idealens 0.1.0__tar.gz

idealens-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 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.

idealens-0.1.0/PKG-INFO

@@ -0,0 +1,243 @@
+Metadata-Version: 2.4
+Name: idealens
+Version: 0.1.0
+Summary: Detect whether a document's ideas came from a person or an AI model (IdeaLens), and related detectors.
+Author: Rishanth Rajendhran
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/RishanthRajendhran/IdeaLens
+Project-URL: Source, https://github.com/RishanthRajendhran/IdeaLens
+Project-URL: Model, https://huggingface.co/rishanthrajendhran/IdeaLens
+Project-URL: Corpus, https://huggingface.co/datasets/rishanthrajendhran/WildOutlines
+Keywords: ai-text-detection,idea-provenance,machine-generated-text
+Classifier: Programming Language :: Python :: 3
+Classifier: Intended Audience :: Science/Research
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.24
+Requires-Dist: huggingface_hub>=0.30
+Provides-Extra: hf
+Requires-Dist: torch>=2.4; extra == "hf"
+Requires-Dist: transformers>=5.15; extra == "hf"
+Requires-Dist: safetensors>=0.4; extra == "hf"
+Requires-Dist: accelerate>=1.0; extra == "hf"
+Provides-Extra: vllm
+Requires-Dist: vllm==0.21.*; extra == "vllm"
+Requires-Dist: transformers>=5.15; extra == "vllm"
+Requires-Dist: xgrammar==0.2.1; extra == "vllm"
+Provides-Extra: vertex
+Requires-Dist: google-auth>=2.20; extra == "vertex"
+Requires-Dist: google-cloud-storage>=2.10; extra == "vertex"
+Provides-Extra: openai
+Requires-Dist: openai>=1.40; extra == "openai"
+Provides-Extra: anthropic
+Requires-Dist: anthropic>=1.0; extra == "anthropic"
+Provides-Extra: weborganizer
+Requires-Dist: torch>=2.4; extra == "weborganizer"
+Requires-Dist: transformers>=5.15; extra == "weborganizer"
+Provides-Extra: dev
+Requires-Dist: pytest>=8; extra == "dev"
+Provides-Extra: all
+Requires-Dist: idealens[anthropic,hf,openai,vertex,vllm,weborganizer]; extra == "all"
+Dynamic: license-file
+
+# idealens
+
+`idealens` detects whether a document's **ideas** came from a person or an AI model, whoever wrote the words. It runs
+the IdeaLens pipeline end to end:
+
+1. **Classify** the document into one of eight long-form formats (an LLM with WebOrganizer's annotation prompt, or
+   WebOrganizer's encoder).
+2. **Extract** an outline: an LLM writes a role-labelled list of the document's ideas, using the same prompt and six
+   worked examples the detector was trained with.
+3. **Score** the outline with a detector, which returns P(human) and a verdict at a calibrated false-positive rate.
+
+The same package also scores the prose-level detector ProseLens and the baselines released with the paper.
+
+Models: [IdeaLens](https://huggingface.co/rishanthrajendhran/IdeaLens) and the other detectors listed under
+[Models](#models). Data: [WildOutlines](https://huggingface.co/datasets/rishanthrajendhran/WildOutlines) (training
+corpus), [IdeaShift](https://huggingface.co/datasets/rishanthrajendhran/IdeaShift),
+[IdeaShift-X](https://huggingface.co/datasets/rishanthrajendhran/IdeaShift-X) and
+[TwiceTold](https://huggingface.co/datasets/rishanthrajendhran/TwiceTold) (evaluation sets).
+
+## Install
+
+```bash
+pip install "idealens[vllm]"          # the default scoring backend (Nemotron models, one 80 GB GPU)
+pip install "idealens[hf]"            # transformers backend; needed for the ModernBERT and Qwen models
+pip install "idealens[openai]"        # OpenAI, OpenRouter or any OpenAI-compatible endpoint (also the logistic models)
+pip install "idealens[anthropic]"     # Claude
+pip install "idealens[vertex]"        # Gemini on Vertex AI (batch mode also needs a GCS bucket)
+pip install "idealens[weborganizer]"  # WebOrganizer's encoder for format classification
+pip install "idealens[all]"
+```
+
+Python 3.10 or later. The vLLM extra pins `vllm==0.21.*`, `transformers>=5.15` and `xgrammar==0.2.1`.
+
+The model repos are gated: request access on the model's Hugging Face page, then `huggingface-cli login`.
