Publish complete Electronics submission: A100, MI250X, paper and raw evidence

Freeze electronics-submission-2026-10-22-r3. Update the dataset card and 29 loadable views; publish matching MDPI PDFs and source ZIP; add the historical, independent-offload, A100 follow-up, and MI250X evidence archives with hashes and CPU reconstruction reports. Preserve earlier releases.

CITATION.cff

@@ -2,7 +2,7 @@ cff-version: 1.2.0
 message: Please cite the RLVRAMBench evidence release and associated paper.
 type: dataset
 title: 'RLVRAMBench: GPU memory feasibility and transfer benchmark'
-version: electronics-submission-2026-10-22-r2
+version: electronics-submission-2026-10-22-r3
 date-released: '2026-10-22'
 license: MIT
 authors:
@@ -14,4 +14,4 @@ authors:
   orcid: https://orcid.org/0000-0003-0626-2303
 repository-code: https://github.com/kobzaond/RLVRAMBench-reproduce
 doi: 10.57967/hf/10780
-url: https://huggingface.co/datasets/kobzaond/RLVRAMBench/tree/electronics-submission-2026-10-22-r2
+url: https://huggingface.co/datasets/kobzaond/RLVRAMBench/tree/electronics-submission-2026-10-22-r3

MANIFEST.sha256

@@ -1,11 +1,11 @@
 94046015d51d7f1ab8e700a09b333fc1e5bb525311a0a6bd5a22cba6f859c663  BENCHMARK.md
-23306037696c32d990987d6c54e235a5ca8a5e3f625951e6e9e74b321a24df33  CITATION.cff
+71ab41e15c6acdcaf099518b2813d788ceaa1054594862d65ac8a8313b4164b2  CITATION.cff
 0d1f722917238e3a5189e4d0de8dfb66bd8a09f91181050529829a2723b70245  DATA_DICTIONARY.md
 7518af731dc7cf4f069effaabb5bef12dfb0142becaddbf7adf4ac873adec98f  LICENSE
-a3090ab63fcb42bac263d40838b57872e86c55a7e46f8220eddb472ef0bf72c3  README.md
+d6905a34e71cb7e02dcbd26efd9db7fcb5c2c16fdf329f7a869f9df5f2f3c418  README.md
 7c1f948ecd8082ad2d0cbe334e576566d6e26a5ccfdb5ac19b69611044fafe96  RIGHTS.md
-762d03d5ed2181b540eb3e748add96046f626e07543786e51e2aaf4422e8863d  SUBMISSION_DATA_DICTIONARY.md
-57455520bd66f625ccb27a7d5ad4345d43acadd7f1d9f38f730dc8cd06005a40  SUBMISSION_REPRODUCTION.md
+3b8ee3ae44b97164fb890df152eb23f0da42cf7cb7460e8ec3f88e94594cde72  SUBMISSION_DATA_DICTIONARY.md
+3aa978005a1f887b9100dd5c3f6b3a31fd35a63d084ed3a91e9ffee31e7207aa  SUBMISSION_REPRODUCTION.md
 9de37c2f38760c276f04e54e2d12d64d8ef95b57f2ea720cc1cea0ba4011daac  amd_mi250x/README.md
 9a929af449a2a922729d8da73b2d4074bf544bc7b6e348c76048ca226f6d2e62  amd_mi250x/analyze.py
 726684913078fffb8da72885f9efca19f0b90e823d12119e2142232447a69551  amd_mi250x/collect.py
@@ -199,9 +199,10 @@ b23baafdf95f91d95668116c53483878db8b702db1a1254a18c6775beef5ff9c  benchmark/revi
 c6e42d81e3da672faad298bb721b0d22f14dec99670a32149633aa6b6e5f478d  benchmark/stage_measurements.csv
 3378dbd45be90928ab839f9a4f16cc807b03168665ab169a8fcd98f1c9012c80  benchmark/summary.json
 57cc5bff368eb5a981faaea300a097d11c9b7734031a8bf4a3a40cfef408686d  benchmark/transfer_configurations.csv
+859c54846ff064c90b60d31ba2738f97f2593e4216806f56944601403aaf2455  dataset-client-requirements.txt
 7fcee3295069b601b901dbdc24bb087ca96e305949d069107b54c8d1014b120d  electronics_revision_2026_10/MANIFEST.sha256
 273db17f97473dab01f1500f974c7c92082307b4e07ca15a8c44ee4dc5e3f706  electronics_revision_2026_10/README.md
-1810e252dba54ed0dc427e1bfb03f40a256b07f5ddd1cfae1b027e5a2d00478e  electronics_revision_2026_10/SUPERSEDED.md
+4c9de7d8de3b629e76eec7bfb28897bcb8ba6ac0eabedc865bc93ea7b69fcdee  electronics_revision_2026_10/SUPERSEDED.md
 397c8caaf0d3abe06587fd649c181fe71d5b8b5f5e24e0591cc7e56d38e275c1  electronics_revision_2026_10/benchmark/attempts.csv
 0c099b96f7ab71c03e02f7fcfca4dd581e5a5997bca9ea473b75d6c616848078  electronics_revision_2026_10/benchmark/baseline_predictions.csv
 b5ceba2f04aa2ceda5b236c220e342ba1d637fd7c326c61af791cd991bbdaea6  electronics_revision_2026_10/benchmark/baseline_scores.json
@@ -732,16 +733,16 @@ c7fffe0231922e5521dc360889fafe941ee26d154c1ce4e167e4ec9f2d13498f  paper/Definiti
 a77a994f978c1860ff14dbcae74631f3ed835b418adbdcf84de2f06d98c34599  paper/Definitions/mdpi_apacite.sty
 4a73faf6e0cf1a225d1b9a41a6955d3cc29a55c5fdce662177d00eee3946ce36  paper/Definitions/mdpi_chicago.bst
 1a41f6c85db06401a4582008e6b8933c7212d17541497701797a55fe473d19e0  paper/Definitions/unicode.tex
-9db9205fa3300aa9a5761e881215bf052bb58ff2ac7cbe5c6d42bec8a216d125  paper/README_revision.md
-335790338a773beeb74f21b53e7a8f43f20734153ae44d830cd6d5e92e32bd4a  paper/RLVRAMBench_Electronics_revision_source.zip
+691079c5f11bccd75c1137f5a0a62ee5e4ee197813f8b54c8067017731f6242c  paper/README_revision.md
+bb14363f7a135b4c32aeeec1a039072858bba610fee3c6c8ecb04628f561d4d1  paper/RLVRAMBench_Electronics_revision_source.zip
