nano-jev 0.1.0__tar.gz

nano_jev-0.1.0/.gitignore

@@ -0,0 +1,18 @@
+# Python
+__pycache__/
+*.pyc
+.venv/
+
+# Model outputs (weights go to Hugging Face, not git) and smoke-test scratch data
+runs/
+data_smoke/
+
+# Editors / OS
+.vscode/
+.idea/
+Thumbs.db
+
+# Packaging / tests
+dist/
+*.egg-info/
+.pytest_cache/

nano_jev-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 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 those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   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 Source form or
+          documentation, 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 attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   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 the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   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 to You for 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 rights consistent with this
+      License. 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
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   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 [yyyy] [name of copyright owner]
+
+   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.

nano_jev-0.1.0/PKG-INFO

@@ -0,0 +1,310 @@
+Metadata-Version: 2.4
+Name: nano-jev
+Version: 0.1.0
+Summary: Tiny calibrated typed-decision model for RAG pipelines: relevance, sufficiency, groundedness
+Project-URL: Homepage, https://github.com/shubham10divakar/nano-jev
+Project-URL: Repository, https://github.com/shubham10divakar/nano-jev
+Project-URL: Issues, https://github.com/shubham10divakar/nano-jev/issues
+Project-URL: Model, https://huggingface.co/sdmlai/nano-jev
+Author: Subham
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: calibration,decision-model,jev,llm,rag,reranker,retrieval
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Requires-Dist: huggingface-hub>=0.24
+Requires-Dist: torch>=2.1
+Requires-Dist: transformers>=4.45
+Provides-Extra: test
+Requires-Dist: pytest>=8; extra == 'test'
+Provides-Extra: train
+Requires-Dist: accelerate; extra == 'train'
+Requires-Dist: datasets>=3.0; extra == 'train'
+Requires-Dist: numpy; extra == 'train'
+Requires-Dist: scikit-learn; extra == 'train'
+Requires-Dist: tqdm; extra == 'train'
+Description-Content-Type: text/markdown
+
+# Nano-Jev
+
+[![PyPI version](https://img.shields.io/pypi/v/nano-jev?label=PyPI&color=blue)](https://pypi.org/project/nano-jev/)
+[![Python](https://img.shields.io/pypi/pyversions/nano-jev)](https://pypi.org/project/nano-jev/)
+[![Downloads](https://static.pepy.tech/badge/nano-jev)](https://pepy.tech/project/nano-jev)
+[![Monthly downloads](https://img.shields.io/pypi/dm/nano-jev?label=downloads%2Fmonth)](https://pypistats.org/packages/nano-jev)
+[![Model version](https://img.shields.io/badge/model-v0.1-yellow?logo=huggingface)](https://huggingface.co/sdmlai/nano-jev/tree/v0.1)
+[![HF downloads](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fhuggingface.co%2Fapi%2Fmodels%2Fsdmlai%2Fnano-jev&query=%24.downloads&label=HF%20downloads&logo=huggingface&color=yellow)](https://huggingface.co/sdmlai/nano-jev)
+[![License](https://img.shields.io/badge/license-Apache--2.0-green)](https://github.com/shubham10divakar/nano-jev/blob/main/LICENSE)
+[![Status](https://img.shields.io/badge/status-research%20preview-orange)](#limitations)
+[![Open in Kaggle](https://kaggle.com/static/images/open-in-kaggle.svg)](https://kaggle.com/kernels/welcome?src=https://github.com/shubham10divakar/nano-jev/blob/main/examples/kaggle_smoke_test.ipynb)
+
+> Jev-style decision model. Independent; not affiliated with TypeSafe AI or the NanoJev
+> GitHub project.
+
+**A tiny (22.7M parameter) calibrated decision model for RAG pipelines.** Give it a
+question, some options and some context; it returns a **probability for each option** in
+under a millisecond on a GPU (a few ms on CPU). It never generates text, so it can't
+return a malformed answer.
