intact. External dependencies retain their licenses
+  and are installed separately; their binaries and model weights are absent.
+- **GQA upstream questions, full programs, scene graphs and photographs:**
+  not redistributed. The official GQA site and downloaded archive READMEs
+  checked on 2026-09-28 do not state an explicit annotation redistribution
+  license. A local configuration's `cc-by-4.0` declaration and third-party
+  mirror cards are insufficient evidence to settle that grant. Users obtain
+  these inputs directly from the upstream source under its applicable terms.
+  The GQA index contains our labels, factual source IDs and our edit deltas.
+- **Manuscripts:** not bundled. Their publication/license selection is separate
+  from this dataset preparation.
+
+Full license texts are in `licenses/CC-BY-4.0.txt` and
+`licenses/Apache-2.0.txt`. The upstream PlotQA MIT and CLEVR BSD notices are
+also retained for clarity; they do not replace the dataset CC BY terms.
+
+CC BY users must credit the creators, link the license, identify modifications,
+retain applicable notices, and not imply upstream endorsement. No extra legal
+or technical restriction is placed on the rights granted by that license.
+The official Visual Genome site links CC BY 4.0, but this does not resolve the
+separate GQA annotation grant. Original image-owner and other third-party rights
+are not overridden. See `ATTRIBUTION.md` for the source links and modifications.

