sciev 0.2.1__tar.gz

sciev-0.2.1/LICENSE

@@ -0,0 +1,186 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to patents and to the extent such
+      patentable contributions are licensed hereunder, provided that the
+      Work done by such Contributor is not otherwise restricted under the
+      License.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the documentation for
+          the Derivative Works, if provided along with the Derivative Works;
+          or, within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for Your modifications.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work by
+      You shall be subject to the same rights and obligations as the
+      copyright holder to the original owner of the Work, and any Contribution
+      submitted for inclusion in the Work by You shall be deemed to be free
+      of additional terms or conditions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing and you assume all risks
+      associated with use of Your Licensed Applications.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable for any damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all other
+      commercial damages or losses), even if such Contributor has been
+      advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or other warranty rights.
+      However, in accepting such obligations, You may act only on Your
+      own behalf and on Your sole responsibility, not on behalf of any
+      other Contributor, and only if You agree to indemnify, defend,
+      and hold each Contributor harmless for any liability or expenses
+      incurred by the use of Your rights under this License.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 alrobles
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

sciev-0.2.1/PKG-INFO

@@ -0,0 +1,196 @@
+Metadata-Version: 2.4
+Name: sciev
+Version: 0.2.1
+Summary: Typed decision heads on frozen masked-diffusion backbones with held-out calibration.
+Author: Angel Robles
+License: Apache-2.0
+Project-URL: Homepage, https://github.com/alrobles/sciev-devel
+Project-URL: Repository, https://github.com/alrobles/sciev-devel
+Project-URL: Issues, https://github.com/alrobles/sciev-devel/issues
+Project-URL: Releases, https://github.com/alrobles/sciev-devel/releases
+Keywords: scientific-computing,decision-models,masked-diffusion,calibration,classification
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: torch>=2.0.0
+Requires-Dist: transformers>=4.45.0
+Requires-Dist: numpy>=1.20.0
+Requires-Dist: pyyaml>=6.0
+Provides-Extra: serve
+Requires-Dist: fastapi>=0.100; extra == "serve"
+Requires-Dist: uvicorn>=0.23; extra == "serve"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Dynamic: license-file
+
+# Sciev
+
+**Sciev** (`sci`·ence + `ev`·aluation) — open **System-One-style
+decision models** on a frozen masked-diffusion backbone. Typed
+questions (`choice` / `noul` / `score`) over a `state`, answered with
+probabilities in **one forward pass**, calibrated with held-out dev data —
+no text generation or output parsing. Classification errors remain possible.
+
+*(repo: `alrobles/sciev-devel`; `pip install sciev`)*
+
+General-purpose by design, with a scientific specialty: the heads are
+trained on passage-grounded scientific decisions and transfer zero-shot
+to general classification.
+
+Sciev uses a **masked-diffusion language model** (LLaDA-8B) and canonical
+option sorting. Distinct option token sequences produce the same model
+input under permutation; duplicate or truncated-to-identical options need
+an explicit tie policy if their external identifiers must be distinguished.
+
+## Results (v0.2, `systemone-v2` encoding, 3 seeds)
+
+| eval | type | fr (release) mean ± sd | da (DAPT, experimental) |
+|---|---|---:|---:|
+| sci battery elite | choice | **0.970 ± 0.017** | 0.961 ± 0.011 |
+| sci battery elite | noul | **0.925 ± 0.006** | 0.913 ± 0.007 |
+| sci battery elite | score | **0.859 ± 0.075** | 0.779 ± 0.109 |
+| GPQA main | choice | 0.295 ± 0.014 | 0.288 ± 0.018 |
+| GPQA diamond | choice | 0.307 ± 0.012 | 0.297 ± 0.011 |
+| SciFact dev | noul | 0.483 ± 0.023 | **0.709 ± 0.020** |
+| SciFact dev | choice | 0.479 ± 0.018 | 0.444 ± 0.062 |
+
+Under matched seeds, recipe and data, DAPT helps out-of-domain
+verification (SciFact noul +22.6pt) but degrades in-domain ordinal
+scoring — task-dependent, not uniform. Evidence controls (empty /
+shuffled passages) show `choice` retains option-side leakage
+(agreement-with-reference ~0.80 without evidence) while `score`
+collapses toward chance; see the paper and `manifests/` in the release
+assets. Earlier v0.1.1 zero-shot numbers (SST-2 0.93, AG News 0.85)
+remain archived in the release JSONs.
+
+| benchmark (archived v0.1.x, `c_*` heads) | type | n | acc | ECE | auto@5%err |
+|---|---|---:|---:|---:|---:|
+| sci battery (elite) | choice | 653 | **0.8698** | 0.0253 | **0.8070** |
+| sci battery (elite) | noul | 1959 | 0.7825 | 0.0957 | 0.2777 |
+| sci battery (elite) | score | 1959 | 0.6075 | 0.1221 | 0.4334 |