 c7eef7d90561cae3961ec4c98b4cd6d1076ff63b17cf2be2d7604af8016efc1d  paper/RLVRAMBench_IEEE_Access_source.zip
 9b28d833742a1befddc4b55b67968fb449d469e95b57f3451adeb509a5909865  paper/RLVRAMBench_IEEE_Access_submission.pdf
 a9f86b81a93a377f47d84a8c8b361fd338679f48839633fca6021e21c57ff233  paper/RLVRAMBench_IEEE_Access_submission.tex
-90d46065b8d84570db402c1adcc7a59c0d244945f1d928eee1e0eaa34b0b832f  paper/RLVRAMBench_MDPI.bbl
-97891c8d6e2e5b0d289bf6d192ea6fff82081065c4b768e917cbb4fadb55d926  paper/RLVRAMBench_MDPI.pdf
-751c59f46bf5aa5102fb1d99566ee4bbf2fbc2c7e43e97e1dba66144fc24a054  paper/RLVRAMBench_MDPI.tex
-1035e92acbb579eee7767ab8ac5f1066fa9286f00b8f616c9721a019c2295fd4  paper/RLVRAMBench_reviewer_response.pdf
-c8d6bceb9785d0bf654c24543ff1e4a4a6d697d14f72163a23bd7c98b024d3b5  paper/RLVRAMBench_reviewer_response.tex
+b2e5a4ccc8f83c3861224bc193b465b67e1f2af3e0b371362dba3d8548760bdd  paper/RLVRAMBench_MDPI.bbl
+4c2ca4795ca43804f2c70850ea81cc33e1ecaa7f98db552d2613e00d963d0772  paper/RLVRAMBench_MDPI.pdf
+664e8443994c7564c9ae5ed10e8c4bc808c96083f5dc86f8b78a27bb834715a4  paper/RLVRAMBench_MDPI.tex
+ecbfb413196c26094b9cfd53ad192caa62fb82dd92f43c56a255c0d541c8e7ee  paper/RLVRAMBench_reviewer_response.pdf
+b11b78eccca034a229d4577fa7f4c48d026ac251ca94543f00e2b0c74739bdd6  paper/RLVRAMBench_reviewer_response.tex
 0a389b5b6c0a51c20796312592e45df2d87af1629c2010baed20ec4e0832dcbb  paper/RLVRAMBench_supplement.pdf
 800544fd0de5b14d94f5324a6d4d58390c13503ee67a5dab643d5af529ca32e1  paper/RLVRAMBench_supplement.tex
 34986c6aa0ac2028056dd1c6813ab1e68068493ebbfe679ca29149627e212c44  paper/electronics_revision.bib
@@ -764,8 +765,8 @@ fc217479985a1e8e871fb377f522c83b21ea60a0c2603480a2a8a25efd25dd09  paper/generate
 495001a07ce54a71c94e58c6444e663b624b519000074c7bc3ab195d52e76177  paper/generated/electronics_reservation.tex
 a2bcc83ac256e8944528562a049eda00d660c7508462747f1864a2adf126002a  paper/generated/electronics_timing.tex
 16b9baaae4987970bef1421ca7266dafba0b32cae8eeea9539955476161bfad9  paper/ieee_access_additions.bib
-97624a960a4837e14dbd03cba6a8e11e727389a6ed4bca525a0aa30672f5bcb3  paper/references.bib
-c53ac467ae8211a113739d8027981007f334b7c988cbdafaa806819fe2ea9f5f  paper/revision_build_verification.json
+e564cecfeb04a0296dada198334adc0a9d094f81f914c3696a18a8ea8faf7a44  paper/references.bib
+51b5037ed4c7f8a873b8e0a1464d80f2953aaeaff2ee3b8bd1727efc06913f61  paper/revision_build_verification.json
 f7c5a3ce45ef9496622af981618a310011bbbf09573e4f874111ed59e450b377  paper/supplement_deepspeed.tex
 e093449c46e5864b9cd97c563539f3c8031c4431afbe652d93066bd59007ae30  paper/supplement_mi250x.tex
 67361722e05e7d2561728f005621fd906acbdce413d54349ee2dc4217032bdc3  profiles/electronics_revision/adaptation/cases.csv
@@ -811,7 +812,7 @@ d45ab7337f86cc4ffd53c3aabf6aca5333c96c9c2578c4abfd51ad630119ea1a  profiles/elect
 a9b6de05d68c17143c23d3ba766fc686f972c8bc8b0199293678183b7bd0e9bd  profiles/electronics_revision/timing/summary.json
 b9299810ea76e2199e3489f73847e85959dbbf2777778188ef57e9c98014ae72  profiles/electronics_revision/timing_condition_summary.csv
 c11edabd061028e385944e85996da0136b583379fb80fbc607802bc14085853d  publication-requirements.txt
-cb486e40beadfd61fccd3f54d671c88be1dfd8f76bb8d24fbb923b06932da064  release.json
+d4aedc9c64d0ed21f67a5668bcb8cc41780089b03eeeb2440358a5b9ed77c934  release.json
 400f75b1c3015d9903b3e898b0ed2dc3771b07363bff7f8761aec0c798ed67ae  releases/RLVRAMBench-mi250x-raw-20261022.tar.zst
 8c7c6979c7652dc1ad413fb4e4fd25642f77a856d4d64b06ca999b0c4ce25a63  releases/RLVRAMBench-mi250x-raw-20261022.tar.zst.sha256
 ac2c69bae8d7e0af07e27de9c182e095fa77a28f5b0d0fc906fbfff302040041  releases/RLVRAMBench-offload-raw-20261022.tar.zst
@@ -907,7 +908,7 @@ c196d8e78ed90586e3824cb5fbdfd0316005253a50c8bed7cfc27d3dc8ea0a99  verification/m
 6356306e94cdf55890bc40246cac87da273c045e749c6d64d5e72e5af30c72db  verification/offload-packaging.json
 f5bb0b872c758d2b4f5a640fc4772f97a8397785499ef11a44d4cce1a7ecb688  verification/offload-reconstruction.json
 bc21f03a41e9a989520260f705fa0a46abe6671630ec7ff568cbda63d83b2692  verification/package-preparation.json
-c53ac467ae8211a113739d8027981007f334b7c988cbdafaa806819fe2ea9f5f  verification/paper-build.json
+51b5037ed4c7f8a873b8e0a1464d80f2953aaeaff2ee3b8bd1727efc06913f61  verification/paper-build.json
 4e5cf66c174c8d7292fb585e522ce17927df4c13a5734548040c7fceed92e733  verification/reproduction-tests.json
 a367674ae14f0bbab157f8c9e6fede2a8e5fa7d88d2c30ad9c596e48a923be4e  verification/revised-reproduction.json
 a066598a8b5565f94d6f0dcb08bb14149d0c31b7da274e6d3bf47a0768be444b  verification/revised-source-package.json