MANIFEST.json

@@ -0,0 +1,950 @@
+{
+  "version": "answerability-v1.0.0-20260928",
+  "purpose": "Release integrity and provenance, separate from semantic/pixel witness evidence",
+  "files": {
+    ".gitattributes": {
+      "bytes": 87,
+      "sha256": "504e4b75f27f6e10572b9ae69a12a996148e56381b54ac8de5d85c25a7767217"
+    },
+    "ATTRIBUTION.md": {
+      "bytes": 3588,
+      "sha256": "52a8ebff46143de561e1a646a0129e324840b2b8b87393165d720f780e010418"
+    },
+    "CITATION.bib": {
+      "bytes": 1638,
+      "sha256": "aa63479fd052b8ba297b4c5a139c62c1dc30c3e0c048f70c55a0971a8257169f"
+    },
+    "CITATION.cff": {
+      "bytes": 793,
+      "sha256": "0fa5de2a06290f2cb8bb5f297970f6549f1b17f0b594c39dac136379d80061bc"
+    },
+    "EVALUATION.md": {
+      "bytes": 4422,
+      "sha256": "8feffcf506bece1af3b00378019be76260c4a00d0e47388b39731ebf407e5197"
+    },
+    "INFERENCE_ENVIRONMENTS.json": {
+      "bytes": 3403,
+      "sha256": "d8d28db3ea7796bda3dbff1e46a75ed0875a24d8412e3f3bb31329635abd73f9"
+    },
+    "LICENSE.md": {
+      "bytes": 2718,
+      "sha256": "5883626232453f584bdcc167fc38d7c345727e2f73ea0b3af38532bf35774f4c"
+    },
+    "PROTOCOL.md": {
+      "bytes": 2936,
+      "sha256": "286fdc8db646d61f7f4d1293de2d2f376a9031cf4de6ae7559ae597827d12a70"
+    },
+    "README.md": {
+      "bytes": 9481,
+      "sha256": "d92c25b7e3e0c0d24d5f54c028b42ef2f15d09ae35f7013367b90490bb6a5f7e"
+    },
+    "RECONSTRUCTION.md": {
+      "bytes": 3353,
+      "sha256": "7a5255e8ed12b82caddcf43594465b59391f30e3ced896c284269a9d72f34983"
+    },
+    "RELEASE.json": {
+      "bytes": 3289,
+      "sha256": "97457ebd2b91a8faef9706a6d148dc8ea9fa0cd65b65f8a99a7604ccaca3a2e3"
+    },
+    "UPLOAD.md": {
+      "bytes": 1775,
+      "sha256": "78105a9b7b8353f33cad0ea0bfb4ec8350049deaf5a3a28b11e0c29ea0c7dfa6"
+    },
+    "data/clevr/test-00000-of-00008.parquet": {
+      "bytes": 122029816,
+      "sha256": "e2e57317940534dbc746da2a1d4d722d66b45ad58dad7c556b5c813d433f10e6"
+    },
+    "data/clevr/test-00001-of-00008.parquet": {
+      "bytes": 122560874,
+      "sha256": "058eb964b60f36fc000e757c4f6e0233e0cb2fc13bcc0d6a150b104cc28f31f5"
+    },
+    "data/clevr/test-00002-of-00008.parquet": {
+      "bytes": 122649027,
+      "sha256": "40ae3fe057b5c148aab17f8971da95cae465210a7ed0817ff34d6f01aa41b2dc"
+    },
+    "data/clevr/test-00003-of-00008.parquet": {
+      "bytes": 122427965,
+      "sha256": "4bd8efa25a9d49b6b3e02516c9f99b89819cb5e5fd4098f996cd427dcdb07d58"
+    },
+    "data/clevr/test-00004-of-00008.parquet": {
+      "bytes": 122424211,
+      "sha256": "e54588c2274b4daedea474581c7928b8d5021d18bfe687b90564a8c5f7447373"
+    },
+    "data/clevr/test-00005-of-00008.parquet": {
+      "bytes": 122703602,
+      "sha256": "c4f526760e70022961a2c35af2fcb3dc5e35e208a6fb5d3f1fa0ae2f9c2febd1"
+    },
+    "data/clevr/test-00006-of-00008.parquet": {
+      "bytes": 121692845,
+      "sha256": "00626dd6a34ff6f1b759d5ac84236ed8d2157b9579a3ac76dd749babeaae8ab7"
+    },
+    "data/clevr/test-00007-of-00008.parquet": {
+      "bytes": 122581306,
+      "sha256": "4f7534965d03dce462288fc9015a08e363d7272fed24c4acfcf1d3b5639f3f07"
+    },
+    "data/gqa/test-00000-of-00006.parquet": {
+      "bytes": 189726,
+      "sha256": "5656d0e4ac4199781239cad88f997a20dd6c527b84e8df62974a7193a1259f96"
+    },
+    "data/gqa/test-00001-of-00006.parquet": {
+      "bytes": 190510,
+      "sha256": "e0efdcd8d6046c6b146d19298d44b35a5404f6706599b6ceee625232a0a0804f"
+    },
+    "data/gqa/test-00002-of-00006.parquet": {
+      "bytes": 190609,
+      "sha256": "0c500b4ff9d37192abae70e6771df2f57f92c1c9fd1dcb3855ed7a89f21aa192"
+    },
+    "data/gqa/test-00003-of-00006.parquet": {
+      "bytes": 190660,
+      "sha256": "84cfdfedf3658813113f2b4d013ec26dc45593d76cccd9ea0445e1e2e95c8975"
+    },
+    "data/gqa/test-00004-of-00006.parquet": {
+      "bytes": 189491,
+      "sha256": "6f2c3e8e9717e830270b1969283fd0910e177e751d87e81d4c7f206f03ef6c22"
+    },
+    "data/gqa/test-00005-of-00006.parquet": {
+      "bytes": 190550,
+      "sha256": "25dbbf3042f06949bb928d9e850be6610f2111ab908d9cb2463608fa5286a538"
+    },
+    "data/plotqa/test-00000-of-00010.parquet": {
+      "bytes": 12135769,
+      "sha256": "3749b9bd6090ba503858a82ede85367fd43fb2b883f39b773391c29475bd0d2e"
+    },
+    "data/plotqa/test-00001-of-00010.parquet": {
+      "bytes": 12497883,
+      "sha256": "cd5a526876de16d4049217eaedb72bb72635efb0d0622517b282561eaeaa8c5d"
+    },
+    "data/plotqa/test-00002-of-00010.parquet": {
+      "bytes": 12339728,
+      "sha256": "b31d69de0cc562adc3bb78e226af434c8473b45b57004d667cecce2d53747af1"
+    },
+    "data/plotqa/test-00003-of-00010.parquet": {
+      "bytes": 12136417,
+      "sha256": "fcb24c4921d6a4739ee72f010e7e568996e58c4b73d81b68fb5dedec75f350e1"
+    },
+    "data/plotqa/test-00004-of-00010.parquet": {
+      "bytes": 12141572,
+      "sha256": "1e848d6faa4823c5286c2850d4d893a1df537fe969814d90dc2eb7b667b29137"
+    },
+    "data/plotqa/test-00005-of-00010.parquet": {
+      "bytes": 12123738,
+      "sha256": "df6110e710d613098a2302bd03e2bd79a76af0816644b5ea0ca877779c28a998"
+    },
+    "data/plotqa/test-00006-of-00010.parquet": {
+      "bytes": 12285517,
+      "sha256": "c3f1abd5f33c9f5da99f380c3ed0c657893514ab732f83da689ce12eb019d506"
+    },
+    "data/plotqa/test-00007-of-00010.parquet": {
+      "bytes": 12336336,
+      "sha256": "48441f34fd91e259e81a09ce1d1d1a252c14b4c6407ad0913e14d12b16aa77fc"
+    },
+    "data/plotqa/test-00008-of-00010.parquet": {
+      "bytes": 12261869,
+      "sha256": "ebec199a245671a12a677f3b962c34028e657f7156b3a00845a5f2f5a1ac64b5"
+    },
+    "data/plotqa/test-00009-of-00010.parquet": {
+      "bytes": 12292552,
+      "sha256": "a1fd438190c07e4e7ba74183a8ea94f4420124f1bd734e3caa175ad1cc9d5ef9"
+    },
+    "evidence/clevr/groups.jsonl.gz": {
+      "bytes": 1052294,
+      "sha256": "dc5bcb5bc5d277ae14c9d0769e1f9524837e08efdc0d1cef456c5d7a446b8573"
+    },
+    "evidence/clevr/views.jsonl.gz": {
+      "bytes": 3028987,
+      "sha256": "398afbadef97b334d78bc5febc083887a79d003183a423810e8cac406094f54f"
+    },
+    "evidence/finite/original-protocol.json": {
+      "bytes": 835,
+      "sha256": "c20f1cca5e61947188298de425de671a1c9f0e141101ec49f1218fa4dea2565b"
+    },
+    "evidence/finite/original-results.jsonl.gz": {
+      "bytes": 27018,
+      "sha256": "de5ed6e92dd0a36067807d970f9974e86657e040796bdf73e63db7c8e3dc874b"
+    },
+    "evidence/finite/solver-protocol.json": {
+      "bytes": 2383,
+      "sha256": "4a538c82c0b4ce8d198d7c1ecbac467a7e3c44b5c22ce7d74f6aa89489925733"
+    },
+    "evidence/finite/solver-results.jsonl.gz": {
+      "bytes": 61448,
+      "sha256": "d383d15f970f43f380a568c30015fe9abb21c853e9ca41cdd1db08465e29f24c"
+    },
+    "evidence/gqa/reconstruction.jsonl.gz": {
+      "bytes": 483217,
+      "sha256": "d0ff47c1a4ea8498dfbf5f880158e409f7b8ab312f862b365d1fcd1b28f4c5e4"
+    },
+    "evidence/plotqa/bounded-witness-proofs.jsonl.gz": {
+      "bytes": 24247132,
+      "sha256": "96086574e2684c40d7d718379af3baf308c7d5cc276bce00fb3b6bedb10188ff"
+    },
+    "evidence/plotqa/groups.jsonl.gz": {
+      "bytes": 1243853,
+      "sha256": "130e8cf3d801dd31253720c3a66db9bba2a06e894c09f862437cece291042a22"
+    },
+    "evidence/plotqa/views.jsonl.gz": {
+      "bytes": 3138359,
+      "sha256": "09cb99c9f91e6668fea404e1624ce24e9ad542afc101e655befb8fc481a699bc"
+    },
+    "expected/crossdomain-analysis.json": {
+      "bytes": 637940,
+      "sha256": "3c0aca374e78f1ad00a89119d96466012e11fcb1d3ab0eccc9b9e3039cd00ffb"
+    },
+    "expected/diagnostic-analysis.json": {
+      "bytes": 6790,
+      "sha256": "396b8913d78ac6ac5d367bd52c874d7d350306b08e62834cfc0f6d5b9843191e"
+    },
+    "expected/plotqa-analysis.json": {
+      "bytes": 2335701,
+      "sha256": "ac38a791fbdc758ed5fcda94984bfa58efcdc5da3c512c212e3a07bb5470ccea"
+    },
+    "expected/plotqa-joint.json": {
+      "bytes": 5916,
+      "sha256": "0326de7c7dcbf9602580aba2f76093edbb0aab52888c09d8c069374763dc3ebc"
+    },
+    "expected/revision-analysis.json": {
+      "bytes": 183153,
+      "sha256": "d5029f44630fb298179fa0d2b255e1cd1d9bfb91f9290da61bc8385efe152789"
+    },
+    "expected/state-standardized-selection.json": {
+      "bytes": 2679,
+      "sha256": "6b77e5bc5903fff8e9e02fddd435907854f247176d047accf8f4d97e7a5813ee"
+    },
+    "inputs/crossdomain/GLM-5.3-Flash.jsonl.gz": {
+      "bytes": 224038,
+      "sha256": "78f84095efe2bed3c95c242a085cab1ba0f770f194b54c61a64bb494b7631d6e"
+    },
+    "inputs/crossdomain/gemma4-26b-a4b-it.jsonl.gz": {
+      "bytes": 224816,
+      "sha256": "58d8fcca0d0edbc4fdadb27a2fb3bcc3d3807c5ab485a5c9eacf924c0be0cb33"
+    },
+    "inputs/crossdomain/gemma4-31b-it.jsonl.gz": {
+      "bytes": 195344,
+      "sha256": "c42edcebe71f4d2b36e6cfbd33370f322210317e638390f3a22ba79ea18a4362"
+    },
+    "inputs/crossdomain/group-metadata.json": {
+      "bytes": 617604,
+      "sha256": "166637095f03822aeee87a7cf83f5b312905411448e6ab5a4b8b6c42ef50dc7f"
+    },
+    "inputs/crossdomain/labels.json": {
+      "bytes": 3105097,
+      "sha256": "8c6395de2b2cac74cd8bf767c677808c8480dc83f9181648c04e6353e9c922ec"
+    },
+    "inputs/crossdomain/manifest.jsonl": {
+      "bytes": 1148961,
+      "sha256": "f450580932439af22a88aad4c4f8042e432b30bc5985e700fbfdc569e99fcc38"
+    },
+    "inputs/crossdomain/molmo2-8b.jsonl.gz": {
+      "bytes": 187046,
+      "sha256": "2f217ab5e241303f24928386790ec54c82e3adb902799a722a6d6f8888f2a6df"
+    },
+    "inputs/crossdomain/qwen-backend-provenance.json": {
+      "bytes": 1999,
+      "sha256": "1fe92c3cebe8002acab360b9ef79d93d80dcfe2091990a16066b4d3e0487c053"
+    },
+    "inputs/crossdomain/qwen3.8-27b-fp8.jsonl.gz": {
+      "bytes": 231892,
+      "sha256": "804cdcec607d133642a17f526cf07ec659766dda31418710aa6759ae9ba0f663"
+    },
+    "inputs/crossdomain/qwen38-flash-next.jsonl.gz": {
+      "bytes": 222887,
+      "sha256": "83baecd100633cf252c1caaf8f72c41c56914b0cdcaa5d183d923f5479523993"
+    },
+    "inputs/plotqa/gemma4-26b-a4b-it.jsonl.gz": {
+      "bytes": 195773,
+      "sha256": "02a3176a7829ec8350308c158c729d22491e1fd652ae44055a67e776db3e2a78"
+    },
+    "inputs/plotqa/gemma4-31b-it.jsonl.gz": {
+      "bytes": 180300,
+      "sha256": "af639938d0f8991b6428c0081e46c0527be50ecbdd316cd16a9e1e3c1736fc8d"
+    },
+    "inputs/plotqa/glm.jsonl.gz": {
+      "bytes": 204931,
+      "sha256": "1cb584c58bbffa27d487465158409e4283a72bc0fb96ae8f57ccdfeae11e02ee"
+    },
+    "inputs/plotqa/labels.json": {
+      "bytes": 1594307,
+      "sha256": "8009891a842bc9c0f83fc95a4cad93ccd663504891a519e60877ad57adc63499"
+    },
+    "inputs/plotqa/manifest.jsonl": {
+      "bytes": 511080,
+      "sha256": "0fdb6e4ae27aa995c2efaf9a54f72eaae5b5d23bd43b643c8af42f316c4fedbb"
+    },
+    "inputs/plotqa/molmo2-8b.jsonl.gz": {
+      "bytes": 176868,
+      "sha256": "a13e0fddc7858734c7d94fe1854d7da93151d00490976a78c21b610d6eff7f2c"
+    },
+    "inputs/plotqa/qwen.jsonl.gz": {
+      "bytes": 196544,
+      "sha256": "09c4d72afc0a0f5734534f276bd0fe9839c8195c0ac0feb40dab5a1ed04e1afb"
+    },
+    "inputs/plotqa/qwen38-flash-next.jsonl.gz": {
+      "bytes": 192900,
+      "sha256": "1405017054bb2817234ddce977ed22b6b8789abcca03a62b524a3a8afd20a1cc"
+    },
+    "licenses/Apache-2.0.txt": {
+      "bytes": 11358,
+      "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30"
+    },
+    "licenses/CC-BY-4.0.txt": {
+      "bytes": 17023,
+      "sha256": "d557539df68e771cc1eedcc91d13f70fca930e508d11eedcafa4b15db49e3744"
+    },
+    "licenses/CLEVR-BSD-3-Clause.txt": {
+      "bytes": 1538,
+      "sha256": "6bd73ce90d18fb0a4ba3db636c8d23ad3cfe0dadce050e8564dfc8ed0fcd1ccb"
+    },
+    "licenses/PlotQA-MIT.txt": {
+      "bytes": 1122,
+      "sha256": "a9709c9eb62b807525e8a5778cb97a37e3746bf6d062e8e6fb6d8fd2c0ec2305"
+    },
+    "licenses/license-text-sources.json": {
+      "bytes": 1025,
+      "sha256": "6220810e05502ae7b27d3a539f6e55797501be154f7678cec4943966d9ed65dd"
+    },
+    "licenses/upstream-review.json": {
+      "bytes": 1441,
+      "sha256": "ae34941693d4acdff6668c27a32db53a078ae98e4ba1321185a92f3e0df5df6b"
+    },
+    "metadata/views.jsonl.gz": {
+      "bytes": 1042468,
+      "sha256": "3524a1d0e038ccd3e81fe1cd4a2c6b1cae404f3bde2b3ed18a9d785dd3ea0b50"
+    },
+    "protocol/manuscript.json": {
+      "bytes": 1554,
+      "sha256": "abbe5046b7ba16144ca0775e64b662c0f98fbdd955d6d8bb2d76ad1eb9626172"
+    },
+    "protocol/prompt.json": {
+      "bytes": 1270,
+      "sha256": "4eebce174b3d64e6c2b230b06eb0e9ce04265451b0dc11b087ece1e5e01f69bf"
+    },
+    "protocol/source-resources.json": {
+      "bytes": 5536,
+      "sha256": "fdd58100f8410f00874e40d75d7977ee59504d87b6733ed18efb23d9b04e3d23"