+
+## Credentials
+
+| Provider (`--provider`) | Variable |
+|---|---|
+| `gemini` (default) | `GEMINI_API_KEY` or `GOOGLE_API_KEY` |
+| `vertex` | `GOOGLE_CLOUD_PROJECT` and application-default credentials; `IDEALENS_GCS_BUCKET` for batch mode |
+| `openai` | `OPENAI_API_KEY` |
+| `anthropic` | `ANTHROPIC_API_KEY` |
+| `openrouter` | `OPENROUTER_API_KEY` |
+| `compatible` | `--base-url` and, if the endpoint needs one, `--api-key` |
+
+The logistic models embed outlines with OpenAI's `text-embedding-3-large`, so they also need `OPENAI_API_KEY`.
+
+## Quick start
+
+Input is JSONL with a `text` field and, optionally, `id`, `url`, `format` and `topic`.
+
+```bash
+idealens run docs.jsonl -o scores.jsonl --dry-run   # estimate tokens and cost; sends nothing
+idealens run docs.jsonl -o scores.jsonl             # classify, extract and score with IdeaLens
+```
+
+Or one step at a time, which lets you extract once and score with several models:
+
+```bash
+idealens classify docs.jsonl     -o formats.jsonl
+idealens extract  formats.jsonl  -o outlines.jsonl
+idealens score    outlines.jsonl -o scores.jsonl --model IdeaLens
+idealens score    outlines.jsonl -o scores_np.jsonl --model IdeaLens-NoParaphrase
+```
+
+Every command keeps the input record's fields and appends to its output. Re-running a command resumes it: records
+already in the output are skipped, and records whose provider call failed are retried (the failed attempts are
+moved to `<output>.failed.jsonl`).
+
+In Python:
+
+```python
+import idealens as il
+
+texts = ["...", "..."]
+formats = il.classify(texts)                   # gemini-3.7-flash by default
+outlines = il.extract(texts, formats)
+with il.Detector("IdeaLens") as det:           # vLLM; the repo's published thresholds
+    records = det.score_outlines(outlines, format=formats)
+
+for r in records:
+    print(r["p_human"], r["verdict"]["ai"])
+```
+
+`il.run(texts, det)` does all three steps. Any provider can be passed as an object:
+
+```python
+from idealens.providers import make
+prov = make("openai", "gpt-6-sol")
+outlines = il.extract(texts, formats, provider=prov, mode="batch")
+```
+
+## Models
+
+`idealens models` lists them. All read the output of the same extraction step, except the two ProseLens models,
+which read the document itself.
+
+| Model | Reads | Backends |
+|---|---|---|
+| **IdeaLens** (default) | outline | vllm, hf |
+| ProseLens | document | vllm, hf |
+| IdeaLens-NoParaphrase | outline | vllm, hf |
+| IdeaLens-Qwen3.5-9B | outline | hf |
+| IdeaLens-Qwen3.5-9B-PerItem | each outline item | hf |
+| IdeaLens-ModernBERT-L | outline | hf |
+| IdeaLens-ModernBERT-L-NoParaphrase | outline | hf |
+| IdeaLens-ModernBERT-L-RolesOnly | the outline's role sequence | hf |
+| IdeaLens-ModernBERT-L-PerItem | each outline item | hf |
+| ProseLens-ModernBERT-L | document | hf |
+| IdeaLens-LogisticClassifier | outline | logistic |
+| IdeaLens-LogisticClassifier-PerItem | each outline item | logistic |
+
+Per-item models score each item and pool the item scores by their mean log-odds. Each model uses its own backend
+unless you pass `--backend`.
+
+The package does not paraphrase outlines. IdeaLens was trained and calibrated on paraphrased outlines;
+IdeaLens-NoParaphrase was trained on outlines as extracted, and is the closer match to what this package produces.
+
+## Verdicts and thresholds
+
+A document is flagged as AI when its P(human) is **strictly below** the cut. The default verdict uses the 1% global
+cut: the score below which 1% of the human calibration documents fall. Each record carries:
+
+| Field | Contents |
+|---|---|
+| `p_human` | the detector's P(human) |
+| `verdict` | the default verdict: `fpr`, `scheme`, `cut`, `ai` |
+| `verdicts` | every calibrated false-positive rate (0.1%, 0.5%, 1%, 2%, 5%) under each scheme: `global`, `per_format`, `per_topic`, and `group:<field>` for cuts you fitted yourself |
+| `warnings` | departures from how the cuts were fitted, such as a different extractor model |
+| `item_p_human` | per-item scores (per-item models) |
+
+Choose another operating point with `--fpr 0.005 --scheme per_format`. When a document has no cut under a scheme
+(its format or topic was not calibrated, or its format was force-fitted), that verdict is `None` with a reason. It
+never falls back to the global cut.