+| SciFact dev | noul | 340 | **0.8500** | 0.0998 | 0.2706 |
+| SciFact dev | score | 340 | 0.6294 | 0.0796 | 0.2559 |
+| GPQA main | choice | 448 | 0.3125 | 0.3180 | 0.0112 |
+| GPQA diamond | choice | 198 | 0.3182 | 0.3089 | 0.0152 |
+| SST-2 (zero-shot) | choice | 872 | **0.9300** | 0.0682 | **0.9415** |
+| AG News (zero-shot) | choice | 7600 | **0.8537** | 0.0556 | 0.3674 |
+| Enron spam (zero-shot) | choice | 2000 | 0.7510 | 0.1046 | 0.0000 |
+| Banking77 (K=77) | choice | 3080 | 0.2289 | 0.0809 | 0.0195 |
+
+The archived v0.1.x table is kept for provenance; `paper/results.json`
+records source hashes, metrics, and training settings. Third-party
+scores with unverified protocol equivalence are not included as direct
+comparisons.
+
+`auto@5%err` is the largest confidence-ranked prefix with realized error
+≤ 5% on labeled evaluation data, not a deployment error guarantee.
+v0.2 fits temperatures on dev data only, and calibration artifacts are
+bound to the checkpoint SHA-256.
+
+Known limits: weak K>4 performance (Banking77, v0.1.x), low GPQA
+accuracy, weaker elite score than choice, and the option-side leakage
+quantified by the evidence controls.
+
+## Architecture
+
+```
+ctx + question + options (canonically sorted by token-ids)
+  -> frozen LLaDA-8B (bidirectional masked-diffusion LM)
+  -> hidden states, layers {8,16,24,32}
+  -> specialist head per decision type
+       choice: AttnPoolHead — 67.2M parameters, four-layer span pooling
+       noul:   MLP — 16.8M parameters, final-layer mean-pooled span
+       score:  MLP — 16.8M parameters + ordinal auxiliary loss (P(y>=j))
+  -> per-option logits -> softmax (temperature fitted on dev)
+```
+
+Checkpoint sizes are 269.4 MB (choice) and 68.6 MB each (noul, score),
+measured from the serialized heads. The backbone stays frozen and is
+fetched separately from HF. No DAPT adapter is required for the release.
+
+## Quickstart
+
+Install the package, download the release heads, and evaluate a prepared
+decision JSONL (the example assumes `my_decisions.jsonl` already exists):
+
+```bash
+pip install sciev
+gh release download v0.2.1 --repo alrobles/sciev-devel \
+    --pattern 'fr_*.pt' --dir release
+python -m sciev.eval \
+    --ckpt release/fr_choice.pt --decision-type choice \
+    --r2-mode spanpool --r2-layers=-1,-9,-17,-25 --canonical-order \
+    --r2-temp-fit-decisions my_decisions_dev.jsonl \
+    --decisions-eval my_decisions.jsonl \
+    --device cuda --out eval.json
+```
+
+Temperatures are fitted on dev splits only (`--r2-temp-fit-decisions`);
+the calibration artifact is bound to the checkpoint hash. Pass an
+explicit `--r2-temp` to reuse a recorded value instead.
+
+Decision records are `{ctx: [token-ids], opts: [[token-ids]...], gold}`
+(`systemone-v2` also accepts text rows; see `sciev/decisions.py`).
+Converters for GPQA / SciFact / SST-2 / AG News / Enron / Banking77 live
+in `data/convert_benchmarks.py`; the scientific battery builder is
+`data/build_sci_decisions.py`.
+
+## Domain adaptation (DAPT)
+
+`sciev/dapt.py` implements an experimental masked-token LoRA
+adaptation loop for HF backbones (requires PEFT). Under the matched
+comparison (same recipe, seeds and `systemone-v2` data), the adapted
+backbone improves out-of-domain verification (SciFact noul +22.6pt) but
+degrades in-domain ordinal scoring — the released heads therefore stay
+frozen (`fr_*`), with `da_*` published as an experimental arm. See the
+matched table and protocol caveats in the paper.
+
+```bash
+python -m sciev.dapt \
+    --hf-backbone GSAI-ML/LLaDA-8B-Instruct \
+    --corpus papers.jsonl --field text \
+    --out runs/dapt --steps 6000 --bs 16 --seq-len 1024
+# then retrain heads with --lora-adapter runs/dapt/lora-final
+```
+
+`--resume` loads adapter weights and `--start-step` skips batches and
+advances the schedule; it does **not** restore optimizer or RNG state.
+It is a warm start, not an exact training-state resume. The DAPT module
+accepts `--hf-backbone`, not `--ckpt`; native MdLMMoE checkpoints can be
+used separately with `sciev.train --ckpt`.
+
+## Repo map
+
+| path | qué |
+|---|---|
+| `sciev/model.py` | MdLMMoE + HFBackbone + heads + canonical layout |
+| `sciev/train.py` | head training (CE, ordinal, RL-lite) |
+| `sciev/eval.py` | acc/ECE/flip/auto@5%, temp-fit, remote API eval |
+| `sciev/dapt.py` | domain-adaptive pretraining (LoRA) |
+| `data/` | battery builders + benchmark converters |
+| `scripts/*.slurm` | reproducible jobs (KU HPC) |
+| `docs/PROJECT-STATUS.md` | roadmap, milestones, design decisions |
+| `paper/main.tex` | arXiv draft |
+
+## Verification
+
+```bash
+OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 python -m pytest -q
+```
+
+The suite includes option-identity flip regressions and checks that paper
+Tables 3 and 4 agree with `paper/results.json`. Compile the paper twice
+with `pdflatex -interaction=nonstopmode -halt-on-error main.tex` from `paper/`.
+
+## License
+
+Apache-2.0 (code + heads). Backbones and datasets keep their own
+licenses.