+    },
+    "protocol/verified-runtime.json": {
+      "bytes": 411,
+      "sha256": "43a4e8e34fd2c1f2fad5cb0fc9328b63cde77a5501603a9bbdb3366bd680eb96"
+    },
+    "requirements.txt": {
+      "bytes": 172,
+      "sha256": "c258a9d9c7df6e6952ec937ae432a26f13b01a0f06497e10f7547fee3e238b9b"
+    },
+    "schemas/automated_audit.schema.json": {
+      "bytes": 3579,
+      "sha256": "f5a66de8ef8075b709d577c031487396a1384c58fa2041ac11a40617cf2d7a10"
+    },
+    "schemas/compiled_program.schema.json": {
+      "bytes": 3351,
+      "sha256": "b6605e4c7dedd00a476468ea84f1ef1f30bfa0e99ddfcf5e5b24e3653e0819f3"
+    },
+    "schemas/dependency_proposal.schema.json": {
+      "bytes": 8524,
+      "sha256": "d06a0fcadad5bfbd02559c09258f0757eda0bee9478fc3483b11c6b292155328"
+    },
+    "schemas/intervention_certificate.schema.json": {
+      "bytes": 11478,
+      "sha256": "7e98be332b2fb8cd340c455301641396afd99cb1e32abe0868f3d8909b5bf7cd"
+    },
+    "schemas/intervention_group.schema.json": {
+      "bytes": 8831,
+      "sha256": "49cb548e038cf06f413ccddb80a4f3b20353aa4baa4af663654ab949a42fbe56"
+    },
+    "schemas/normalized_item.schema.json": {
+      "bytes": 2674,
+      "sha256": "c03c98d776b5892ee34750bccd300f73a3a72eff102787e59967ce726bf01816"
+    },
+    "scripts/analyze_crossdomain_paper.py": {
+      "bytes": 16908,
+      "sha256": "fcd3030fbc465dccfa4125faa38479faf771d3224576e31c58a2a5e768a76568"
+    },
+    "scripts/analyze_crossdomain_visual.py": {
+      "bytes": 8675,
+      "sha256": "45df57a48be1c454673ede8d01627a38426839573d39d74cd49163f7126b1822"
+    },
+    "scripts/analyze_visual_judges.py": {
+      "bytes": 15124,
+      "sha256": "ceaddd258c2c1dc142a1e75bbab3bca5fa2a177ea30804e30a8e842cbc5e91cd"
+    },
+    "scripts/build_answerability_rewrite_assets.py": {
+      "bytes": 7788,
+      "sha256": "34f591209f00b3107e74cf99dc5d2137964dc9193839812ff896b8edf8d9fb7c"
+    },
+    "scripts/build_crossdomain_paper_assets.py": {
+      "bytes": 6942,
+      "sha256": "f5229618dd44fd2264bb79aa58116939a69315485ac4f1973d374bea9ac45f1a"
+    },
+    "scripts/freeze_hf_dataset.py": {
+      "bytes": 1995,
+      "sha256": "573c59ea3f34d837085d6c9e7869834695c3d6053141a545e99f88c61b98ae0f"
+    },
+    "scripts/hf_release_common.py": {
+      "bytes": 2781,
+      "sha256": "0bcd2916732d54feaea65c98cbf7d60f3a6a8c897bd82b11a8a56e66d8ba2af2"
+    },
+    "scripts/reconstruct_hf_gqa.py": {
+      "bytes": 6810,
+      "sha256": "1f5df181ff7ae12ac42e20020cb521221721db4e5e3dd1d25aaba68dee79860c"
+    },
+    "scripts/replay_hf_statistics.py": {
+      "bytes": 3977,
+      "sha256": "9efeea3c9232a948dc64eb49f654f964a95d61bb95f5755a3283c2b4bf31cbe3"
+    },
+    "scripts/score_hf_dataset.py": {
+      "bytes": 6095,
+      "sha256": "a7ad86eb48f98bf48774a61ef006b657a4473eeecba6a4fccb5d832f25418e54"
+    },
+    "scripts/verify_argument_evidence.py": {
+      "bytes": 4948,
+      "sha256": "f45edeb4b9bc0c42f69695f0e65196053c0bbe66625b24e13af2c598cedf3706"
+    },
+    "scripts/verify_bounded_witness_bundle.py": {
+      "bytes": 2533,
+      "sha256": "cbc2de3f526c51bdae70aa62b1357fbbfb5f0b2b229a7234110c8c71595b7f28"
+    },
+    "scripts/verify_hf_dataset.py": {
+      "bytes": 8812,
+      "sha256": "a5eee4098dbc8d6f3c5c0c1d0137836e9381777c3a06e1aab6dd737f9e5aad0d"
+    },
+    "src/explicit_learning/__init__.py": {
+      "bytes": 340,
+      "sha256": "c9f2e88116f6cdb36a2438486cba4b24d88f8e37cf9a89bc897595b26af02b5f"
+    },
+    "src/explicit_learning/atomic_io.py": {
+      "bytes": 5399,
+      "sha256": "15733ce8419c1510ad87bffa0b2cd4c4f5f24acff7408054e6ac8d793aa48d4c"
+    },
+    "src/explicit_learning/c2/__init__.py": {
+      "bytes": 1700,
+      "sha256": "85680cc598b67592347a9cc89f93d6353e0cfbafae1ca03c2cacd691f476a7a3"
+    },
+    "src/explicit_learning/c2/client.py": {
+      "bytes": 6683,
+      "sha256": "2764140e0ccfac4bff7699294c2c4a8ba6dd272b565e3e8bf10c7f3ef5bc2c2a"
+    },
+    "src/explicit_learning/c2/compiler.py": {
+      "bytes": 10053,
+      "sha256": "aa7e75f9e12589794608cb87028c2b37838923965459513c4fe7e1cf73cbf895"
+    },
+    "src/explicit_learning/c2/mutations.py": {
+      "bytes": 10101,
+      "sha256": "848eae8df209b7c8f4ce4b67f11fe5927193d2caefe4059550b05674f37a6e09"
+    },
+    "src/explicit_learning/c2/parse.py": {
+      "bytes": 5179,
+      "sha256": "e2ac0dff67d2825cf4b3fdf9349b0767865a5a86b46a078e214eb2e05cd88fa6"
+    },
+    "src/explicit_learning/c2/record.py": {
+      "bytes": 4742,
+      "sha256": "33bdec41ca4b7de402adf07187abe4dc30326b7bfe9c27f005d377de1fb1ef44"
+    },
+    "src/explicit_learning/c2/request.py": {
+      "bytes": 7402,
+      "sha256": "4a5ed22bfa93aacc716003e341fe7c8c6497ff08e20c4a276d3e915b7d237df6"
+    },
+    "src/explicit_learning/c2/state.py": {
+      "bytes": 4031,
+      "sha256": "80a7754b00ccd3ca414dfaf5628f1c40ee862ed416758d50f9dc40c55d59b524"
+    },
+    "src/explicit_learning/certificates/__init__.py": {
+      "bytes": 720,
+      "sha256": "d879c668c92357777b9bdd22ecdfdf4f0986537cb623230211d66ff3c1ccdc99"
+    },
+    "src/explicit_learning/certificates/bounded_plot.py": {
+      "bytes": 6584,
+      "sha256": "1f61330d05e9b16196b2c8b949816029b4c497b60afff4a5d0ad54dd6c49b436"
+    },
+    "src/explicit_learning/certificates/build.py": {
+      "bytes": 11462,
+      "sha256": "cac7d6d071eea5cbefdba651647c03f585dc50fdc8e5c0e08f591314fa4ac353"
+    },
+    "src/explicit_learning/certificates/evi_core.py": {
+      "bytes": 20421,
+      "sha256": "1b46a035ede636f79384cd3d87a00dc25afbf8acad0c2b40d04bcf93a2c8815f"
+    },
+    "src/explicit_learning/certificates/finite_program.py": {
+      "bytes": 8965,
+      "sha256": "6c644aeb3640a3dc6d3bcf9e1c24cefd2b99d0c7e39141498134845fe7090480"
+    },
+    "src/explicit_learning/certificates/plot_observation.py": {
+      "bytes": 2016,
+      "sha256": "6b273e9a1d7197af04cec083104079518f56661e31441a9b3ddcc44ac9013fa4"
+    },
+    "src/explicit_learning/certificates/replay.py": {
+      "bytes": 10325,
+      "sha256": "9ca9b0fa0b1f82b5919991d4a9f316f003d56d0c89414aaced88a17099e8a2cd"
+    },
+    "src/explicit_learning/certificates/smt_program.py": {
+      "bytes": 2465,
+      "sha256": "b0d9660f1702a771b76ef7f4ceb218f50c87598a8ad38af309abba785a0f133b"
+    },
+    "src/explicit_learning/cli/__init__.py": {
+      "bytes": 68,
+      "sha256": "6b7675c2d1927a0038a6c843e0e6456f6cf2ef2e32a8fcfbf0bdacfabb53dfe8"
+    },
+    "src/explicit_learning/cli/_stub.py": {
+      "bytes": 1600,
+      "sha256": "b836b9c89d629a48a96f1431f1feea5ca7a39aa73268c941924e69ef0626a076"
+    },
+    "src/explicit_learning/cli/c2.py": {
+      "bytes": 9799,
+      "sha256": "4b1a63b37e2665f259131b598cdf0c229234a9e02b198d08cbb65c650149823c"
+    },
+    "src/explicit_learning/cli/data.py": {
+      "bytes": 27831,
+      "sha256": "1d287b1b03aaee8ca83f72e271a702c787a15a8fc83c3d88d50a23d4b07a2aa0"
+    },
+    "src/explicit_learning/cli/evaluate.py": {
+      "bytes": 10445,
+      "sha256": "5158342bae8ec9279fa6c08538a19047bad04a357d924a885b70c21e30641643"
+    },
+    "src/explicit_learning/cli/preflight.py": {
+      "bytes": 9646,
+      "sha256": "04f60faf0587178195e0babb5fe0a3a7004e8bdaed91369aabe2edb03425ecc7"
+    },
+    "src/explicit_learning/cli/prepare.py": {
+      "bytes": 29813,
+      "sha256": "a8a8da1ef6f01f8beae4a0570cfd97a50eece2d8986b1de398b542240933baa6"
+    },
+    "src/explicit_learning/cli/resource.py": {
+      "bytes": 7324,
+      "sha256": "1d732af5ad5e1dc84ed74c172940fed2619f3569c37f966aa2da7b4099f5eac0"
+    },
+    "src/explicit_learning/cli/smoke.py": {
+      "bytes": 3076,
+      "sha256": "9f9032275faa11f402ac911185d21638e52d75a3f9ffad91287994f991f0b7c9"
+    },
+    "src/explicit_learning/cli/train.py": {
+      "bytes": 57306,
+      "sha256": "b24b99510f465ffc128e4165d2eb209208f3ff1e25dedb70cc1029fb04f7521d"
+    },
+    "src/explicit_learning/config.py": {
+      "bytes": 14688,
+      "sha256": "3e275403fd6ffea71c37b3754efd2bd79cdc06e31dcc69382ac9f9dd095ec5ce"
+    },
+    "src/explicit_learning/contamination/__init__.py": {
+      "bytes": 1172,
+      "sha256": "c077eac71f7d5f04f644ded768a24e076a89a019f03eed3e3d4eff8113aa0b6f"
+    },
+    "src/explicit_learning/contamination/fingerprints.py": {
+      "bytes": 9384,
+      "sha256": "1052a9fc0619fb3682526ed2303d14678a720cc5b3ae00ea5266d9517e5688b2"
+    },
+    "src/explicit_learning/contamination/index.py": {
+      "bytes": 8484,
+      "sha256": "5584fbea26357b61fde7b6056a67fadb41753ed04c67e95163ca35ceeb65e701"
+    },
+    "src/explicit_learning/contamination/match.py": {
+      "bytes": 10840,
+      "sha256": "0e531a45020add40402259934eefcb4445087eaa63a23cdb490e85775fff32cd"
+    },
+    "src/explicit_learning/datasets/__init__.py": {
+      "bytes": 1199,
+      "sha256": "4123225105e86bf9752771911e940dbd5bbdf7956a16d0884cb4e05c8c6d0b18"
+    },
+    "src/explicit_learning/datasets/clevr.py": {
+      "bytes": 12549,
+      "sha256": "d7f23aa223524af414438c5ae802233c557db44483889b0308dd63fdbc5d1d81"
+    },
+    "src/explicit_learning/datasets/evi_core_release.py": {
+      "bytes": 86451,
+      "sha256": "733541f391b382d5f601615b8ec62b00a04c55b956b2c9d4f7a0e3ab28ffbab0"
+    },
+    "src/explicit_learning/datasets/gqa.py": {
+      "bytes": 13601,
+      "sha256": "4197942614f700190a0929581d0d5bdd52ba0cd755444650d4f4a7e954561238"
+    },
+    "src/explicit_learning/datasets/plotqa_prefix.py": {
+      "bytes": 16227,
+      "sha256": "6c20f64bf83e65cf29ebfaa074088af69e2c09844951802792184424ad5d0b59"
+    },
+    "src/explicit_learning/datasets/release.py": {
+      "bytes": 37980,
+      "sha256": "416cdfed1343e2ee3d3431bfa3ab915c575411b90544ef1c0becc7f7effc94bf"
+    },
+    "src/explicit_learning/dsl/__init__.py": {
+      "bytes": 1238,
+      "sha256": "f33cea841746c38acae5849cf92a5b9103644ce36206d53ba67c1ae5cf43ff78"
+    },
+    "src/explicit_learning/dsl/ast.py": {
+      "bytes": 12096,
+      "sha256": "b120babec28aea901247b7977f3024b20bcddd2aed0e6ab22b911cc56d85c2ae"
+    },
+    "src/explicit_learning/dsl/canonicalize.py": {
+      "bytes": 10363,
+      "sha256": "30e9163cedd039851e05b3b29510e460f853ad627673c42c8f79c6c38f8d6385"
+    },
+    "src/explicit_learning/dsl/clevrdsl.py": {
+      "bytes": 8785,
+      "sha256": "e4930a249903f590c63e9f7cfec63ac936969da6189cf4b232774114a145c168"
+    },
+    "src/explicit_learning/dsl/geometrydsl.py": {
+      "bytes": 7970,
+      "sha256": "693809060bf648b373aebccc9847daaccb32d4e23d98b7e94c9e4ce721a8a6b3"
+    },
+    "src/explicit_learning/dsl/plotdsl.py": {
+      "bytes": 7370,
+      "sha256": "7db4058362bd5fd5fdfcad0a35ed70fbc55a351a9ee226749945c15178736b91"
+    },
+    "src/explicit_learning/dsl/tabledsl.py": {
+      "bytes": 2422,
+      "sha256": "66d206aabfa119217223d45273bcc3da539e5142b53ba8d5ccd4b7bf408b0525"
+    },
+    "src/explicit_learning/dsl/typecheck.py": {
+      "bytes": 7212,
+      "sha256": "4f0bc9b71c54e26e5d2515b43d0e929e7c95b9fa95a835cd1a789f587a00844a"
+    },
+    "src/explicit_learning/evaluation/__init__.py": {
+      "bytes": 570,
+      "sha256": "8d8822e342bbde485907fce0c9b43db2fb4b851dfba4f6b67b2534d61f495341"
+    },
+    "src/explicit_learning/evaluation/core.py": {
+      "bytes": 23525,
+      "sha256": "a9988819f05b7f061a8498f99650f13691fb9fc59f2f61096947720bd20fa7dc"
+    },
+    "src/explicit_learning/evaluation/inference.py": {
+      "bytes": 12914,
+      "sha256": "22772f4f058b6086486f0532157846e5e68de6d5dbe12529505d662e9320eb69"
+    },
+    "src/explicit_learning/evaluation/retention_hf.py": {
+      "bytes": 10871,
+      "sha256": "ef9e5c016ccfc7f2004a7c754587ea1f5aa8b5f5af61bd5837e6a95ef41641f9"
+    },
+    "src/explicit_learning/evaluation/statistics.py": {
+      "bytes": 5565,
+      "sha256": "9752995ca2f8ed975cda0479dc4e61b1174a3d3bd3b977e28ccc539ab759298d"
+    },
+    "src/explicit_learning/executors/__init__.py": {
+      "bytes": 1926,
+      "sha256": "35832987a16defba13ba247c063f1b8c9ea079ea2bf0c8870cb6378f160356d2"
+    },
+    "src/explicit_learning/executors/base.py": {
+      "bytes": 9733,
+      "sha256": "ba390f54beb8e5dcc2bcd9b20747b79ab39321e6334f4b254397c17c2c4682cb"
+    },
+    "src/explicit_learning/executors/clevr.py": {
+      "bytes": 13867,
+      "sha256": "0c2c6efb2648bd73a2c6e36812211ff25c7b60d46a4901d7643e49fe11198be9"
+    },
+    "src/explicit_learning/executors/geometry.py": {
+      "bytes": 10822,
+      "sha256": "1df8c9df1f12112432b5a6ed066e4763072abdbabb0bb63ef1ebaba44514f010"
+    },
+    "src/explicit_learning/executors/gqa.py": {
+      "bytes": 30439,
+      "sha256": "84a0c13db5d64363271191d8cbe801dce1c8c229ebba3ec6dbdf3425a56e6a35"
+    },
+    "src/explicit_learning/executors/plot.py": {
+      "bytes": 12523,