+
+The published cuts were fitted on 80,000 human documents with outlines extracted by `gemini-3.7-flash` with the six
+worked examples. Other extractor models work, but their outlines can shift the score distribution, so the
+calibrated false-positive rate is no longer guaranteed. Every record says so in `warnings`.
+
+### Calibrating on your own data
+
+If your documents differ from web text, fit cuts on human documents from your own domain:
+
+```bash
+idealens calibrate my_humans.jsonl --save-as my_domain --model IdeaLens
+idealens score outlines.jsonl -o scores.jsonl --thresholds my_domain
+```
+
+The input can be scored records, outlines or documents; the missing steps are run first. `--group-by FIELD` fits a
+cut per value of a field, and `--label-field` / `--human-value` let you pass a mixed file and calibrate on its human
+rows. A cut is fitted only where there are enough documents to estimate it (at least 25 expected below the cut and
+200 in the group). Profiles are saved under `~/.config/idealens/thresholds` (or `$IDEALENS_HOME/thresholds`).
+
+## Formats
+
+The detectors were trained on eight formats: Academic Writing, Creative Writing, Knowledge Article, News Article,
+Nonfiction Writing, Personal About Page, Personal Blog and User Reviews. A document in another format is assigned
+the closest of the eight by a second LLM pass (or by the encoder's probabilities). Such a document has
+`forced_format: true`, keeps its original label in `original_format`, and gets no per-format verdict.
+
+If you know a document's format, give it in the record's `format` field or with `--format`; classification is then
+skipped. `idealens run --check-format` classifies anyway and reports disagreements.
+
+## Cost
+
+`--dry-run` estimates the tokens and cost of a run before anything is sent, and `idealens cost output.jsonl` totals
+a finished one. Extraction dominates: the prompt carries six worked examples, and reasoning tokens make up most of
+the output. Batch mode (`--mode batch`) costs half as much where the provider has a batch API (Gemini on Vertex,
+OpenAI, Anthropic); not every model is served by it.
+
+## Hardware
+
+| Model | Needs |
+|---|---|
+| IdeaLens, ProseLens, IdeaLens-NoParaphrase | one 80 GB GPU (A100 80GB or H100 80GB); the weights take 59 GiB; about 60 GiB of CPU RAM while loading |
+| IdeaLens-Qwen3.5-9B models | one GPU; the weights take about 19 GB in bf16 (tested on 80 GB GPUs) |
+| ModernBERT models | any GPU; a CPU works for small jobs |
+| Logistic models | CPU only |
+
+The vLLM backend checks that the weights fit before it starts, and stops with a message otherwise. Use
+`tensor_parallel_size` (`Detector(..., tensor_parallel_size=2)`) to spread a model over two smaller GPUs.
+
+## Development
+
+```bash
+pip install -e ".[dev]"
+pytest tests
+```
+
+The tests need no GPU, network or API key.
+
+## License
+
+The code is released under the Apache License 2.0. Each model has its own license, given on its model page.
+
+## Citation
+
+A citation will be added here once the paper is on arXiv.

idealens-0.1.0/README.md

@@ -0,0 +1,199 @@
+# idealens
+
+`idealens` detects whether a document's **ideas** came from a person or an AI model, whoever wrote the words. It runs
+the IdeaLens pipeline end to end:
+
+1. **Classify** the document into one of eight long-form formats (an LLM with WebOrganizer's annotation prompt, or
+   WebOrganizer's encoder).
+2. **Extract** an outline: an LLM writes a role-labelled list of the document's ideas, using the same prompt and six
+   worked examples the detector was trained with.
+3. **Score** the outline with a detector, which returns P(human) and a verdict at a calibrated false-positive rate.
+
+The same package also scores the prose-level detector ProseLens and the baselines released with the paper.
+
+Models: [IdeaLens](https://huggingface.co/rishanthrajendhran/IdeaLens) and the other detectors listed under
+[Models](#models). Data: [WildOutlines](https://huggingface.co/datasets/rishanthrajendhran/WildOutlines) (training
+corpus), [IdeaShift](https://huggingface.co/datasets/rishanthrajendhran/IdeaShift),
+[IdeaShift-X](https://huggingface.co/datasets/rishanthrajendhran/IdeaShift-X) and
+[TwiceTold](https://huggingface.co/datasets/rishanthrajendhran/TwiceTold) (evaluation sets).