sciev-0.2.1/README.md

@@ -0,0 +1,163 @@
+# Sciev
+
+**Sciev** (`sci`·ence + `ev`·aluation) — open **System-One-style
+decision models** on a frozen masked-diffusion backbone. Typed
+questions (`choice` / `noul` / `score`) over a `state`, answered with
+probabilities in **one forward pass**, calibrated with held-out dev data —
+no text generation or output parsing. Classification errors remain possible.
+
+*(repo: `alrobles/sciev-devel`; `pip install sciev`)*
+
+General-purpose by design, with a scientific specialty: the heads are
+trained on passage-grounded scientific decisions and transfer zero-shot
+to general classification.
+
+Sciev uses a **masked-diffusion language model** (LLaDA-8B) and canonical
+option sorting. Distinct option token sequences produce the same model
+input under permutation; duplicate or truncated-to-identical options need
+an explicit tie policy if their external identifiers must be distinguished.
+
+## Results (v0.2, `systemone-v2` encoding, 3 seeds)
+
+| eval | type | fr (release) mean ± sd | da (DAPT, experimental) |
+|---|---|---:|---:|
+| sci battery elite | choice | **0.970 ± 0.017** | 0.961 ± 0.011 |
+| sci battery elite | noul | **0.925 ± 0.006** | 0.913 ± 0.007 |
+| sci battery elite | score | **0.859 ± 0.075** | 0.779 ± 0.109 |
+| GPQA main | choice | 0.295 ± 0.014 | 0.288 ± 0.018 |
+| GPQA diamond | choice | 0.307 ± 0.012 | 0.297 ± 0.011 |
+| SciFact dev | noul | 0.483 ± 0.023 | **0.709 ± 0.020** |
+| SciFact dev | choice | 0.479 ± 0.018 | 0.444 ± 0.062 |
+
+Under matched seeds, recipe and data, DAPT helps out-of-domain
+verification (SciFact noul +22.6pt) but degrades in-domain ordinal
+scoring — task-dependent, not uniform. Evidence controls (empty /
+shuffled passages) show `choice` retains option-side leakage
+(agreement-with-reference ~0.80 without evidence) while `score`
+collapses toward chance; see the paper and `manifests/` in the release
+assets. Earlier v0.1.1 zero-shot numbers (SST-2 0.93, AG News 0.85)
+remain archived in the release JSONs.
+
+| benchmark (archived v0.1.x, `c_*` heads) | type | n | acc | ECE | auto@5%err |
+|---|---|---:|---:|---:|---:|
+| sci battery (elite) | choice | 653 | **0.8698** | 0.0253 | **0.8070** |
+| sci battery (elite) | noul | 1959 | 0.7825 | 0.0957 | 0.2777 |
+| sci battery (elite) | score | 1959 | 0.6075 | 0.1221 | 0.4334 |
+| SciFact dev | noul | 340 | **0.8500** | 0.0998 | 0.2706 |
+| SciFact dev | score | 340 | 0.6294 | 0.0796 | 0.2559 |
+| GPQA main | choice | 448 | 0.3125 | 0.3180 | 0.0112 |
+| GPQA diamond | choice | 198 | 0.3182 | 0.3089 | 0.0152 |
+| SST-2 (zero-shot) | choice | 872 | **0.9300** | 0.0682 | **0.9415** |
+| AG News (zero-shot) | choice | 7600 | **0.8537** | 0.0556 | 0.3674 |
+| Enron spam (zero-shot) | choice | 2000 | 0.7510 | 0.1046 | 0.0000 |
+| Banking77 (K=77) | choice | 3080 | 0.2289 | 0.0809 | 0.0195 |
+
+The archived v0.1.x table is kept for provenance; `paper/results.json`
+records source hashes, metrics, and training settings. Third-party
+scores with unverified protocol equivalence are not included as direct
+comparisons.
+
+`auto@5%err` is the largest confidence-ranked prefix with realized error
+≤ 5% on labeled evaluation data, not a deployment error guarantee.
+v0.2 fits temperatures on dev data only, and calibration artifacts are
+bound to the checkpoint SHA-256.
+
+Known limits: weak K>4 performance (Banking77, v0.1.x), low GPQA
+accuracy, weaker elite score than choice, and the option-side leakage
+quantified by the evidence controls.
+
+## Architecture
+
+```
+ctx + question + options (canonically sorted by token-ids)
+  -> frozen LLaDA-8B (bidirectional masked-diffusion LM)
+  -> hidden states, layers {8,16,24,32}