+      "sha256": "23af23d95b20d61ea58f43bcf0668bc27ad3cd8c1a6296baaea9b5fdcc9b6d2e"
+    },
+    "src/explicit_learning/executors/reference_plot.py": {
+      "bytes": 4101,
+      "sha256": "7c88d08971780e9c8440928b583366421632550b667ebf46407d2654e086cf61"
+    },
+    "src/explicit_learning/executors/table.py": {
+      "bytes": 7252,
+      "sha256": "df1629db1f0bf61384478b5a908feb7fb01e24457a0e24227cbc6887c940271f"
+    },
+    "src/explicit_learning/hashing.py": {
+      "bytes": 5776,
+      "sha256": "5d01ec2fda7212cb369caf7ab17e4bf9cdc758be6c22f7eae1df4410b4380708"
+    },
+    "src/explicit_learning/ingest/__init__.py": {
+      "bytes": 17483,
+      "sha256": "63a4609d4130b87088111221263e740eb21335bf430eb56df35de6a8960676a9"
+    },
+    "src/explicit_learning/ingest/base.py": {
+      "bytes": 17334,
+      "sha256": "1e1a9e6e7eb994d31c966a98b3c788956bbc4dc874034c47b1bad7decb0f440d"
+    },
+    "src/explicit_learning/ingest/bbox_docvqa.py": {
+      "bytes": 3024,
+      "sha256": "b3be97e1aa184efa32f14bd610b686d216759e690b21ac468fe66ee327cc21cd"
+    },
+    "src/explicit_learning/ingest/chartqa.py": {
+      "bytes": 2615,
+      "sha256": "3409f6de725d86698d0bb1a31a4a4b70c413cb3427172ac732237d09987999db"
+    },
+    "src/explicit_learning/ingest/eval_sources.py": {
+      "bytes": 5523,
+      "sha256": "b2d7f321c6ea778d3a5f0c9d1fa1f0040b8b55ed59b449c1cf51b67c61d13bf4"
+    },
+    "src/explicit_learning/ingest/mmk12.py": {
+      "bytes": 2379,
+      "sha256": "a5c570565c61965aad82345afdedb9c3526c693909d69b50cf899129df449ac2"
+    },
+    "src/explicit_learning/ingest/mmmu.py": {
+      "bytes": 2393,
+      "sha256": "06aaab4339ca5b1d4100f5c35b86740f7b48c2bdba58a8b913c3bc49e75b8e60"
+    },
+    "src/explicit_learning/ingest/mmr1.py": {
+      "bytes": 2563,
+      "sha256": "25bb82dab4a43a21360471e0b12d8a0c77c6836154a72199ff9ebd19be04cf57"
+    },
+    "src/explicit_learning/interventions/__init__.py": {
+      "bytes": 4088,
+      "sha256": "19288008abcd39772b96d1336db5c04368495c1bf7446f239b882fdf4259bcc3"
+    },
+    "src/explicit_learning/interventions/base.py": {
+      "bytes": 12798,
+      "sha256": "bd48ddc1dcded045940009ac885908c555d7e105014651c5d9aaed5cfd339881"
+    },
+    "src/explicit_learning/interventions/clean_delete.py": {
+      "bytes": 4002,
+      "sha256": "ab7592dd88b1f506721d8b26567be961825de6914394abd6d6f6b43097742464"
+    },
+    "src/explicit_learning/interventions/clevr.py": {
+      "bytes": 5893,
+      "sha256": "ba8adbc0e52c6245d68733bd86110b7b4d6adc7245c7114d66bc1f598306817c"
+    },
+    "src/explicit_learning/interventions/completion.py": {
+      "bytes": 7176,
+      "sha256": "76d9e95f6375f089ee4ff1e75fab2813deb4430d0f56c8048d0672f645a4b8f8"
+    },
+    "src/explicit_learning/interventions/control.py": {
+      "bytes": 7491,
+      "sha256": "467f349fbfd8d3c3154f498e8c47c59ee9d48cba5e47a58a390ecfe7cff7ccf4"
+    },
+    "src/explicit_learning/interventions/feasibility.py": {
+      "bytes": 5380,
+      "sha256": "8d22a750a02241caf7df4a47cde62c8300414e8142bfaabfaa56ad13c6806739"
+    },
+    "src/explicit_learning/interventions/redact_solid.py": {
+      "bytes": 2055,
+      "sha256": "21769383187c7bd1e38e42c086fb10fe9c9a5b1ba02aed64188b9a641b17bff2"
+    },
+    "src/explicit_learning/interventions/substitute.py": {
+      "bytes": 11331,
+      "sha256": "462abcc340d216c3e777076655c8602fc003f78e452f45cb6bfa14f9b9b0a512"
+    },
+    "src/explicit_learning/logging_setup.py": {
+      "bytes": 3260,
+      "sha256": "9d46eb38fcfb1176c11aa92482d655120f80f8ed0bc80ef3eeb3e5827d5b2e95"
+    },
+    "src/explicit_learning/manifests.py": {
+      "bytes": 6492,
+      "sha256": "798544f845bd0b85f85a90dd492cbaf3f1a16124712d6148644c0ca66ff50401"
+    },
+    "src/explicit_learning/paths.py": {
+      "bytes": 2521,
+      "sha256": "c2a07dd38a5f6777d6be7a779e349f82bb46cc6f8eb73c8460c12b1a7b1c625e"
+    },
+    "src/explicit_learning/py.typed": {
+      "bytes": 0,
+      "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
+    },
+    "src/explicit_learning/renderers/__init__.py": {
+      "bytes": 4462,
+      "sha256": "4485d1c301f727e4ba028c6945fcd2d7f1aa36fbe6a88d96004c1f68bd904c12"
+    },
+    "src/explicit_learning/renderers/base.py": {
+      "bytes": 18253,
+      "sha256": "b001ea674a34f45e95afd1e18bfca9a70416f462001e4a0eb4945d358224ae0c"
+    },
+    "src/explicit_learning/renderers/clevr.py": {
+      "bytes": 13595,
+      "sha256": "ac883be6bff411d463de9c9373232a11c7e7d7103375eae50735dcff7ae51c24"
+    },
+    "src/explicit_learning/renderers/clevr_blender.py": {
+      "bytes": 14382,
+      "sha256": "679f14662c56e54004bc001f2a05b37856bea5dc05a8e3a702d0d7d6de14f8d6"
+    },
+    "src/explicit_learning/renderers/geometry.py": {
+      "bytes": 21983,
+      "sha256": "25b28dfca998564e159963cebc6d0d759ee9ba38ae159487519abb0666af4b96"
+    },
+    "src/explicit_learning/renderers/gqa.py": {
+      "bytes": 9885,
+      "sha256": "442aa0680ffaec473f459b059c5af02b97bdc4cee4cd54537498ab237e09c6cb"
+    },
+    "src/explicit_learning/renderers/node_map.py": {
+      "bytes": 1640,
+      "sha256": "f30baeedbca6ae9278309f638ddf1622bf39ed0b59ce025224c5f2740f715041"
+    },
+    "src/explicit_learning/renderers/plot.py": {
+      "bytes": 40126,
+      "sha256": "5b45dd8fc5c35f1926fa485546c443446641ea6eaed178a500247ab1c0a09414"
+    },
+    "src/explicit_learning/renderers/text_manifest.py": {
+      "bytes": 931,
+      "sha256": "a5958e6ac40fe880750f3edbf404b4b1676825c5471265a6a74c6f2db4da77e6"
+    },
+    "src/explicit_learning/resources.py": {
+      "bytes": 12488,
+      "sha256": "88ae0bed875fae12160cff09c2a4cbbc59bebc91cddd536ff668db8843eb6faa"
+    },
+    "src/explicit_learning/schema_io.py": {
+      "bytes": 3790,
+      "sha256": "33cd48255563cb4b224cb364330288891a1dbcb8eb31f18632d104c838c8d04d"
+    },
+    "src/explicit_learning/sources/__init__.py": {
+      "bytes": 1409,
+      "sha256": "37af64e9e5bf1666e8d76b40d626877ba4f6aa9ee4296cd302c5c75b66236a0e"
+    },
+    "src/explicit_learning/sources/base.py": {
+      "bytes": 10147,
+      "sha256": "954cdd6a5d31a7ed903b2cc00567574726201b3885aff6f9f242a54efdaa8f6c"
+    },
+    "src/explicit_learning/sources/clevr.py": {
+      "bytes": 17089,
+      "sha256": "dd2134cecad32c8cbfeb9951d38406c2c2edb7484b597cf2b060f147252d9d7f"
+    },
+    "src/explicit_learning/sources/geometry3k.py": {
+      "bytes": 15103,
+      "sha256": "0442b07af4605f5308b547fe4f6774387fbea717aa25ebcb8ebe51a98063e4eb"
+    },
+    "src/explicit_learning/sources/gqa.py": {
+      "bytes": 17601,
+      "sha256": "d13f4e63b7b2e8c3ebd9c58200ebe7097e52200e9fb267c3a6af15d2b81be5f4"
+    },
+    "src/explicit_learning/sources/plotqa.py": {
+      "bytes": 16525,
+      "sha256": "2d43c73f29848fa5c50e850effc354056ccd518def64d7839aac9b1d14d8524b"
+    },
+    "src/explicit_learning/sources/world_ops.py": {
+      "bytes": 5795,
+      "sha256": "73ce9ff8052aafdcde0c525c8c0b89c841adbcf6a9f86a7ffe9c40541268461c"
+    },
+    "src/explicit_learning/training/__init__.py": {
+      "bytes": 1702,
+      "sha256": "b4dc76a4ee187af070d7a94424ac626dcbc8793be47132e95c505f7534fb6228"
+    },
+    "src/explicit_learning/training/aligned_grpo.py": {
+      "bytes": 10724,
+      "sha256": "156a40c887bca9c2c6017d7e0a9d019208bdce28474bb5316c4154531bc669de"
+    },
+    "src/explicit_learning/training/answers.py": {
+      "bytes": 9595,
+      "sha256": "b2b1e2d50344d98b19eb53b598837eb9755aeec6b5a4075cdf44aa63972ca135"
+    },
+    "src/explicit_learning/training/artifacts.py": {
+      "bytes": 21776,
+      "sha256": "a1c81c53d593e7809be1c7db3a400ac36dfc87f641c8979d29dcb8e74b1c49b8"
+    },
+    "src/explicit_learning/training/backend.py": {
+      "bytes": 87432,
+      "sha256": "5f36fe39fd27fafe3849dfa5c8a7f3c2b86b093c8a85c366de1f7c3f13d067c9"
+    },
+    "src/explicit_learning/training/data.py": {
+      "bytes": 22210,
+      "sha256": "3f168ec85ccc5493ebd335d7b96b0379e1a49bfcdadbb4de3a9e993b2075b3e0"
+    },
+    "src/explicit_learning/training/environment.py": {
+      "bytes": 4033,
+      "sha256": "e86c1e90ba29729f9c6d6339f9ab466dcdfd6ab9e9037f6ba19588e8f281ee96"
+    },
+    "src/explicit_learning/training/evi_po.py": {
+      "bytes": 40802,
+      "sha256": "62f25db078e3fcf57afbb769cbfb5f2ed934ce6cd54ed47003619033f73e6412"
+    },
+    "src/explicit_learning/training/evi_po_contract.py": {
+      "bytes": 17774,
+      "sha256": "fcab73b870fb0492b3aae584518b6fde47da19e3eb6cff44d36cb027fddbbd72"
+    },
+    "src/explicit_learning/training/launcher.py": {
+      "bytes": 17723,
+      "sha256": "7c6c7aae59a8ad0ec27a4d00f03f51a206dc1fdcd98e9ca0b784b314cf4d97db"
+    },
+    "src/explicit_learning/training/lazy_dataset.py": {
+      "bytes": 5328,
+      "sha256": "f141c30f5f5683e00e2fef4084e1c37f24b657e36723072bbe57e739411f3d03"
+    },
+    "src/explicit_learning/training/ledger.py": {
+      "bytes": 11316,
+      "sha256": "46e1c6ec7a35b7da6e90dd23128c28f73a508c09c264846e37bf61849eac16c4"
+    },
+    "src/explicit_learning/training/offline.py": {
+      "bytes": 6099,
+      "sha256": "bc4054d1a6b4a4df89081ca6bf082f877878377a78abd88049e28c318479f91c"
+    },
+    "src/explicit_learning/training/papo.py": {
+      "bytes": 14436,
+      "sha256": "9bc5eab27b2b12d61ef62249c4574be67194b0a09f0358ec64b177bce36ac4bc"
+    },
+    "src/explicit_learning/training/papo_contract.py": {
+      "bytes": 4897,
+      "sha256": "abede0c72f8f0aefc75b7c4789ffe1c800608c0ff3b044db9025f1a120490684"
+    },
+    "src/explicit_learning/training/peft_contract.py": {
+      "bytes": 4784,
+      "sha256": "33c52e4e41a96328de948d22570b6d445f7526a3f497b653583ac0f19c704033"
+    },
+    "src/explicit_learning/training/preflight.py": {
+      "bytes": 14886,
+      "sha256": "040d0beec11deef03457d2c0453130d0deb145f5ed46520543d892be5e4301fa"
+    },
+    "src/explicit_learning/training/rewards.py": {
+      "bytes": 13150,
+      "sha256": "acda00fb88e54df06798e12635dff76f4bce3458b9517b867fdd8ff1c2e31d08"
+    },
+    "src/explicit_learning/training/sft_schedule.py": {
+      "bytes": 10807,
+      "sha256": "2a739d25f6785d4c4e5f44da9467bdd77a102b068d0c73a1b28961dfe672aed3"
+    },
+    "src/explicit_learning/training/slots.py": {
+      "bytes": 7370,
+      "sha256": "b90a091e63e67381ff560b525ba86833066cb37309044abb80cf948ea0027cb9"
+    },
+    "src/explicit_learning/training/smoke_gate.py": {
+      "bytes": 35942,
+      "sha256": "2c73e2c907265c605e0faa40d5e7780cf75d2d559a639417675530acf3306544"
+    },
+    "src/explicit_learning/training/targets.py": {
+      "bytes": 5067,
+      "sha256": "395049d3ba9458e75e4898cb6ec05b40a4e394e46d7ddb7468d55852348bf1d8"
+    },
+    "src/explicit_learning/training/vllm_runtime.py": {
+      "bytes": 6794,
+      "sha256": "466e7bb6a2d4c4154a763c745bdf99ce7f5473e2838a12d35da38b2e9c8a6c35"
+    },
+    "src/explicit_learning/vcs.py": {
+      "bytes": 927,
+      "sha256": "781537162eced23bb6c19feeddfa605c9a75c1d45387fc77dc46272225418846"
+    },
+    "validation/dataset-validation.json": {
+      "bytes": 539,
+      "sha256": "52e6ae623727d31cef11daf97b9483b80bd760b95d0ba5a259d3e3e01163efba"
+    },
+    "validation/gqa-reconstruction-validation.json": {
+      "bytes": 313,
+      "sha256": "9f071fb9066a5e9a4f8d5404a7cbdb99e7723640dd6839eb1101d1a6bc050b7a"
+    },
+    "validation/manuscript-table-comparison.json": {
+      "bytes": 268,
+      "sha256": "b1eb520ed20cb2d94fe8c2328da3c29acf01426951f62680c096e7d8dca8d962"
+    },
+    "validation/proof-build.json": {
+      "bytes": 310,
+      "sha256": "43bd015412ab24eee69b05b7b16d21d18805182182417c1cd08b1c9e90e8afb4"
+    },
+    "validation/publication-boundary-audit.json": {
+      "bytes": 218,
+      "sha256": "d6af1989ded233425bf6165fd742f60eaf4338be043df729ada4dd6c32b1c0d9"
+    },
+    "validation/reconstruction-loader.json": {
+      "bytes": 169,
+      "sha256": "3e4c4bc9e96651b2d60f2b2e85f6a00c09e6fa2a9ff3d36731ffe11eafa36858"
+    },
+    "validation/replay-validation.json": {
+      "bytes": 401,
+      "sha256": "2deefe91918dd1e859c1503e916021e38d213e4045da2cae42c43bd1858e521a"
+    },
+    "validation/scoring-cli.json": {
+      "bytes": 176,
+      "sha256": "4bc09c55f8db2d0139be53d4e7810ea4b4e10f0689ccac5d36432ad21621ea4d"
+    },
+    "validation/supplementary-statistics.json": {
+      "bytes": 924,
+      "sha256": "83512314a71c3dafd21bb13ca9fada0ac2a46bbf510face0eb6a1bc7c2c273bc"
+    },
+    "validation/test-summary.json": {
+      "bytes": 299,
+      "sha256": "4386377bc1b92d766fdb3257146972a6d625b11de0029adb7bb2c5f4526ba1ce"
+    }
+  },
+  "total_bytes": 1151312498,
+  "repository": "sungguk/visual-answerability",
+  "release_tag": "v1.0.0",
+  "prepared_manifest_sha256": "3959fb0365bcf14c55e01123cca90a5bce1744407e70b80511192b3842139f03"
+}