+
+Use it for the small decisions an agentic RAG loop makes over and over:
+
+| Decision | Options | Asks |
+|---|---|---|
+| `relevance` | irrelevant / partially relevant / directly answers | Is this passage useful for the query? |
+| `sufficient` | yes / no | Do these passages contain enough to answer? |
+| `grounded` | yes / no | Is this claim supported by the context? |
+| `decide` | **your own options** | Anything else, e.g. topic or route |
+
+## At a glance
+
+| | |
+|---|---|
+| Package | `nano-jev` 0.1.0 on [PyPI](https://pypi.org/project/nano-jev/) (import `nanojev`, command `nano-jev`) |
+| Weights | [`sdmlai/nano-jev`](https://huggingface.co/sdmlai/nano-jev), version `v0.1` (default), downloaded automatically on first use |
+| Model | MiniLM cross-encoder, 6 layers, hidden size 384, **22.7M parameters**, max 512 tokens |
+| Accuracy (test split) | relevance **0.798**, sufficient **0.759**, grounded **0.804** ([results](#results)) |
+| Calibration error (ECE) | 0.020 / 0.019 / 0.045 after temperature scaling |
+| Speed | ~0.4–2.3 ms per decision batched on an RTX 3060; runs on CPU |
+| Python | 3.10+ · PyTorch 2.1+ · Transformers 4.45+ |
+| Licence | Apache-2.0 (see the [MS MARCO note](#licence)) |
+| Status | **v0.1 research preview**; v1.0 planned |
+
+## Install
+
+```bash
+pip install nano-jev
+```
+
+PyTorch is installed from PyPI if you don't have it. For a CUDA build, install PyTorch
+first following [pytorch.org](https://pytorch.org/get-started/locally/).
+
+## Quick start
+
+```python
+import nanojev
+
+d = nanojev.load()   # default weights (v0.1); downloads ~90 MB the first time, then cached
+
+d.relevance("When was UCL founded?", [
+    "University College London was founded in 1826.",
+    "The Analytical Engine was a proposed mechanical computer.",
+])
+# [{'irrelevant': 0.021, 'partially relevant': 0.041, 'directly answers': 0.938},
+#  {'irrelevant': 0.822, 'partially relevant': 0.143, 'directly answers': 0.035}]
+
+d.sufficient("In which year was the university attended by Ada Lovelace's tutor founded?", [
+    "Augustus De Morgan tutored Ada Lovelace. He was a professor at University College London.",
+    "University College London was founded in 1826.",
+])
+# {'yes': 0.899, 'no': 0.101}      (with only the first passage: {'yes': 0.055, 'no': 0.945})
+
+d.grounded("UCL was founded in 1900.", "University College London was founded in 1826.")
+# {'yes': 0.051, 'no': 0.949}
+
+d.decide("Which topic is this passage about?", ["computing", "cooking", "football"],
+         "The Analytical Engine was a proposed mechanical computer.")
+# {'computing': 0.783, 'cooking': 0.143, 'football': 0.074}
+```
+
+### In a RAG loop
+
+```python
+d = nanojev.load()
+
+def answer(query, retrieve, llm):
+    passages = retrieve(query, k=10)
+    scores = d.relevance(query, passages)
+    kept = [p for p, s in zip(passages, scores) if s["irrelevant"] < 0.5]
+
+    if d.sufficient(query, kept)["yes"] < 0.85:          # not enough context
+        kept += retrieve(query, k=10, page=2)            # retrieve more / rewrite the query / abstain
+
+    reply = llm(query, kept)
+    if d.grounded(reply, "\n".join(kept))["yes"] < 0.5:  # unsupported answer
+        reply = llm(query, kept, strict=True)            # regenerate, flag, or escalate
+    return reply
+```
+
+The thresholds are examples. Calibrated probabilities make thresholds meaningful, but
+**check them on your own data**.
+
+## Choosing weights
+
+Weights live on Hugging Face at [`sdmlai/nano-jev`](https://huggingface.co/sdmlai/nano-jev),
+one git tag per version. A version downloads the first time you use it and then loads
+from the local cache (works offline).