+
+## Install
+
+```bash
+pip install "idealens[vllm]"          # the default scoring backend (Nemotron models, one 80 GB GPU)
+pip install "idealens[hf]"            # transformers backend; needed for the ModernBERT and Qwen models
+pip install "idealens[openai]"        # OpenAI, OpenRouter or any OpenAI-compatible endpoint (also the logistic models)
+pip install "idealens[anthropic]"     # Claude
+pip install "idealens[vertex]"        # Gemini on Vertex AI (batch mode also needs a GCS bucket)
+pip install "idealens[weborganizer]"  # WebOrganizer's encoder for format classification
+pip install "idealens[all]"
+```
+
+Python 3.10 or later. The vLLM extra pins `vllm==0.21.*`, `transformers>=5.15` and `xgrammar==0.2.1`.
+
+The model repos are gated: request access on the model's Hugging Face page, then `huggingface-cli login`.
+
+## Credentials
+
+| Provider (`--provider`) | Variable |
+|---|---|
+| `gemini` (default) | `GEMINI_API_KEY` or `GOOGLE_API_KEY` |
+| `vertex` | `GOOGLE_CLOUD_PROJECT` and application-default credentials; `IDEALENS_GCS_BUCKET` for batch mode |
+| `openai` | `OPENAI_API_KEY` |
+| `anthropic` | `ANTHROPIC_API_KEY` |
+| `openrouter` | `OPENROUTER_API_KEY` |
+| `compatible` | `--base-url` and, if the endpoint needs one, `--api-key` |
+
+The logistic models embed outlines with OpenAI's `text-embedding-3-large`, so they also need `OPENAI_API_KEY`.
+
+## Quick start
+
+Input is JSONL with a `text` field and, optionally, `id`, `url`, `format` and `topic`.
+
+```bash
+idealens run docs.jsonl -o scores.jsonl --dry-run   # estimate tokens and cost; sends nothing
+idealens run docs.jsonl -o scores.jsonl             # classify, extract and score with IdeaLens
+```
+
+Or one step at a time, which lets you extract once and score with several models:
+
+```bash
+idealens classify docs.jsonl     -o formats.jsonl
+idealens extract  formats.jsonl  -o outlines.jsonl
+idealens score    outlines.jsonl -o scores.jsonl --model IdeaLens
+idealens score    outlines.jsonl -o scores_np.jsonl --model IdeaLens-NoParaphrase
+```
+
+Every command keeps the input record's fields and appends to its output. Re-running a command resumes it: records
+already in the output are skipped, and records whose provider call failed are retried (the failed attempts are
+moved to `<output>.failed.jsonl`).
+
+In Python:
+
+```python
+import idealens as il
+
+texts = ["...", "..."]
+formats = il.classify(texts)                   # gemini-3.7-flash by default
+outlines = il.extract(texts, formats)
+with il.Detector("IdeaLens") as det:           # vLLM; the repo's published thresholds
+    records = det.score_outlines(outlines, format=formats)
+
+for r in records:
+    print(r["p_human"], r["verdict"]["ai"])
+```
+
+`il.run(texts, det)` does all three steps. Any provider can be passed as an object:
+
+```python
+from idealens.providers import make
+prov = make("openai", "gpt-6-sol")
+outlines = il.extract(texts, formats, provider=prov, mode="batch")
+```
+
+## Models
+
+`idealens models` lists them. All read the output of the same extraction step, except the two ProseLens models,
+which read the document itself.
+
+| Model | Reads | Backends |
+|---|---|---|
+| **IdeaLens** (default) | outline | vllm, hf |
+| ProseLens | document | vllm, hf |
+| IdeaLens-NoParaphrase | outline | vllm, hf |
+| IdeaLens-Qwen3.5-9B | outline | hf |
+| IdeaLens-Qwen3.5-9B-PerItem | each outline item | hf |
+| IdeaLens-ModernBERT-L | outline | hf |
+| IdeaLens-ModernBERT-L-NoParaphrase | outline | hf |
+| IdeaLens-ModernBERT-L-RolesOnly | the outline's role sequence | hf |
+| IdeaLens-ModernBERT-L-PerItem | each outline item | hf |
+| ProseLens-ModernBERT-L | document | hf |
+| IdeaLens-LogisticClassifier | outline | logistic |
+| IdeaLens-LogisticClassifier-PerItem | each outline item | logistic |
+
+Per-item models score each item and pool the item scores by their mean log-odds. Each model uses its own backend
+unless you pass `--backend`.