PROTOCOL.md

@@ -0,0 +1,50 @@
+# Frozen model protocol and environment
+
+`protocol/prompt.json` holds the exact user prompt template and requested JSON
+schema. Present one image and the unchanged source question independently;
+group states only after inference. All six reported configurations used
+temperature 0 and a 512-token output cap through vLLM-based serving. The saved
+analysis treats them as serving configurations; it does not equate API and
+local checkpoint revisions that lack matching provenance.
+
+| Scoring key | Paper name | Model identifier |
+|---|---|---|
+| `qwen38-flash-next` | Qwen 3.8 Flash Next | `Qwen/Qwen3.8-Flash-Next` |
+| `gemma4-31b-it` | Gemma 4 31B | `google/gemma-4-31B-it` |
+| `qwen` | Qwen 3.8 27B | `Qwen/Qwen3.8-27B-FP8` |
+| `gemma4-26b-a4b-it` | Gemma 4 26B A4B | `google/gemma-4-26B-A4B-it` |
+| `glm` | GLM 5.3 Flash | `GLM-5.3-Flash` |
+| `molmo2-8b` | Molmo2-8B | `allenai/Molmo2-8B` |
+
+In `inputs/crossdomain/`, `qwen` is named `qwen3.8-27b-fp8` and `glm` is named
+`GLM-5.3-Flash`. Each configuration contributes 5,000 chart and 7,000 scene
+responses. The 11 terminal invalids comprise eight chart outputs from Molmo2
+and one CLEVR output each from Qwen Next, Qwen 27B and Gemma 26B. All GQA final
+responses are schema-valid.
+
+Qwen 27B scene outputs combine 2,425 preserved bridge responses and 4,575 local
+FP8 fallback responses. The local fallback's model revision and serving settings
+are in `inputs/crossdomain/qwen-backend-provenance.json`. The bridge checkpoint
+revision is unknown. Mixed-backend question groups are retained and reported;
+do not interpret a backend difference as a causal model/domain effect.
+
+The verified **CPU reproduction** environment is pinned in `requirements.txt`
+and `protocol/verified-runtime.json`. It reproduces statistics and validates
+the packaged evidence; it is not claimed to be every model's inference
+environment. Available historical inference revisions/versions and decoding
+settings are recorded in `INFERENCE_ENVIRONMENTS.json`; unknown
+values stay unknown. No model weights, API credentials or generation traces
+are included, and preparing this release ran no new inference.
+
+For future GPU replication, start with a small preflight and record GPU memory,
+utilization and throughput. The original local runners used vLLM where supported,
+`gpu_memory_utilization=0.90`, greedy decoding, and model-native image processing.
+Batching/parallelism and backend-specific compatibility affect the runtime;
+the frozen results can be reproduced from the supplied scoring records without
+access to any original endpoint.
+
+Strict scene answers normalize rational numbers, Unicode NFKC, case and spaces
+only. Invalid outputs fail full-denominator scores and all-sibling group success.
+All post-hoc strata, class-balanced metrics and bootstrap settings retain their
+descriptive status. The finite-program construction records are independent of
+the six-configuration model response denominator.

