groundgate 0.1.0__tar.gz

groundgate-0.1.0/.gitignore

@@ -0,0 +1,22 @@
+__pycache__/
+*.py[cod]
+.venv/
+.venv-*/
+dist/
+build/
+*.egg-info/
+.coverage
+coverage.xml
+htmlcov/
+.mypy_cache/
+.pytest_cache/
+.ruff_cache/
+.DS_Store
+.env
+
+# spike raw sources (fetched and hash-verified by spikes/m0_5/fetch.py)
+spikes/*/sources/*.pdf
+spikes/*/sources/*.xml
+
+# benchmark downloads (hash-pinned in bench/sources.json)
+bench/.cache/

groundgate-0.1.0/CHANGELOG.md

@@ -0,0 +1,21 @@
+# Changelog
+
+Versions follow [SemVer](https://semver.org). A change to how any candidate is decided is a new
+spec version, and receipts name the spec version they were decided under.
+
+## 0.1.0 (2026-09-29)
+
+First release. Implements spec v0.1.
+
+- `admit` and `verify`: deterministic admission with ten rejection codes, five flags, required
+  field coverage and re-anchoring, and receipts hashed over RFC 8785 canonical JSON.
+- `groundgate extract`: PDF (with `[pdf]`), HTML, XML and text to NFC text, with page and word
+  boxes for PDFs. Text a reader can't see is dropped, and superscripts are marked.
+- A LangExtract adapter that checks each extraction's value and unit at the place LangExtract
+  aligned it.
+- `groundgate report`: a self-contained HTML review page that re-derives the receipt before it
+  renders.
+- 16 conformance vectors and 6 invalid packets, language-neutral, so another implementation can
+  prove it agrees.
+- A benchmark on 30 public-domain FDA, NTSB and IRS documents with person-checked gold and seven
+  models: [bench/RESULTS.md](https://github.com/mohanraj00/groundgate/blob/main/bench/RESULTS.md).

groundgate-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.