+
+```python
+import nanojev
+
+models, online = nanojev.list_models()           # versions on the Hub (+ which are downloaded)
+for m in models:
+    print(m.name, m.status, m.params, "downloaded" if m.downloaded else "")
+
+d = nanojev.load("v0.1")                          # a specific version
+d = nanojev.load("sdmlai/nano-jev@v0.1")          # any Hub repo, optionally @tag
+d = nanojev.load("path/to/my-trained-model")      # a local folder
+print(d.version, d.config["base"])
+```
+
+Which weights `nanojev.load()` uses when you don't say:
+
+1. the `NANOJEV_MODEL` environment variable
+2. the choice saved with `nano-jev use ...` (stored in `~/.nanojev/config.json`)
+3. the package default, `v0.1` (pinned per package release, so results stay reproducible)
+
+## Command line
+
+Installing the package adds a `nano-jev` command (`python -m nanojev` does the same).
+
+```bash
+nano-jev --version
+nano-jev list                         # available weights; * marks the selected one
+nano-jev list --local runs            # also list your own trained folders under runs/
+nano-jev download v0.1                # fetch ahead of time, e.g. before going offline
+nano-jev use v0.1                     # choose default weights: a version, Hub repo[@tag] or folder
+nano-jev use --reset                  # back to the package default
+nano-jev current                      # what's selected and where it is on disk
+
+nano-jev relevance -q "When was UCL founded?" -p "University College London was founded in 1826." -p "Another passage."
+nano-jev sufficient -q "QUERY" -p "PASSAGE 1" -p "PASSAGE 2"
+nano-jev grounded --claim "UCL was founded in 1900." --context "University College London was founded in 1826."
+nano-jev decide --question "Which topic is this passage about?" -o computing -o cooking --state "TEXT"
+```
+
+Decision commands accept `--model` (override the selection for one call) and `--json`.
+
+```
+$ nano-jev list
+   MODEL  SOURCE  STATUS            BASE                                 PARAMS  RELEASED    DOWNLOADED
+*  v0.1   hub     research preview  cross-encoder/ms-marco-MiniLM-L6-v2  22.7M   2026-09-25  yes
+
+selected: v0.1  (from default)
+
+$ nano-jev grounded --claim "UCL was founded in 1900." --context "University College London was founded in 1826."
+yes=0.051  no=0.949
+```
+
+## How it works
+
+Each (question, option, context) triple goes through a small cross-encoder that outputs
+one logit:
+
+```
+[CLS] question: <q> option: <opt> [SEP] <context> [SEP]  → encoder → linear → logit
+```
+
+A decision's option logits are divided by that decision's temperature (fitted on
+held-out data) and softmaxed, so the probabilities are calibrated. Options are scored
+independently, which is why you can pass option sets the model never saw in training.
+
+## Results
+
+Test split = held-out halves of the HotpotQA / SQuAD 2.0 / MultiNLI validation sets.
+Every model is temperature-calibrated on the same calibration split. Accuracy:
+
+| Decision | **Nano-Jev v0.1** (22.7M) | bge-reranker-v2-m3 (568M) | Qwen3-4B-Instruct, prompted (4B) | untrained base |
+|---|---|---|---|---|
+| relevance (3-way) | **0.798** | 0.694 | 0.660 | 0.273 |
+| sufficient | **0.759** | 0.692 | 0.706 | 0.503 |
+| grounded | 0.804 | 0.798 | **0.844** | 0.540 |
+
+Nano-Jev is ~90–165× faster than the prompted LLM. The baselines are zero-shot on these
+datasets and Nano-Jev is not, so read the caveats in
+[results/v0.1_comparison.md](https://github.com/shubham10divakar/nano-jev/blob/main/results/v0.1_comparison.md).