README.md

@@ -0,0 +1,194 @@
+---
+pretty_name: Visual Answerability across Charts and Scenes
+language:
+- en
+license: other
+license_name: component-specific-see-license
+license_link: LICENSE.md
+task_categories:
+- visual-question-answering
+size_categories:
+- 10K<n<100K
+tags:
+- visual-answerability
+- abstention
+- executable-certificates
+- plotqa
+- clevr
+- gqa
+configs:
+- config_name: plotqa
+  default: true
+  data_files:
+  - split: test
+    path: data/plotqa/*.parquet
+- config_name: clevr
+  data_files:
+  - split: test
+    path: data/clevr/*.parquet
+- config_name: gqa
+  data_files:
+  - split: test
+    path: data/gqa/*.parquet
+---
+
+# Visual Answerability across Charts and Scenes
+
+Frozen release: **answerability-v1.0.0-20260928**, Hub tag **v1.0.0**.
+Dataset: [sungguk/visual-answerability](https://huggingface.co/datasets/sungguk/visual-answerability).
+This benchmark accompanies *When Does an Image Determine the Answer?
+Benchmarking Visual Answerability across Charts and Scenes*, using the
+September 28, 2026 manuscript metadata and the unchanged evaluated cohorts.
+No arXiv identifier or DOI has been assigned in this package.
+
+Each source question has several original/edited views. Present **one image and
+one question independently** to a model; group responses only after inference.
+The main task requires correct supported answers and correct abstentions for
+every view of a question. Do not expose `state`, `target`, `answerable`, sibling
+views, or evidence files to the model.
+
+## Contents and access
+
+| Configuration | Groups / views | Image access |
+|---|---|---|
+| `plotqa` | 1,000 / 5,000 | Embedded PNG |
+| `clevr` | 1,000 / 4,000 | Embedded PNG |
+| `gqa` | 1,000 / 3,000 | Local reconstruction |
+
+All three configurations use an evaluation-only `test` split. Source splits
+remain explicit: PlotQA comes from the frozen train-prefix selection; CLEVR and
+GQA come from source validation data. This naming does not imply use of an
+official upstream hidden test split. This release covers the **12,000 evaluated
+views**, not the broader 58,000-view training/development resource.
+
+The chart cohort uses executable D15 direct-lookup questions from the frozen
+PlotQA prefix. CLEVR uses source functional programs; GQA uses the supported
+native semantic-program subset. There is one source image per question group
+within each domain.
+
+The `gqa` configuration supplies IDs, our labels and edit recipes. Its `image`
+and `question` fields are null. The official GQA pages and archive READMEs checked
+for this preparation did not supply an explicit annotation redistribution grant.
+The earlier local CC BY declaration is not treated as upstream permission.
+Obtain the official questions, graphs and images yourself, then follow
+[RECONSTRUCTION.md](RECONSTRUCTION.md). The reconstruction checks source identity,
+programs, labels and exact evaluated image bytes. It writes outside this folder.
+
+PlotQA and CLEVR data retain CC BY 4.0; bundled project code retains Apache 2.0.
+See [LICENSE.md](LICENSE.md) and [ATTRIBUTION.md](ATTRIBUTION.md) for component
+boundaries, upstream authors, modifications and notices.
+
+## Load the dataset
+
+```bash
+python -m pip install datasets pillow
+```
+
+```python
+from datasets import load_dataset
+
+root = "sungguk/visual-answerability"
+charts = load_dataset(root, "plotqa", split="test", revision="v1.0.0")
+scenes = load_dataset(root, "clevr", split="test", revision="v1.0.0")
+gqa_index = load_dataset(root, "gqa", split="test", revision="v1.0.0")
+example = charts[0]
+example["image"].save("example.png")
+print(example["question"])
+```
+
+For the evidence, scoring scripts and pinned dependencies, download the full
+snapshot as described in [UPLOAD.md](UPLOAD.md). The `v1.0.0` tag identifies the
+verified release commit; use that commit hash when an immutable revision is
+required. Local snapshot paths can also be passed to `load_dataset`.
+
+## Fields and states
+
+`item_id` is the stable prediction join key (`item-0001`… or `cross-00000`…).
+`group_id` joins sibling states. `source_question_id`, `source_image_id`,
+`source_split`, and `source_revision` preserve upstream identity.
+`question_sha256` and `image_sha256` validate reconstruction and release
+integrity; hashes are not the semantic proof of a label. `image` is a Hugging
+Face Image feature backed by embedded bytes, except for GQA. `target` is a
+string, with `<UNANSWERABLE>` for unanswerable states. `dependency_depth` is the
+historical field name; for scene programs it counts steps, not tree depth.
+
+- `FULL`: the original supported question.
+- `A_SAME`: a control edit preserves the supported answer.
+- `A_CHANGED`: an edit changes the supported answer (charts and CLEVR).
+- `U_MISSING`: the construction removes evidence needed for a unique answer.
+- `U_INVALID`: an edit invalidates the chart question's referent (charts only).
+
+`evidence_kind`, `evidence_file`, `operator`, `source_view_id`, and
+`gqa_location_stratum` describe the construction and its evidence. The location
+flag identifies 376 GQA groups for the paper's sensitivity analysis; it is not
+a finding that those labels are wrong or that the remaining 624 groups are clean.
+
+## Evidence and evaluation
+
+- **Chart proofs:** `evidence/plotqa/bounded-witness-proofs.jsonl.gz` contains
+  all 1,000 corrected bounded witnesses, reconstructed from frozen FULL and
+  A_CHANGED worlds and verified against each evaluated missing-view PNG.
+  The checker re-renders complete worlds, verifies axis admissibility and two
+  executors, then checks identical intervened pixels with different answers.
+  These are freshly serialized proofs, not a claim of byte identity with an
+  unavailable historical proof archive.
+  Original per-view certificate descriptors are retained as historical
+  construction metadata. The corrected chart missing-information claim is
+  verified from the separate bounded-witness records above.
+- **Scene evidence:** CLEVR programs, worlds, alternatives and render/edit
+  metadata are in `evidence/clevr/`. GQA source evidence is reconstructed from
+  official inputs using `evidence/gqa/reconstruction.jsonl.gz`. Neither scene
+  subset is presented as having the chart complete-world pixel guarantee.
+- **Scoring records:** `inputs/` contains six configurations' 72,000 selected
+  final scoring projections, including 11 terminal invalid outputs. These are
+  parsed scoring fields, not raw generations, token traces or fresh inference.
+  GQA question text is replaced by stable question-ID references; its rationale
+  text is omitted. Scoring and all reported numerical results are preserved.
+  Image paths inside the historical scoring manifests and group descriptors
+  describe the original construction layout; use the Parquet `image` column
+  and stable `item_id` for public model inputs.
+- **Statistics:** exact replay of all 18 model/domain cells and 10,000-draw
+  paired group bootstrap intervals; also saved finite-program records for
+  1,024 original and 1,456 combined solver-study cases. Replaying these records
+  is separate from rerunning the solver constructors.
+
+See [EVALUATION.md](EVALUATION.md) for new predictions and replay commands,
+[PROTOCOL.md](PROTOCOL.md) for prompts and serving configurations, and
+[UPLOAD.md](UPLOAD.md) for snapshot download and release verification.
+
+## Intended use and limitations
+
+Use for evaluation of image-conditioned answers and abstention, construction
+audits, and reproduction of reported analyses. Do not train or tune on this
+evaluation split when reporting its benchmark results. Questions were selected
+by deterministic executable-program and rendering gates, not representative
+sampling of all visual questions. New per-example edits/labels used no human
+adjudication or reasoning-model judgments.
+
+The chart claim assumes correct source truth, question/program compilation,
+the fixed-axis rational-valued world class, and the drawing primitives. The
+two executors are independent implementations; they share rendering code.
+CLEVR labels use source programs and edits. GQA graphs and object boxes may be
+incomplete; position questions can retain visible cues. Report GQA scores as
+agreement with these source-derived labels, including both sensitivity strata.
+There is no independent human adjudication of every view. Configuration
+comparisons include different serving backends; Qwen 27B scenes mix preserved
+API outputs and a local FP8 fallback. Intervals are descriptive, paired by
+question, and not corrected for multiple comparisons.
+
+Upstream photographs may depict people and everyday places. The release adds
+no identity annotation or sensitive-attribute classification. All original
+rights and any third-party rights remain with their respective holders.
+
+## Citation and provenance
+
+Use [CITATION.cff](CITATION.cff) or [CITATION.bib](CITATION.bib), and cite the
+upstream datasets in [ATTRIBUTION.md](ATTRIBUTION.md). The author list follows
+the prepared manuscript; no conference acceptance or published identifier is
+asserted. `RELEASE.json` records the build scope and source snapshot.
+`MANIFEST.json` identifies every frozen file, including the actual working-tree
+code used despite existing uncommitted manuscript revisions. Reports under
+`validation/` state what was actually checked during local preparation. Their
+historical `publication_performed: false` fields describe those offline checks,
+not the current Hub visibility. The release retains the exact evaluated data,
+proofs, scoring records and software of the frozen preparation candidate.