+  -> specialist head per decision type
+       choice: AttnPoolHead — 67.2M parameters, four-layer span pooling
+       noul:   MLP — 16.8M parameters, final-layer mean-pooled span
+       score:  MLP — 16.8M parameters + ordinal auxiliary loss (P(y>=j))
+  -> per-option logits -> softmax (temperature fitted on dev)
+```
+
+Checkpoint sizes are 269.4 MB (choice) and 68.6 MB each (noul, score),
+measured from the serialized heads. The backbone stays frozen and is
+fetched separately from HF. No DAPT adapter is required for the release.
+
+## Quickstart
+
+Install the package, download the release heads, and evaluate a prepared
+decision JSONL (the example assumes `my_decisions.jsonl` already exists):
+
+```bash
+pip install sciev
+gh release download v0.2.1 --repo alrobles/sciev-devel \
+    --pattern 'fr_*.pt' --dir release
+python -m sciev.eval \
+    --ckpt release/fr_choice.pt --decision-type choice \
+    --r2-mode spanpool --r2-layers=-1,-9,-17,-25 --canonical-order \
+    --r2-temp-fit-decisions my_decisions_dev.jsonl \
+    --decisions-eval my_decisions.jsonl \
+    --device cuda --out eval.json
+```
+
+Temperatures are fitted on dev splits only (`--r2-temp-fit-decisions`);
+the calibration artifact is bound to the checkpoint hash. Pass an
+explicit `--r2-temp` to reuse a recorded value instead.
+
+Decision records are `{ctx: [token-ids], opts: [[token-ids]...], gold}`
+(`systemone-v2` also accepts text rows; see `sciev/decisions.py`).
+Converters for GPQA / SciFact / SST-2 / AG News / Enron / Banking77 live
+in `data/convert_benchmarks.py`; the scientific battery builder is
+`data/build_sci_decisions.py`.
+
+## Domain adaptation (DAPT)
+
+`sciev/dapt.py` implements an experimental masked-token LoRA
+adaptation loop for HF backbones (requires PEFT). Under the matched
+comparison (same recipe, seeds and `systemone-v2` data), the adapted
+backbone improves out-of-domain verification (SciFact noul +22.6pt) but
+degrades in-domain ordinal scoring — the released heads therefore stay
+frozen (`fr_*`), with `da_*` published as an experimental arm. See the
+matched table and protocol caveats in the paper.
+
+```bash
+python -m sciev.dapt \
+    --hf-backbone GSAI-ML/LLaDA-8B-Instruct \
+    --corpus papers.jsonl --field text \
+    --out runs/dapt --steps 6000 --bs 16 --seq-len 1024
+# then retrain heads with --lora-adapter runs/dapt/lora-final
+```
+
+`--resume` loads adapter weights and `--start-step` skips batches and
+advances the schedule; it does **not** restore optimizer or RNG state.
+It is a warm start, not an exact training-state resume. The DAPT module
+accepts `--hf-backbone`, not `--ckpt`; native MdLMMoE checkpoints can be
+used separately with `sciev.train --ckpt`.
+
+## Repo map
+
+| path | qué |
+|---|---|
+| `sciev/model.py` | MdLMMoE + HFBackbone + heads + canonical layout |
+| `sciev/train.py` | head training (CE, ordinal, RL-lite) |
+| `sciev/eval.py` | acc/ECE/flip/auto@5%, temp-fit, remote API eval |
+| `sciev/dapt.py` | domain-adaptive pretraining (LoRA) |
+| `data/` | battery builders + benchmark converters |
+| `scripts/*.slurm` | reproducible jobs (KU HPC) |
+| `docs/PROJECT-STATUS.md` | roadmap, milestones, design decisions |
+| `paper/main.tex` | arXiv draft |
+
+## Verification
+
+```bash
+OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 python -m pytest -q
+```
+
+The suite includes option-identity flip regressions and checks that paper
+Tables 3 and 4 agree with `paper/results.json`. Compile the paper twice
+with `pdflatex -interaction=nonstopmode -halt-on-error main.tex` from `paper/`.
+
+## License
+
+Apache-2.0 (code + heads). Backbones and datasets keep their own
+licenses.