README.md

@@ -139,7 +139,7 @@ This is the dataset accompanying **Memory Feasibility in Colocated Language-Mode
 Reinforcement Learning: A Failure-Aware Measurement Study**, prepared for the
 MDPI *Electronics* revision on **22 October 2026**.
 
-**Submission version: `electronics-submission-2026-10-22-r2`.** This version is
+**Submission version: `electronics-submission-2026-10-22-r3`.** This version is
 published on `main` and pinned by the same tag. Dataset DOI:
 [10.57967/hf/10780](https://doi.org/10.57967/hf/10780).
 Earlier releases remain in the repository history and `releases/`.
@@ -148,7 +148,7 @@ Earlier releases remain in the repository history and `releases/`.
 - [Scientific supplement](paper/RLVRAMBench_supplement.pdf)
 - [Point-by-point reviewer response](paper/RLVRAMBench_reviewer_response.pdf)
 - [Editable TeX source ZIP](paper/RLVRAMBench_Electronics_revision_source.zip)
-- [Matching reproduction code](https://github.com/kobzaond/RLVRAMBench-reproduce/tree/electronics-submission-2026-10-22-r2)
+- [Matching reproduction code](https://github.com/kobzaond/RLVRAMBench-reproduce/tree/electronics-submission-2026-10-22-r3)
 - [Release manifest and archive hashes](release.json)
 - [Submission data dictionary](SUBMISSION_DATA_DICTIONARY.md)
 - [Reconstruction guide](SUBMISSION_REPRODUCTION.md)
@@ -219,13 +219,24 @@ are missing evidence or inapplicable quantities, not zero memory.
 
 ## Load a view
 
+The direct Hub-loading interface is tested with Python 3.11,
+`datasets==5.1.0` and `huggingface_hub==2.2.0`. Install these from
+`dataset-client-requirements.txt` in a separate reader environment. Older
+clients such as `datasets==2.12.0` ignore the card's configuration mapping
+and may try to combine unrelated audit JSON files. The frozen raw-analysis
+environment is documented separately in the reproduction guide.
+
+```sh
+python -m pip install -r dataset-client-requirements.txt
+```
+
 ```python
 from datasets import load_dataset
 
 configurations = load_dataset(
     "kobzaond/RLVRAMBench",
     "mi250x_configurations",
-    revision="electronics-submission-2026-10-22-r2",
+    revision="electronics-submission-2026-10-22-r3",
     split="measured",
 )
 ```

SUBMISSION_DATA_DICTIONARY.md

@@ -1,6 +1,6 @@
 # Submission data dictionary
 
-Version: `electronics-submission-2026-10-22-r2`.
+Version: `electronics-submission-2026-10-22-r3`.
 The original [data dictionary](DATA_DICTIONARY.md) documents the historical
 benchmark and remains unchanged where it is bound to earlier source hashes.
 This document covers the submission extensions and their relationships.

SUBMISSION_REPRODUCTION.md

@@ -1,7 +1,7 @@
 # Electronics submission release — 22 October 2026
 
 The paper, dataset, and reproduction code use the version tag
-`electronics-submission-2026-10-22-r2`. The submission dataset is published on
+`electronics-submission-2026-10-22-r3`. The submission dataset is published on
 the Hugging Face `main` branch and pinned with the same tag:
 <https://huggingface.co/datasets/kobzaond/RLVRAMBench>.
 The dataset DOI is `10.57967/hf/10780`; use the version tag to identify
@@ -14,13 +14,24 @@ No model weights or optimizer checkpoints are distributed.
 