RECONSTRUCTION.md

@@ -0,0 +1,65 @@
+# Reconstruct the GQA evaluation inputs locally
+
+The public candidate includes our labels, factual source IDs, image/question
+checksums and our edit deltas for all 3,000 GQA views. It omits upstream questions,
+full semantic programs, full graphs and photographs because an explicit GQA
+annotation redistribution grant was not found. Read the upstream terms when
+obtaining the source files. Do not upload the reconstructed directory as part
+of this candidate without resolving the missing grant.
+
+1. Obtain `questions1.2.zip` and `sceneGraphs.zip` from the
+   [official GQA downloads](https://cs.stanford.edu/people/dorarad/gqa/download.html).
+   Extract `val_balanced_questions.json` and `val_sceneGraphs.json`.
+2. Obtain the official `images.zip`, or the equivalent original images from
+   [Visual Genome](https://homes.cs.washington.edu/~ranjay/visualgenome/api.html).
+   Put the selected original JPEGs in one directory named `{image_id}.jpg`.
+   The needed 1,000 IDs are in the `gqa` configuration and reconstruction recipes.
+   Re-encoded third-party mirror images may not reproduce the frozen bytes.
+3. Install the pinned `requirements.txt` and run:
+
+```bash
+python scripts/reconstruct_hf_gqa.py \
+  --questions /path/to/val_balanced_questions.json \
+  --scene-graphs /path/to/val_sceneGraphs.json \
+  --images /path/to/original-images \
+  --output ../gqa-reconstructed
+```
+
+The script performs no network download or upload. It requires a new output
+directory outside the release. It checks original question strings and source
+programs, canonical graphs, normalized photograph bytes, and every final image
+against the frozen evaluated identities. It replays the source program and the
+symbolic alternatives, then writes an ImageFolder-compatible `metadata.jsonl`,
+3,000 JPEGs and the reconstructed program/edit evidence. Any mismatch fails
+explicitly; it does not substitute newly generated examples.
+
+```python
+from datasets import load_dataset
+gqa = load_dataset(
+    "imagefolder",
+    data_files={"test": [
+        "../gqa-reconstructed/images/*.jpg",
+        "../gqa-reconstructed/metadata.jsonl",
+    ]},
+    split="test",
+)
+```
+
+Use this explicit file/split selection. Directory-wide split inference can pick
+up validation-report filenames; those reports are not model inputs. These
+3,000 examples belong to the benchmark's evaluation-only test cohort.
+
+The GQA renderer decodes the source JPEG to RGB, writes a normalized JPEG at
+quality 88 with 4:2:0 subsampling, and applies a `(127,127,127)` patch over one
+recorded object box for edited views. FULL avoids a second JPEG encoding.
+Patched views are encoded once from the normalized source. Exact codec behavior
+is why the pinned Pillow version matters. CLEVR and PlotQA need no reconstruction
+for model evaluation because their evaluated pixels are embedded in Parquet.
+
+The CLEVR source worlds/programs and renderer settings are bundled for auditing.
+Optional regeneration of the rendered CLEVR scenes additionally requires the
+official generator/assets at revision
+`f0ce2c81750bfae09b5bf94d009f42e055f2cb3a`, Blender 2.79b/Cycles and the settings
+in the view evidence. These dependencies are not needed to load the dataset or
+to replay its source-program labels. GQA graph alternatives are symbolic scene
+evidence and do not establish identical pixels for hypothetical photographs.

RELEASE.json

@@ -0,0 +1,78 @@
+{
+  "version": "answerability-v1.0.0-20260928",
+  "status": "public_release_payload",
+  "paper_snapshot": "2026-09-28",
+  "code_git_head": "31ba785af150ea48c8dc090bc4e1ab403a6e7162",
+  "code_worktree_dirty": true,
+  "code_identity": "Frozen per-file digests in MANIFEST.json identify the actual copied working-tree code, including uncommitted revisions.",
+  "views": {
+    "plotqa": 5000,
+    "clevr": 4000,
+    "gqa": 3000
+  },
+  "groups": 3000,
+  "embedded_images": 9000,
+  "gqa_reconstruction_views": 3000,
+  "embedded_image_bytes": 1097646097,
+  "saved_scoring_records": 72000,
+  "chart_proofs": 1000,
+  "scope": "12,000 evaluated views only; the broader 58,000-view training/development resource is not this test release.",
+  "gqa_policy": "No upstream question text, full semantic programs, full graphs, or photographs are redistributed. GQA question references replace text in numerical-replay projections.",
+  "source_archive": "evi-core-v2-official-blender-2026-07-30",
+  "original_source_manifest": {
+    "contract": "evi_core_multidomain_v2",
+    "group_counts": {
+      "certified_eval": 1000,
+      "train": 12000,
+      "validation": 2000
+    },
+    "per_example_human_decisions": 0,
+    "pipeline_code_commit": "e4db1a2de24991898ca4c2eddb29dd89664261f5",
+    "pipeline_config_sha256": "d228b44f82ab09b38b0fd94671979283f3d5ff7272771fbb243a2a57ad02e3ae",
+    "reasoning_vlm_calls": 0,
+    "rejection_counts": {
+      "clevr_train:duplicate_source_image": 876,
+      "clevr_train:group_reject:ClevrGroupError": 6733,
+      "clevr_train_render:ClevrBlenderVisibilityError": 18,
+      "clevr_validation:duplicate_source_image": 268,
+      "clevr_validation:group_reject:ClevrGroupError": 1614,
+      "clevr_validation_render:ClevrBlenderVisibilityError": 4,
+      "gqa_train:duplicate_source_image": 1257,
+      "gqa_train:group_reject:GqaGroupError": 15042,
+      "gqa_train:group_reject:IngestError": 3436,
+      "gqa_train:unsupported_or_malformed_program": 141794,
+      "gqa_validation:duplicate_source_image": 373,
+      "gqa_validation:group_reject:GqaGroupError": 2968,
+      "gqa_validation:group_reject:IngestError": 685,
+      "gqa_validation:unsupported_or_malformed_program": 20027
+    },
+    "resource_manifest_sha256": "01e99f1187025455bf0743544e662a8b179e29e446dc1062b13b4fffec5faedc",
+    "schema_version": 1,
+    "source_group_counts": {
+      "clevr": 5000,
+      "gqa": 6000,
+      "plotqa": 4000
+    },
+    "source_view_counts": {
+      "clevr": 20000,
+      "gqa": 18000,
+      "plotqa": 20000
+    },
+    "split_group_jsonl_sha256": {
+      "certified_eval": "c94e5fa3507a8c218a7f1ed0a85686cf8df3bacd39869febed2d728fdd489bc8",
+      "train": "cb93762241d68fbfd163ae2168060c1af8b980f707fdc4afeb1bf0dae8692880",
+      "validation": "816f28010134e32707720bd592dc9729f83f925629282b71c4510e8674ff2605"
+    },
+    "split_view_counts": {
+      "certified_eval": 5000,
+      "train": 46000,
+      "validation": 7000
+    },
+    "total_groups": 15000,
+    "total_views": 58000
+  },
+  "repository": "sungguk/visual-answerability",
+  "release_tag": "v1.0.0",
+  "prepared_manifest_sha256": "3959fb0365bcf14c55e01123cca90a5bce1744407e70b80511192b3842139f03",
+  "publication_note": "Remote commit and completion checks are recorded in the external publication receipt."
+}

UPLOAD.md

@@ -0,0 +1,36 @@
+# Release download and verification
+
+The public dataset is
+[sungguk/visual-answerability](https://huggingface.co/datasets/sungguk/visual-answerability).
+The frozen release tag is `v1.0.0`; the internal cohort version remains
+`answerability-v1.0.0-20260928`. The separate private training handoff is not
+part of this public repository.
+
+Download the complete release, including evidence and evaluation scripts:
+
+```python
+from huggingface_hub import HfApi, snapshot_download
+
+repo_id = "sungguk/visual-answerability"
+commit = HfApi(token=False).dataset_info(repo_id, revision="v1.0.0").sha
+print("Release commit:", commit)
+root = snapshot_download(repo_id, repo_type="dataset", revision=commit,
+                         local_dir="hf-dataset", token=False)
+```
+
+Install `hf-dataset/requirements.txt`, then follow `EVALUATION.md` to verify
+the frozen manifest, decode and hash the 9,000 embedded images, check all
+1,000 bounded chart proofs, and replay the reported numerical results.
+Write validation outputs and reconstructed GQA data outside `hf-dataset`.
+`MANIFEST.json` records SHA-256 digests of all release files except itself.
+
+The public release changes only publication documentation and metadata from
+the verified local preparation candidate. Upstream GQA questions, full
+programs, graphs and photographs are excluded. Obtain the official inputs
+and follow `RECONSTRUCTION.md` for that configuration.
+
+Maintainer checks before assigning the release tag include an anonymous
+download of every file, exact file-set and digest comparison against the
+manifest, and loading all three configurations from the Hub commit. Viewer
+processing is checked separately. Future data changes require a new version;
+the `v1.0.0` tag must continue to identify this verified snapshot.

data/clevr/test-00000-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2e57317940534dbc746da2a1d4d722d66b45ad58dad7c556b5c813d433f10e6
+size 122029816

data/clevr/test-00001-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:058eb964b60f36fc000e757c4f6e0233e0cb2fc13bcc0d6a150b104cc28f31f5
+size 122560874

data/clevr/test-00002-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40ae3fe057b5c148aab17f8971da95cae465210a7ed0817ff34d6f01aa41b2dc
+size 122649027

data/clevr/test-00003-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bd8efa25a9d49b6b3e02516c9f99b89819cb5e5fd4098f996cd427dcdb07d58
+size 122427965

data/clevr/test-00004-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e54588c2274b4daedea474581c7928b8d5021d18bfe687b90564a8c5f7447373
+size 122424211

data/clevr/test-00005-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4f526760e70022961a2c35af2fcb3dc5e35e208a6fb5d3f1fa0ae2f9c2febd1
+size 122703602

data/clevr/test-00006-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00626dd6a34ff6f1b759d5ac84236ed8d2157b9579a3ac76dd749babeaae8ab7
+size 121692845

data/clevr/test-00007-of-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f7534965d03dce462288fc9015a08e363d7272fed24c4acfcf1d3b5639f3f07
+size 122581306

data/gqa/test-00000-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5656d0e4ac4199781239cad88f997a20dd6c527b84e8df62974a7193a1259f96
+size 189726

data/gqa/test-00001-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0efdcd8d6046c6b146d19298d44b35a5404f6706599b6ceee625232a0a0804f
+size 190510

data/gqa/test-00002-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c500b4ff9d37192abae70e6771df2f57f92c1c9fd1dcb3855ed7a89f21aa192
+size 190609

data/gqa/test-00003-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84cfdfedf3658813113f2b4d013ec26dc45593d76cccd9ea0445e1e2e95c8975
+size 190660

data/gqa/test-00004-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f2c3e8e9717e830270b1969283fd0910e177e751d87e81d4c7f206f03ef6c22
+size 189491

data/gqa/test-00005-of-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25dbbf3042f06949bb928d9e850be6610f2111ab908d9cb2463608fa5286a538
+size 190550

data/plotqa/test-00000-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3749b9bd6090ba503858a82ede85367fd43fb2b883f39b773391c29475bd0d2e
+size 12135769

data/plotqa/test-00001-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd5a526876de16d4049217eaedb72bb72635efb0d0622517b282561eaeaa8c5d
+size 12497883

data/plotqa/test-00002-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b31d69de0cc562adc3bb78e226af434c8473b45b57004d667cecce2d53747af1
+size 12339728

data/plotqa/test-00003-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcb24c4921d6a4739ee72f010e7e568996e58c4b73d81b68fb5dedec75f350e1
+size 12136417

data/plotqa/test-00004-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e848d6faa4823c5286c2850d4d893a1df537fe969814d90dc2eb7b667b29137
+size 12141572

data/plotqa/test-00005-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df6110e710d613098a2302bd03e2bd79a76af0816644b5ea0ca877779c28a998
+size 12123738

data/plotqa/test-00006-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3f1abd5f33c9f5da99f380c3ed0c657893514ab732f83da689ce12eb019d506
+size 12285517