sciev-0.2.1/pyproject.toml

@@ -0,0 +1,50 @@
+[project]
+name = "sciev"
+version = "0.2.1"
+description = "Typed decision heads on frozen masked-diffusion backbones with held-out calibration."
+readme = "README.md"
+requires-python = ">=3.10"
+license = { text = "Apache-2.0" }
+authors = [
+    { name = "Angel Robles" },
+]
+keywords = [
+    "scientific-computing",
+    "decision-models",
+    "masked-diffusion",
+    "calibration",
+    "classification",
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Science/Research",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = [
+    "torch>=2.0.0",
+    "transformers>=4.45.0",
+    "numpy>=1.20.0",
+    "pyyaml>=6.0",
+]
+
+[project.optional-dependencies]
+serve = ["fastapi>=0.100", "uvicorn>=0.23"]
+dev = ["pytest>=7.0"]
+
+[project.urls]
+Homepage = "https://github.com/alrobles/sciev-devel"
+Repository = "https://github.com/alrobles/sciev-devel"
+Issues = "https://github.com/alrobles/sciev-devel/issues"
+Releases = "https://github.com/alrobles/sciev-devel/releases"
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools]
+packages = ["sciev"]

sciev-0.2.1/sciev/__init__.py

@@ -0,0 +1,6 @@
+"""sciev — open System-One-style decision models on the EcoReasoner dLLM.
+
+R1 token-slot readout on ecoreasoner MdLMMoE checkpoints: state + question +
+[MASK] -> calibrated distribution over typed options (noul/choice/score).
+"""
+__version__ = "0.2.1"