 ## Download the submission data
 
+The direct Hub-loading interface is tested with Python 3.11,
+`datasets==5.1.0` and `huggingface_hub==2.2.0`. Install these from
+the dataset's `dataset-client-requirements.txt` in a separate reader environment. Older
+clients such as `datasets==2.12.0` ignore the card's configuration mapping
+and may try to combine unrelated audit JSON files. The frozen raw-analysis
+environment is documented separately in the reproduction guide.
+
+```sh
+python -m pip install -r dataset-client-requirements.txt
+```
+
 ```python
 from huggingface_hub import snapshot_download
 
 snapshot_download(
     "kobzaond/RLVRAMBench",
     repo_type="dataset",
-    revision="electronics-submission-2026-10-22-r2",
+    revision="electronics-submission-2026-10-22-r3",
     local_dir="RLVRAMBench-data",
 )
 ```

dataset-client-requirements.txt

@@ -0,0 +1,2 @@
+datasets==5.1.0
+huggingface_hub==2.2.0

electronics_revision_2026_10/SUPERSEDED.md

@@ -1,4 +1,4 @@
 > **Earlier curated snapshot.** The 22 October Electronics submission is
 > defined by the root README, `release.json`, and tag
-> `electronics-submission-2026-10-22-r2`. Use the root-level data paths for
+> `electronics-submission-2026-10-22-r3`. Use the root-level data paths for
 > that submission; this folder preserves its earlier contents.

paper/README_revision.md

@@ -41,7 +41,7 @@ The point-by-point response and external-estimator audit live in
 source hashes are under `benchmark/electronics_revision_20261016/`.
 Derived analyses are under `profiles/electronics_revision/`. The MI250X
 harness, attempt ledger, and results, and the DeepSpeed estimator comparison,
-are in the reproduction repository (tag `electronics-submission-2026-10-22-r2`).
+are in the reproduction repository (tag `electronics-submission-2026-10-22-r3`).
 The Hugging Face dataset uses the same version tag and publishes this
 submission version on `main`, with the historical A100, independent-offload,
 A100 follow-up, and MI250X raw-evidence archives and their hashes.

paper/RLVRAMBench_Electronics_revision_source.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:335790338a773beeb74f21b53e7a8f43f20734153ae44d830cd6d5e92e32bd4a
-size 837738
+oid sha256:bb14363f7a135b4c32aeeec1a039072858bba610fee3c6c8ecb04628f561d4d1
+size 837736

paper/RLVRAMBench_MDPI.bbl

@@ -164,13 +164,13 @@ Agarwal, R.; Schwarzer, M.; Castro, P.S.; Courville, A.; Bellemare, M.G.
 Kobza, O.; {\v S}ediv{\'y}, J.
 \newblock {RLVRAMBench}: Benchmark Evidence Release.
 \newblock Hugging Face dataset,  2026.
-\newblock Version electronics-submission-2026-10-22-r2; released October 22, 2026, {\url{https://doi.org/10.57967/hf/10780}}.
+\newblock Version electronics-submission-2026-10-22-r3; released October 22, 2026, {\url{https://doi.org/10.57967/hf/10780}}.
 
 \bibitem[Kobza and {\v S}ediv{\'y}(2026{\natexlab{b}})]{kobza2026reproduction}
 Kobza, O.; {\v S}ediv{\'y}, J.
 \newblock {RLVRAMBench}: Resource Transfer Benchmark and Reproduction Code.
 \newblock GitHub repository,  2026.
-\newblock Tag electronics-submission-2026-10-22-r2; released October 22, 2026.
+\newblock Tag electronics-submission-2026-10-22-r3; released October 22, 2026.
 
 \bibitem[{OpenAI}(2026)]{openai2026codex}
 {OpenAI}.

paper/RLVRAMBench_MDPI.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97891c8d6e2e5b0d289bf6d192ea6fff82081065c4b768e917cbb4fadb55d926
-size 494758
+oid sha256:4c2ca4795ca43804f2c70850ea81cc33e1ecaa7f98db552d2613e00d963d0772
+size 494860

paper/RLVRAMBench_MDPI.tex

@@ -75,8 +75,8 @@ Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical Univer
 % Abstract
 \abstract{%
 When language-model reinforcement learning (RL) runs generation and
-training on the same graphics processing units (GPUs), whether a run fits
-in GPU memory depends on state kept between execution stages. We study
+training on shared graphics processing units (GPUs), memory feasibility
+depends on state kept between execution stages. We study
 this with Group Relative Policy Optimization in the VERL framework, the
 vLLM inference engine, and A100 GPUs, mostly with low-rank adaptation
 (LoRA). We keep failed runs as well as completions within or above a
@@ -90,7 +90,7 @@ copying outcome labels across models or GPU counts can approve settings
 that fail or exceed the limit. RQ3 (run duration): two of twelve one-step
 screens miss a later limit exceedance, although all four three-run labels
 agree with 100-step runs. On AMD MI250X with a newer vLLM, 79 of 89 complete labels
-transfer, but failures occur later, after the actor update, and the
+transfer, but most failures occur after the actor update, and the
 whole-run peak no longer hides actor growth. We release the measurements
 and evaluation software as RLVRAMBench.}
 
@@ -2608,7 +2608,7 @@ quality for code RL with test-based rewards.
 RLVRAMBench releases two items: versioned evidence on Hugging Face
 \cite{kobza2026benchmark} and separate reproduction software
 \cite{kobza2026reproduction}. The submission version is pinned by the
-tag \texttt{electronics-submission-2026-10-22-r2} in both repositories;
+tag \texttt{electronics-submission-2026-10-22-r3} in both repositories;
 the same dataset version is available on the Hugging Face main branch.
 Its release manifest identifies the separate historical A100,
 independent-offload, A100 follow-up, and MI250X raw-evidence archives
@@ -2703,7 +2703,7 @@ untested configurations.
 