data/plotqa/test-00007-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48441f34fd91e259e81a09ce1d1d1a252c14b4c6407ad0913e14d12b16aa77fc
+size 12336336

data/plotqa/test-00008-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebec199a245671a12a677f3b962c34028e657f7156b3a00845a5f2f5a1ac64b5
+size 12261869

data/plotqa/test-00009-of-00010.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1fd438190c07e4e7ba74183a8ea94f4420124f1bd734e3caa175ad1cc9d5ef9
+size 12292552

evidence/clevr/groups.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc5bcb5bc5d277ae14c9d0769e1f9524837e08efdc0d1cef456c5d7a446b8573
+size 1052294

evidence/clevr/views.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:398afbadef97b334d78bc5febc083887a79d003183a423810e8cac406094f54f
+size 3028987

evidence/finite/original-protocol.json

@@ -0,0 +1,35 @@
+{
+  "schema": "compositional_certificate_study_v1",
+  "seed": 20260914,
+  "cases_per_family": 128,
+  "families": [
+    "sum",
+    "difference_of_sums",
+    "mean",
+    "range",
+    "threshold",
+    "equality",
+    "clipped_sum",
+    "product_of_differences"
+  ],
+  "domain": [
+    0,
+    9
+  ],
+  "cells": 4,
+  "hidden_cells": [
+    1,
+    2
+  ],
+  "sampling": "independent_uniform_world_and_mask",
+  "question": "deterministic_program_verbalization",
+  "independence": "fresh worlds; no PlotQA groups, model outputs or hand adjudication",
+  "oracle": "all finite completions; separate postfix executor",
+  "baselines": [
+    "schema_source_replay_render",
+    "dependency_cut",
+    "interval_propagation",
+    "all_low_vs_all_high_witness"
+  ],
+  "scope": "descriptive synthetic finite-domain validation; not real-world prevalence"
+}

evidence/finite/original-results.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de5ed6e92dd0a36067807d970f9974e86657e040796bdf73e63db7c8e3dc874b
+size 27018

evidence/finite/solver-protocol.json

@@ -0,0 +1,102 @@
+{
+  "schema": "finite_solver_comparison_v2",
+  "seed": 20260915,
+  "original_cases": 1024,
+  "holdout_cases": 432,
+  "holdout_generator": "recursive random syntax tree; original family generator is not called",
+  "maximum_depths": [
+    2,
+    4,
+    6
+  ],
+  "maximum_nodes": 31,
+  "domains": [
+    [
+      0,
+      9
+    ],
+    [
+      -3,
+      6
+    ],
+    [
+      0,
+      4
+    ]
+  ],
+  "hidden_counts": [
+    1,
+    2,
+    3
+  ],
+  "cases_per_depth_domain_hidden_cell": 16,
+  "leaf_stop_probability": 0.25,
+  "leaf_cell_probability": 0.75,
+  "leaf_cell_sampling": "uniform over indices 0,1,2,3",
+  "leaf_constants_inclusive": [
+    -3,
+    12
+  ],
+  "leaf_constants_sampling": "uniform integers",
+  "division_denominators_uniform": [
+    -3,
+    -2,
+    -1,
+    1,
+    2,
+    3
+  ],
+  "source_values_sampling": "four independent uniform integers in the declared domain",
+  "hidden_sampling": "uniform subset without replacement, stored in sorted order",
+  "child_node_budget": "floor((parent_budget-1)/2) for each child; depth cap 6 realizes at most depth 4",
+  "operators_uniform": [
+    "add",
+    "sub",
+    "mul",
+    "div",
+    "min",
+    "max",
+    "gt",
+    "eq"
+  ],
+  "smt_timeout_ms": 1000,
+  "smt_random_seed": 0,
+  "unknown_policy": "retain as unresolved, never relabel constant",
+  "oracle": "complete enumeration with separate postfix executor",
+  "scope": "author-generated finite stress test; no natural-image or prevalence claim",
+  "original_protocol": {
+    "schema": "compositional_certificate_study_v1",
+    "seed": 20260914,
+    "cases_per_family": 128,
+    "families": [
+      "sum",
+      "difference_of_sums",
+      "mean",
+      "range",
+      "threshold",
+      "equality",
+      "clipped_sum",
+      "product_of_differences"
+    ],
+    "domain": [
+      0,
+      9
+    ],
+    "cells": 4,
+    "hidden_cells": [
+      1,
+      2
+    ],
+    "sampling": "independent_uniform_world_and_mask",
+    "question": "deterministic_program_verbalization",
+    "independence": "fresh worlds; no PlotQA groups, model outputs or hand adjudication",
+    "oracle": "all finite completions; separate postfix executor",
+    "baselines": [
+      "schema_source_replay_render",
+      "dependency_cut",
+      "interval_propagation",
+      "all_low_vs_all_high_witness"
+    ],
+    "scope": "descriptive synthetic finite-domain validation; not real-world prevalence"
+  }
+}

evidence/finite/solver-results.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d383d15f970f43f380a568c30015fe9abb21c853e9ca41cdd1db08465e29f24c
+size 61448

evidence/gqa/reconstruction.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0ff47c1a4ea8498dfbf5f880158e409f7b8ab312f862b365d1fcd1b28f4c5e4
+size 483217

evidence/plotqa/bounded-witness-proofs.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96086574e2684c40d7d718379af3baf308c7d5cc276bce00fb3b6bedb10188ff
+size 24247132

evidence/plotqa/groups.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:130e8cf3d801dd31253720c3a66db9bba2a06e894c09f862437cece291042a22
+size 1243853

evidence/plotqa/views.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09cb99c9f91e6668fea404e1624ce24e9ad542afc101e655befb8fc481a699bc
+size 3138359

expected/diagnostic-analysis.json

@@ -0,0 +1,255 @@
+{
+  "status": "passed",
+  "inference_calls": 0,
+  "analysis_status": "post-hoc descriptive accounting on frozen responses",
+  "sha_validation_performed": false,
+  "domains": {
+    "plotqa": {
+      "qwen38-flash-next": {
+        "groups_by_number_of_failed_states": [
+          835,
+          147,
+          11,
+          7,
+          0,
+          0
+        ],
+        "failures": 190,
+        "affected_groups": 165,
+        "failures_beyond_first_per_affected_group": 25,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "gemma4-31b-it": {
+        "groups_by_number_of_failed_states": [
+          667,
+          300,
+          19,
+          14,
+          0,
+          0
+        ],
+        "failures": 380,
+        "affected_groups": 333,
+        "failures_beyond_first_per_affected_group": 47,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "qwen": {
+        "groups_by_number_of_failed_states": [
+          665,
+          283,
+          39,
+          13,
+          0,
+          0
+        ],
+        "failures": 400,
+        "affected_groups": 335,
+        "failures_beyond_first_per_affected_group": 65,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "gemma4-26b-a4b-it": {
+        "groups_by_number_of_failed_states": [
+          539,
+          419,
+          22,
+          16,
+          4,
+          0
+        ],
+        "failures": 527,
+        "affected_groups": 461,
+        "failures_beyond_first_per_affected_group": 66,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "glm": {
+        "groups_by_number_of_failed_states": [
+          273,
+          558,
+          167,
+          1,
+          1,
+          0
+        ],
+        "failures": 899,
+        "affected_groups": 727,
+        "failures_beyond_first_per_affected_group": 172,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "molmo2-8b": {
+        "groups_by_number_of_failed_states": [
+          194,
+          509,
+          297,
+          0,
+          0,
+          0
+        ],
+        "failures": 1103,
+        "affected_groups": 806,
+        "failures_beyond_first_per_affected_group": 297,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      }
+    },
+    "clevr": {
+      "qwen38-flash-next": {
+        "groups_by_number_of_failed_states": [
+          454,
+          506,
+          28,
+          11,
+          1
+        ],
+        "failures": 599,
+        "affected_groups": 546,
+        "failures_beyond_first_per_affected_group": 53,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "gemma4-31b-it": {
+        "groups_by_number_of_failed_states": [
+          210,
+          451,
+          89,
+          245,
+          5
+        ],
+        "failures": 1384,
+        "affected_groups": 790,
+        "failures_beyond_first_per_affected_group": 594,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "qwen": {
+        "groups_by_number_of_failed_states": [
+          273,
+          505,
+          144,
+          69,
+          9
+        ],
+        "failures": 1036,
+        "affected_groups": 727,
+        "failures_beyond_first_per_affected_group": 309,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "gemma4-26b-a4b-it": {
+        "groups_by_number_of_failed_states": [
+          189,
+          491,
+          79,
+          228,
+          13
+        ],
+        "failures": 1385,
+        "affected_groups": 811,
+        "failures_beyond_first_per_affected_group": 574,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "glm": {
+        "groups_by_number_of_failed_states": [
+          205,
+          793,
+          2,
+          0,
+          0
+        ],
+        "failures": 797,
+        "affected_groups": 795,
+        "failures_beyond_first_per_affected_group": 2,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "molmo2-8b": {
+        "groups_by_number_of_failed_states": [
+          216,
+          562,
+          107,
+          112,
+          3
+        ],
+        "failures": 1124,
+        "affected_groups": 784,
+        "failures_beyond_first_per_affected_group": 340,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      }
+    },
+    "gqa": {
+      "qwen38-flash-next": {
+        "groups_by_number_of_failed_states": [
+          399,
+          448,
+          137,
+          16
+        ],
+        "failures": 770,
+        "affected_groups": 601,
+        "failures_beyond_first_per_affected_group": 169,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "gemma4-31b-it": {
+        "groups_by_number_of_failed_states": [
+          372,
+          453,
+          160,
+          15
+        ],
+        "failures": 818,
+        "affected_groups": 628,
+        "failures_beyond_first_per_affected_group": 190,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "qwen": {
+        "groups_by_number_of_failed_states": [
+          359,
+          455,
+          170,
+          16
+        ],
+        "failures": 843,
+        "affected_groups": 641,
+        "failures_beyond_first_per_affected_group": 202,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "gemma4-26b-a4b-it": {
+        "groups_by_number_of_failed_states": [
+          392,
+          400,
+          192,
+          16
+        ],
+        "failures": 832,
+        "affected_groups": 608,
+        "failures_beyond_first_per_affected_group": 224,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "glm": {
+        "groups_by_number_of_failed_states": [
+          305,
+          651,
+          42,
+          2
+        ],
+        "failures": 741,
+        "affected_groups": 695,
+        "failures_beyond_first_per_affected_group": 46,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      },
+      "molmo2-8b": {
+        "groups_by_number_of_failed_states": [
+          267,
+          573,
+          146,
+          14
+        ],
+        "failures": 907,
+        "affected_groups": 733,
+        "failures_beyond_first_per_affected_group": 174,
+        "accounting_identity": "k*E = (1-B_k) + repeated_failures/G"
+      }
+    }
+  },
+  "gqa_reweighting": {
+    "definition": "E_lambda=(1-lambda)*E_A+lambda*E_U, fixed predictions and labels",
+    "QwenNext_minus_GLM_answerable_pp": 11.6,
+    "QwenNext_minus_GLM_unanswerable_pp": -20.3,
+    "crossover_lambda_point_estimate": 0.36363636363636365,
+    "interpretation": "arithmetic sensitivity; not a measured deployment prevalence or calibrated threshold"
+  }
+}