+
+## Tests
+
+```bash
+git clone https://github.com/shubham10divakar/nano-jev
+cd nano-jev
+pip install -e ".[test,train]"
+
+pytest                                   # offline: builds a tiny random model, ~10 s
+NANOJEV_NETWORK_TESTS=1 pytest           # also downloads the real v0.1 weights and checks them
+```
+
+What the tests cover (see [`tests/`](https://github.com/shubham10divakar/nano-jev/tree/main/tests)):
+
+| File | Examples of what's checked |
+|---|---|
+| `test_decider.py` | every decision returns a probability distribution over its options; option order doesn't change probabilities; temperatures and config are loaded |
+| `test_registry.py` | version names, `repo@tag` parsing, selection order (env > saved > default), local folders |
+| `test_cli.py` | `--version`, `use` / `current` / `use --reset`, `relevance` and `decide` with `--json`, offline `list` |
+| `test_calibration.py` | temperature fitting recovers a known temperature; ECE of an overconfident model |
+| `test_network.py` | (opt-in) v0.1 is listed on the Hub, downloads, and makes sensible decisions |
+
+### Smoke test on Kaggle / Colab / a fresh machine
+
+- **Kaggle:** click the *Open in Kaggle* badge above (or upload
+  [`examples/kaggle_smoke_test.ipynb`](https://github.com/shubham10divakar/nano-jev/blob/main/examples/kaggle_smoke_test.ipynb)),
+  turn on *Settings → Internet*, then *Run All*.
+- **Anywhere:**
+  ```bash
+  pip install nano-jev
+  wget https://raw.githubusercontent.com/shubham10divakar/nano-jev/main/examples/smoke_test.py
+  python smoke_test.py        # 9 checks, ends with ALL SMOKE TESTS PASSED
+  ```
+
+## Train your own
+
+The repo also contains the full training pipeline (≈15 min on an RTX 3060):
+
+```bash
+pip install -e ".[train]"
+python scripts/prepare_data.py --preset default   # HotpotQA + SQuAD 2.0 + MultiNLI -> data/ (v0.1 splits are included)
+python scripts/train.py                           # -> runs/nano-jev-dev
+python scripts/evaluate.py --model runs/nano-jev-dev
+python scripts/baselines.py --which all           # bge-reranker + prompted Qwen3-4B (~35 min)
+nano-jev use runs/nano-jev-dev                     # use your model everywhere
+```
+
+`--preset smoke` gives a 1-minute end-to-end run. Layout:
+
+```
+nanojev/        the package: decider (API), registry (weights), model, schema,
+                calibration, report, data, __main__ (CLI)
+scripts/        prepare_data, train, evaluate, baselines, demo
+tests/          pytest suite
+examples/       smoke_test.py, kaggle_smoke_test.ipynb
+data/           v0.1 train / calib / test splits (JSONL)
+results/        evaluation tables and baseline comparison
+model_cards/    Hugging Face model card per released version
+```
+
+## Limitations
+
+- **In-distribution evaluation only (v0.1).** Trained on the train splits of the same three
+  datasets it is tested on; the baselines are zero-shot. No out-of-distribution results yet.
+- **Groundedness** is below a prompted 4B LLM. Consider escalating low-confidence
+  `grounded` calls to a larger model.
+- **English, Wikipedia-style text.** Expect worse behaviour on other domains and languages.
+- **Custom option sets** work but aren't trained for; treat those probabilities as rough.
+- "partially relevant" follows HotpotQA rules: a supporting paragraph without the answer string.
+
+## Licence
+
+Code and weights: **Apache-2.0**.
+
+The v0.1 weights start from `cross-encoder/ms-marco-MiniLM-L6-v2`, which was trained on
+MS MARCO, whose terms are **non-commercial**. Commercial users should check those terms, or
+wait for **v1.0**, which will be retrained on an MIT-licensed base with no MS MARCO lineage.
+
+Training and evaluation data derive from [HotpotQA](https://hotpotqa.github.io/) (CC BY-SA 4.0),
+[SQuAD 2.0](https://rajpurkar.github.io/SQuAD-explorer/) (CC BY-SA 4.0) and
+[MultiNLI](https://cims.nyu.edu/~sbowman/multinli/); the files in `data/` carry those licences.
+
+## Roadmap
+
+- **v1.0:** MIT-licensed base, held-out evaluation (MuSiQue, FEVER / RAGTruth), a
+  grounded-decision cascade that escalates unsure cases to an LLM, stable API.
+- Later: a larger model that reads all chunks in one pass, and a `next_action` decision
+  for agentic RAG loops.