 \funding{This research received no external funding.}
 
-\dataavailability{The submission dataset is openly available at \url{https://huggingface.co/datasets/kobzaond/RLVRAMBench} (dataset DOI: \url{https://doi.org/10.57967/hf/10780}; version tag \texttt{electronics-submission-2026-10-22-r2}, also published on the main branch). It contains the curated benchmark tables, A100 and MI250X follow-up results, estimator inputs and outputs, raw-evidence archives, and file-hash manifests. The matching reproduction and analysis code is available at \url{https://github.com/kobzaond/RLVRAMBench-reproduce} under the same version tag. Earlier releases remain separately identified. Model weights and optimizer checkpoints are not included.}
+\dataavailability{The submission dataset is openly available at \url{https://huggingface.co/datasets/kobzaond/RLVRAMBench} (dataset DOI: \url{https://doi.org/10.57967/hf/10780}; version tag \texttt{electronics-submission-2026-10-22-r3}, also published on the main branch). It contains the curated benchmark tables, A100 and MI250X follow-up results, estimator inputs and outputs, raw-evidence archives, and file-hash manifests. The matching reproduction and analysis code is available at \url{https://github.com/kobzaond/RLVRAMBench-reproduce} under the same version tag. Earlier releases remain separately identified. Model weights and optimizer checkpoints are not included.}
 
 \acknowledgments{OpenAI Codex \cite{openai2026codex} and Anthropic's Claude Code were used to assist with programming, experiment orchestration, evidence checks, analysis, and manuscript preparation, including language editing. The authors are responsible for the code, analyses, and content.}
 

paper/RLVRAMBench_reviewer_response.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1035e92acbb579eee7767ab8ac5f1066fa9286f00b8f616c9721a019c2295fd4
+oid sha256:ecbfb413196c26094b9cfd53ad192caa62fb82dd92f43c56a255c0d541c8e7ee
 size 115181

paper/RLVRAMBench_reviewer_response.tex

@@ -49,14 +49,14 @@
   pdfcreator={LaTeX via pandoc}}
 
 \ifXeTeX
-\special{pdf:trailerid [ <7ba286ececa1b752a686c30d6b06e0fa> <7ba286ececa1b752a686c30d6b06e0fa> ]}
+\special{pdf:trailerid [ <2e9a067e7201a7126d309cd52146036d> <2e9a067e7201a7126d309cd52146036d> ]}
 \fi
 \ifPDFTeX
 \pdftrailerid{}
-\pdftrailer{/ID [ <7ba286ececa1b752a686c30d6b06e0fa> <7ba286ececa1b752a686c30d6b06e0fa> ]}
+\pdftrailer{/ID [ <2e9a067e7201a7126d309cd52146036d> <2e9a067e7201a7126d309cd52146036d> ]}
 \fi
 \ifLuaTeX
-\pdfvariable trailerid {[ <7ba286ececa1b752a686c30d6b06e0fa> <7ba286ececa1b752a686c30d6b06e0fa> ]}
+\pdfvariable trailerid {[ <2e9a067e7201a7126d309cd52146036d> <2e9a067e7201a7126d309cd52146036d> ]}
 \fi
 
 \author{}
@@ -83,7 +83,7 @@ Below, each answer first states what we changed, then where the change
 is in the manuscript, and then the main evidence.
 
 The submission dataset and the matching reproduction code are pinned as
-electronics-submission-2026-10-22-r2. The Hugging Face main branch
+electronics-submission-2026-10-22-r3. The Hugging Face main branch
 contains this dataset version, including the raw-evidence archives,
 curated tables, estimator records, and verification reports. The MI250X
 comparison retains one unresolved configuration: 79 of 89 complete

paper/references.bib

@@ -388,7 +388,7 @@
   howpublished = {Hugging Face dataset},
   url = {https://huggingface.co/datasets/kobzaond/RLVRAMBench},
   doi = {10.57967/hf/10780},
-  note = {Version electronics-submission-2026-10-22-r2; released October 22, 2026}
+  note = {Version electronics-submission-2026-10-22-r3; released October 22, 2026}
 }
 
 @misc{kobza2026reproduction,
@@ -397,5 +397,5 @@
   year = {2026},
   howpublished = {GitHub repository},
   url = {https://github.com/kobzaond/RLVRAMBench-reproduce},
-  note = {Tag electronics-submission-2026-10-22-r2; released October 22, 2026}
+  note = {Tag electronics-submission-2026-10-22-r3; released October 22, 2026}
 }