+
+The package does not paraphrase outlines. IdeaLens was trained and calibrated on paraphrased outlines;
+IdeaLens-NoParaphrase was trained on outlines as extracted, and is the closer match to what this package produces.
+
+## Verdicts and thresholds
+
+A document is flagged as AI when its P(human) is **strictly below** the cut. The default verdict uses the 1% global
+cut: the score below which 1% of the human calibration documents fall. Each record carries:
+
+| Field | Contents |
+|---|---|
+| `p_human` | the detector's P(human) |
+| `verdict` | the default verdict: `fpr`, `scheme`, `cut`, `ai` |
+| `verdicts` | every calibrated false-positive rate (0.1%, 0.5%, 1%, 2%, 5%) under each scheme: `global`, `per_format`, `per_topic`, and `group:<field>` for cuts you fitted yourself |
+| `warnings` | departures from how the cuts were fitted, such as a different extractor model |
+| `item_p_human` | per-item scores (per-item models) |
+
+Choose another operating point with `--fpr 0.005 --scheme per_format`. When a document has no cut under a scheme
+(its format or topic was not calibrated, or its format was force-fitted), that verdict is `None` with a reason. It
+never falls back to the global cut.
+
+The published cuts were fitted on 80,000 human documents with outlines extracted by `gemini-3.7-flash` with the six
+worked examples. Other extractor models work, but their outlines can shift the score distribution, so the
+calibrated false-positive rate is no longer guaranteed. Every record says so in `warnings`.
+
+### Calibrating on your own data
+
+If your documents differ from web text, fit cuts on human documents from your own domain:
+
+```bash
+idealens calibrate my_humans.jsonl --save-as my_domain --model IdeaLens
+idealens score outlines.jsonl -o scores.jsonl --thresholds my_domain
+```
+
+The input can be scored records, outlines or documents; the missing steps are run first. `--group-by FIELD` fits a
+cut per value of a field, and `--label-field` / `--human-value` let you pass a mixed file and calibrate on its human
+rows. A cut is fitted only where there are enough documents to estimate it (at least 25 expected below the cut and
+200 in the group). Profiles are saved under `~/.config/idealens/thresholds` (or `$IDEALENS_HOME/thresholds`).
+
+## Formats
+
+The detectors were trained on eight formats: Academic Writing, Creative Writing, Knowledge Article, News Article,
+Nonfiction Writing, Personal About Page, Personal Blog and User Reviews. A document in another format is assigned
+the closest of the eight by a second LLM pass (or by the encoder's probabilities). Such a document has
+`forced_format: true`, keeps its original label in `original_format`, and gets no per-format verdict.
+
+If you know a document's format, give it in the record's `format` field or with `--format`; classification is then
+skipped. `idealens run --check-format` classifies anyway and reports disagreements.
+
+## Cost
+
+`--dry-run` estimates the tokens and cost of a run before anything is sent, and `idealens cost output.jsonl` totals
+a finished one. Extraction dominates: the prompt carries six worked examples, and reasoning tokens make up most of
+the output. Batch mode (`--mode batch`) costs half as much where the provider has a batch API (Gemini on Vertex,
+OpenAI, Anthropic); not every model is served by it.
+
+## Hardware
+
+| Model | Needs |
+|---|---|
+| IdeaLens, ProseLens, IdeaLens-NoParaphrase | one 80 GB GPU (A100 80GB or H100 80GB); the weights take 59 GiB; about 60 GiB of CPU RAM while loading |
+| IdeaLens-Qwen3.5-9B models | one GPU; the weights take about 19 GB in bf16 (tested on 80 GB GPUs) |
+| ModernBERT models | any GPU; a CPU works for small jobs |
+| Logistic models | CPU only |
+
+The vLLM backend checks that the weights fit before it starts, and stops with a message otherwise. Use
+`tensor_parallel_size` (`Detector(..., tensor_parallel_size=2)`) to spread a model over two smaller GPUs.
+
+## Development
+
+```bash
+pip install -e ".[dev]"
+pytest tests
+```
+
+The tests need no GPU, network or API key.
+
+## License
+
+The code is released under the Apache License 2.0. Each model has its own license, given on its model page.
+
+## Citation
+
+A citation will be added here once the paper is on arXiv.