paper/revision_build_verification.json

@@ -4,14 +4,14 @@
   "extracted_pdf_text_matches": true,
   "material_build_warnings": [],
   "review_response_finalized": true,
-  "source_bundle_sha256": "335790338a773beeb74f21b53e7a8f43f20734153ae44d830cd6d5e92e32bd4a",
+  "source_bundle_sha256": "bb14363f7a135b4c32aeeec1a039072858bba610fee3c6c8ecb04628f561d4d1",
   "files": {
-    "RLVRAMBench_MDPI.tex": "751c59f46bf5aa5102fb1d99566ee4bbf2fbc2c7e43e97e1dba66144fc24a054",
-    "RLVRAMBench_MDPI.pdf": "97891c8d6e2e5b0d289bf6d192ea6fff82081065c4b768e917cbb4fadb55d926",
+    "RLVRAMBench_MDPI.tex": "664e8443994c7564c9ae5ed10e8c4bc808c96083f5dc86f8b78a27bb834715a4",
+    "RLVRAMBench_MDPI.pdf": "4c2ca4795ca43804f2c70850ea81cc33e1ecaa7f98db552d2613e00d963d0772",
     "RLVRAMBench_supplement.tex": "800544fd0de5b14d94f5324a6d4d58390c13503ee67a5dab643d5af529ca32e1",
     "RLVRAMBench_supplement.pdf": "0a389b5b6c0a51c20796312592e45df2d87af1629c2010baed20ec4e0832dcbb",
-    "RLVRAMBench_reviewer_response.pdf": "1035e92acbb579eee7767ab8ac5f1066fa9286f00b8f616c9721a019c2295fd4",
-    "electronics_reviewer_response.md": "1f4f9fd070bcc617215823f5ec5576d8844a202f84c2d986b236699718d5c476"
+    "RLVRAMBench_reviewer_response.pdf": "ecbfb413196c26094b9cfd53ad192caa62fb82dd92f43c56a255c0d541c8e7ee",
+    "electronics_reviewer_response.md": "2f83fe4640ba9e40afb18dd282cd5c834381cb94974c70659a0f972e80379cd3"
   },
   "tex_versions": {
     "pdflatex": "pdfTeX 3.141592653-2.6-1.40.29 (TeX Live 2026)",

release.json

@@ -1,13 +1,13 @@
 {
   "schema_version": "rlvrambench-submission-1",
-  "release": "electronics-submission-2026-10-22-r2",
+  "release": "electronics-submission-2026-10-22-r3",
   "date": "2026-10-22",
   "dataset_repository": "kobzaond/RLVRAMBench",
-  "dataset_revision": "electronics-submission-2026-10-22-r2",
+  "dataset_revision": "electronics-submission-2026-10-22-r3",
   "dataset_main_contains_submission": true,
   "dataset_doi": "10.57967/hf/10780",
   "code_repository": "kobzaond/RLVRAMBench-reproduce",
-  "code_revision": "electronics-submission-2026-10-22-r2",
+  "code_revision": "electronics-submission-2026-10-22-r3",
   "manuscript_branch": "electronics-revision-mi250x",
   "reviewed_manuscript_base_commit": "7bc9eb9",
   "historical_reconstruction_contract": "release.json in the reproduction code repository",
@@ -66,10 +66,15 @@
   "new_gpu_experiments_during_release_preparation": false,
   "model_weights_included": false,
   "optimizer_checkpoints_included": false,
-  "supersedes_release": "electronics-submission-2026-10-22",
-  "revision_reason": "Exclude Hugging Face server-managed .gitattributes from dataset checksum manifests; scientific data and results unchanged.",
-  "code_commit": "d5c8fa79365d007098f30bdb8978b755725f5515",
-  "manuscript_commit": "3d386fa8b5fd55a02eede5ccac664866ad39eccc",
+  "supersedes_release": "electronics-submission-2026-10-22-r2",
+  "revision_reason": "Final abstract reduced to 199 words and failure-stage wording qualified; tested Hub reader versions documented. Scientific data and numerical results unchanged.",
+  "dataset_client": {
+    "python": "3.11",
+    "datasets": "5.1.0",
+    "huggingface_hub": "2.2.0"
+  },
+  "code_commit": "53acd80a09acad714d4bbe2f88edf95cb868315f",
+  "manuscript_commit": "7df8c16edd88afa22012c35d513b824f7a35bd4a",
   "verification": {
     "historical": "verification/historical-reconstruction.json",
     "a100_followups": "verification/a100-followups-reconstruction.json",
@@ -81,11 +86,17 @@
     "dataset_views": "verification/dataset-loading.json",
     "tests": "verification/reproduction-tests.json"
   },
+  "manifest_exclusions": [
+    ".gitattributes (managed automatically by Hugging Face)",
+    "MANIFEST.sha256",
+    "upload-manifest.json"
+  ],
+  "abstract_words": 199,
   "submission_files": [
     {
       "path": "paper/RLVRAMBench_MDPI.pdf",
-      "sha256": "97891c8d6e2e5b0d289bf6d192ea6fff82081065c4b768e917cbb4fadb55d926",
-      "size_bytes": 494758
+      "sha256": "4c2ca4795ca43804f2c70850ea81cc33e1ecaa7f98db552d2613e00d963d0772",
+      "size_bytes": 494860
     },
     {
       "path": "paper/RLVRAMBench_supplement.pdf",
@@ -94,18 +105,13 @@
     },
     {
       "path": "paper/RLVRAMBench_reviewer_response.pdf",
-      "sha256": "1035e92acbb579eee7767ab8ac5f1066fa9286f00b8f616c9721a019c2295fd4",
+      "sha256": "ecbfb413196c26094b9cfd53ad192caa62fb82dd92f43c56a255c0d541c8e7ee",
       "size_bytes": 115181
     },
     {
       "path": "paper/RLVRAMBench_Electronics_revision_source.zip",
-      "sha256": "335790338a773beeb74f21b53e7a8f43f20734153ae44d830cd6d5e92e32bd4a",
-      "size_bytes": 837738
+      "sha256": "bb14363f7a135b4c32aeeec1a039072858bba610fee3c6c8ecb04628f561d4d1",
+      "size_bytes": 837736
     }
-  ],
-  "manifest_exclusions": [
-    ".gitattributes (managed automatically by Hugging Face)",
-    "MANIFEST.sha256",
-    "upload-manifest.json"
   ]
 }

upload-manifest.json

@@ -1,5 +1,5 @@
 {
-  "release": "electronics-submission-2026-10-22-r2",
+  "release": "electronics-submission-2026-10-22-r3",
   "scope": "All files staged for this submission, excluding this inventory, MANIFEST.sha256, and server-managed .gitattributes; older remote release assets remain unchanged.",
   "files": [
     {
@@ -10,7 +10,7 @@
     {
       "path": "CITATION.cff",
       "size_bytes": 646,
-      "sha256": "23306037696c32d990987d6c54e235a5ca8a5e3f625951e6e9e74b321a24df33"
+      "sha256": "71ab41e15c6acdcaf099518b2813d788ceaa1054594862d65ac8a8313b4164b2"
     },
     {
       "path": "DATA_DICTIONARY.md",
@@ -24,8 +24,8 @@
     },
     {
       "path": "README.md",
-      "size_bytes": 11936,
-      "sha256": "a3090ab63fcb42bac263d40838b57872e86c55a7e46f8220eddb472ef0bf72c3"
+      "size_bytes": 12421,
+      "sha256": "d6905a34e71cb7e02dcbd26efd9db7fcb5c2c16fdf329f7a869f9df5f2f3c418"
     },
     {
       "path": "RIGHTS.md",
@@ -35,12 +35,12 @@
     {
       "path": "SUBMISSION_DATA_DICTIONARY.md",
       "size_bytes": 6716,
-      "sha256": "762d03d5ed2181b540eb3e748add96046f626e07543786e51e2aaf4422e8863d"
+      "sha256": "3b8ee3ae44b97164fb890df152eb23f0da42cf7cb7460e8ec3f88e94594cde72"
     },
     {
       "path": "SUBMISSION_REPRODUCTION.md",
-      "size_bytes": 5495,
-      "sha256": "57455520bd66f625ccb27a7d5ad4345d43acadd7f1d9f38f730dc8cd06005a40"
+      "size_bytes": 5994,
+      "sha256": "3aa978005a1f887b9100dd5c3f6b3a31fd35a63d084ed3a91e9ffee31e7207aa"
     },
     {
       "path": "amd_mi250x/README.md",
@@ -1007,6 +1007,11 @@
       "size_bytes": 40770,
       "sha256": "57cc5bff368eb5a981faaea300a097d11c9b7734031a8bf4a3a40cfef408686d"
     },
+    {
+      "path": "dataset-client-requirements.txt",
+      "size_bytes": 39,
+      "sha256": "859c54846ff064c90b60d31ba2738f97f2593e4216806f56944601403aaf2455"
+    },
     {
       "path": "electronics_revision_2026_10/MANIFEST.sha256",
       "size_bytes": 48805,
@@ -1020,7 +1025,7 @@
     {
       "path": "electronics_revision_2026_10/SUPERSEDED.md",
       "size_bytes": 266,
-      "sha256": "1810e252dba54ed0dc427e1bfb03f40a256b07f5ddd1cfae1b027e5a2d00478e"
+      "sha256": "4c9de7d8de3b629e76eec7bfb28897bcb8ba6ac0eabedc865bc93ea7b69fcdee"
     },
     {
       "path": "electronics_revision_2026_10/benchmark/attempts.csv",
@@ -3675,12 +3680,12 @@
     {
       "path": "paper/README_revision.md",
       "size_bytes": 2982,
-      "sha256": "9db9205fa3300aa9a5761e881215bf052bb58ff2ac7cbe5c6d42bec8a216d125"
+      "sha256": "691079c5f11bccd75c1137f5a0a62ee5e4ee197813f8b54c8067017731f6242c"
     },
     {
       "path": "paper/RLVRAMBench_Electronics_revision_source.zip",
-      "size_bytes": 837738,
-      "sha256": "335790338a773beeb74f21b53e7a8f43f20734153ae44d830cd6d5e92e32bd4a"
+      "size_bytes": 837736,
+      "sha256": "bb14363f7a135b4c32aeeec1a039072858bba610fee3c6c8ecb04628f561d4d1"
     },
     {
       "path": "paper/RLVRAMBench_IEEE_Access_source.zip",
@@ -3700,27 +3705,27 @@
     {
       "path": "paper/RLVRAMBench_MDPI.bbl",
       "size_bytes": 10933,
-      "sha256": "90d46065b8d84570db402c1adcc7a59c0d244945f1d928eee1e0eaa34b0b832f"
+      "sha256": "b2e5a4ccc8f83c3861224bc193b465b67e1f2af3e0b371362dba3d8548760bdd"
     },
     {
       "path": "paper/RLVRAMBench_MDPI.pdf",
-      "size_bytes": 494758,
-      "sha256": "97891c8d6e2e5b0d289bf6d192ea6fff82081065c4b768e917cbb4fadb55d926"
+      "size_bytes": 494860,
+      "sha256": "4c2ca4795ca43804f2c70850ea81cc33e1ecaa7f98db552d2613e00d963d0772"
     },
     {
       "path": "paper/RLVRAMBench_MDPI.tex",
-      "size_bytes": 159594,
-      "sha256": "751c59f46bf5aa5102fb1d99566ee4bbf2fbc2c7e43e97e1dba66144fc24a054"
+      "size_bytes": 159576,
+      "sha256": "664e8443994c7564c9ae5ed10e8c4bc808c96083f5dc86f8b78a27bb834715a4"
     },
     {
       "path": "paper/RLVRAMBench_reviewer_response.pdf",
       "size_bytes": 115181,
-      "sha256": "1035e92acbb579eee7767ab8ac5f1066fa9286f00b8f616c9721a019c2295fd4"
+      "sha256": "ecbfb413196c26094b9cfd53ad192caa62fb82dd92f43c56a255c0d541c8e7ee"
     },
     {
       "path": "paper/RLVRAMBench_reviewer_response.tex",
       "size_bytes": 23751,
-      "sha256": "c8d6bceb9785d0bf654c24543ff1e4a4a6d697d14f72163a23bd7c98b024d3b5"
+      "sha256": "b11b78eccca034a229d4577fa7f4c48d026ac251ca94543f00e2b0c74739bdd6"
     },
     {
       "path": "paper/RLVRAMBench_supplement.pdf",
@@ -3835,12 +3840,12 @@
     {
       "path": "paper/references.bib",
       "size_bytes": 18331,
-      "sha256": "97624a960a4837e14dbd03cba6a8e11e727389a6ed4bca525a0aa30672f5bcb3"
+      "sha256": "e564cecfeb04a0296dada198334adc0a9d094f81f914c3696a18a8ea8faf7a44"
     },
     {
       "path": "paper/revision_build_verification.json",
       "size_bytes": 1044,
-      "sha256": "c53ac467ae8211a113739d8027981007f334b7c988cbdafaa806819fe2ea9f5f"
+      "sha256": "51b5037ed4c7f8a873b8e0a1464d80f2953aaeaff2ee3b8bd1727efc06913f61"
     },
     {
       "path": "paper/supplement_deepspeed.tex",
@@ -4069,8 +4074,8 @@
     },
     {
       "path": "release.json",
-      "size_bytes": 4191,
-      "sha256": "cb486e40beadfd61fccd3f54d671c88be1dfd8f76bb8d24fbb923b06932da064"
+      "size_bytes": 4362,
+      "sha256": "d4aedc9c64d0ed21f67a5668bcb8cc41780089b03eeeb2440358a5b9ed77c934"
     },
     {
       "path": "releases/RLVRAMBench-mi250x-raw-20261022.tar.zst",
@@ -4550,7 +4555,7 @@
     {
       "path": "verification/paper-build.json",
       "size_bytes": 1044,
-      "sha256": "c53ac467ae8211a113739d8027981007f334b7c988cbdafaa806819fe2ea9f5f"
+      "sha256": "51b5037ed4c7f8a873b8e0a1464d80f2953aaeaff2ee3b8bd1727efc06913f61"
     },
     {
       "path": "verification/reproduction-tests.json",
@@ -4568,5 +4573,5 @@
       "sha256": "a066598a8b5565f94d6f0dcb08bb14149d0c31b7da274e6d3bf47a0768be444b"
     }
   ],
-  "manifest_sha256": "333ea1830509cf3078a2b9c0edf37fe6ff7794b54e2abd823ce002027a30b28c"
+  "manifest_sha256": "a0ad1eaefbe97244525e45c8fded5db87ecf3846cc781c5601c02dc0e6989402"
 }

verification/paper-build.json

@@ -4,14 +4,14 @@
   "extracted_pdf_text_matches": true,
   "material_build_warnings": [],
   "review_response_finalized": true,
-  "source_bundle_sha256": "335790338a773beeb74f21b53e7a8f43f20734153ae44d830cd6d5e92e32bd4a",
+  "source_bundle_sha256": "bb14363f7a135b4c32aeeec1a039072858bba610fee3c6c8ecb04628f561d4d1",
   "files": {
-    "RLVRAMBench_MDPI.tex": "751c59f46bf5aa5102fb1d99566ee4bbf2fbc2c7e43e97e1dba66144fc24a054",
-    "RLVRAMBench_MDPI.pdf": "97891c8d6e2e5b0d289bf6d192ea6fff82081065c4b768e917cbb4fadb55d926",
+    "RLVRAMBench_MDPI.tex": "664e8443994c7564c9ae5ed10e8c4bc808c96083f5dc86f8b78a27bb834715a4",
+    "RLVRAMBench_MDPI.pdf": "4c2ca4795ca43804f2c70850ea81cc33e1ecaa7f98db552d2613e00d963d0772",
     "RLVRAMBench_supplement.tex": "800544fd0de5b14d94f5324a6d4d58390c13503ee67a5dab643d5af529ca32e1",
     "RLVRAMBench_supplement.pdf": "0a389b5b6c0a51c20796312592e45df2d87af1629c2010baed20ec4e0832dcbb",
-    "RLVRAMBench_reviewer_response.pdf": "1035e92acbb579eee7767ab8ac5f1066fa9286f00b8f616c9721a019c2295fd4",
-    "electronics_reviewer_response.md": "1f4f9fd070bcc617215823f5ec5576d8844a202f84c2d986b236699718d5c476"
+    "RLVRAMBench_reviewer_response.pdf": "ecbfb413196c26094b9cfd53ad192caa62fb82dd92f43c56a255c0d541c8e7ee",
+    "electronics_reviewer_response.md": "2f83fe4640ba9e40afb18dd282cd5c834381cb94974c70659a0f972e80379cd3"
   },
   "tex_versions": {
     "pdflatex": "pdfTeX 3.141592653-2.6-1.40.29 (TeX Live 2026)",