groundgate-0.1.0/PKG-INFO

@@ -0,0 +1,275 @@
+Metadata-Version: 2.5
+Name: groundgate
+Version: 0.1.0
+Summary: Deterministic admission for LLM-extracted facts: admit, flag for verification, or reject, with receipts.
+Project-URL: Repository, https://github.com/mohanraj00/groundgate
+Project-URL: Documentation, https://github.com/mohanraj00/groundgate/blob/main/docs/guide.md
+Project-URL: Specification, https://github.com/mohanraj00/groundgate/blob/main/SPEC.md
+Project-URL: Changelog, https://github.com/mohanraj00/groundgate/blob/main/CHANGELOG.md
+Project-URL: Issues, https://github.com/mohanraj00/groundgate/issues
+Author: Mohan Raj
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: extraction,grounding,langextract,llm,verification
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+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
+Classifier: Topic :: Text Processing
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Provides-Extra: langextract
+Requires-Dist: langextract>=1.0; extra == 'langextract'
+Provides-Extra: pdf
+Requires-Dist: pdfminer-six>=20231228; extra == 'pdf'
+Description-Content-Type: text/markdown
+
+# groundgate
+
+**LLMs propose facts. groundgate decides which ones you can trust.**
+
+groundgate is a deterministic admission layer for LLM document extraction. You give it a
+document, a schema, and the facts an extractor proposed. Every fact comes back with one of three
+outcomes:
+
+- **admitted**: every check passed;
+- **needs_verification**: nothing failed, but a person should look, and the decision says where;
+- **rejected**: a check failed, with a stable reason code.
+
+Every run writes a receipt with hashes of every input. Anyone can re-derive it byte for byte.
+
+![The groundgate review page for IRS Publication 590-A. Left: a Roth IRA limit flagged QUALIFIED_VALUE because the model cited the spousal-deduction sentence ("more than $236,000"). Right: the source text, with that citation outlined in amber and the correct Roth sentence, admitted, highlighted in green below it.](https://raw.githubusercontent.com/mohanraj00/groundgate/main/docs/report.png)
+
+## Why this exists
+
+Grounded extraction tools check that the quoted **text** exists in the source. They don't check
+the **value** you asked for.
+
+LangExtract is the clearest example, and a good tool. It aligns each `extraction_text` to the
+source and reports `MATCH_EXACT`, `MATCH_FUZZY` and so on. It never looks at the attributes that
+hold the typed value. So this passes as an exact match:
+
+```text
+extraction_text: "$8,000"     found in the source: MATCH_EXACT
+value:           "80000"      not what the source says
+```
+
+Here is what that looks like on a real document. I ran Gemini 3.6 Flash and GPT-OSS 120B through
+LangExtract on two pages of IRS Publication 590-A. LangExtract aligned all 48 extractions as
+`MATCH_EXACT`. groundgate then:
+
+- **rejected** Gemini's reading of a misprinted `$252,0000` as 2,520,000;
+- **flagged** two GPT-OSS facts that took a Roth IRA limit from the spousal-deduction sentence;
+  the numbers match, but the text says "more than" where the field says "at least";
+- **listed** a required field both models missed;
+- **admitted** 44 facts, all with the right value.
+
+One admitted fact still cites the wrong rule. Its number is correct, its comparator matches, and
+no span check can tell. I checked every decision by hand; the details are in
+[examples/irs-590a](https://github.com/mohanraj00/groundgate/blob/main/examples/irs-590a), which is also the page in the screenshot.
+
+I measured the gap on purpose before writing the library. On 51 facts from two public-domain
+government documents, LangExtract's strictest setting (`MATCH_EXACT` only) accepted **100%** of
+planted value and unit errors whose quote was correct. groundgate caught **all** of them and
+rejected **none** of the correct facts. The method and full tables are in
+[spikes/m0_5](https://github.com/mohanraj00/groundgate/blob/main/spikes/m0_5/README.md).
+
+To be clear about scale: current models rarely get a value wrong. In the benchmark below, 117 of
+4,350 real extractions were wrong. The point is not that models are bad. It is that when
+one is wrong, nothing downstream should have to trust it, and every fact that is admitted comes
+with proof you can re-check.
+
+LangExtract verifies the text. groundgate verifies the value.
+
+## Benchmark
+
+Seven models (Gemini, GPT and Claude, each through its own CLI) ran through LangExtract on 30
+public-domain FDA drug labels, NTSB accident reports and IRS publications, at two chunk sizes. A
+person checked every gold fact. Of 4,350 extractions, 117 were wrong: the wrong value, a value
+for a field the document doesn't state, the wrong unit, or not a number.
+
+![Pooled over all 14 runs: MATCH_EXACT accepted 98% of wrong extractions without review, groundgate 6.8%. MATCH_EXACT rejected 4.6% of correct extractions, groundgate 0.3%. groundgate sent 12% of extractions to a person.](https://raw.githubusercontent.com/mohanraj00/groundgate/main/bench/charts/summary.svg)
+
+| | LangExtract, `MATCH_EXACT` | groundgate |
+|---|---:|---:|
+| wrong extractions accepted without review | 98.3% (115/117) | 6.8% (8/117) |
+| correct extractions rejected | 4.6% (194/4,233) | 0.3% (14/4,233) |
+| extractions sent to a person | 0% | 12.1% (526/4,350) |
+
+Most catches (84 of 109) were `CONFLICTING_CANDIDATES`: a model proposed two values for a
+single-valued field, and groundgate sent both to review instead of picking one.
+
+Six of the 8 that got through are the case groundgate says it cannot catch (see
+[What it does not do](#what-it-does-not-do)). Five models took metoprolol's heart failure maximum
+(200 mg) as the hypertension maximum, and one took lisinopril's renal impairment dose as the usual
+starting dose. Each number is real and cited correctly; it belongs to another condition. The
+other two read levothyroxine's 1.6 mcg/kg/day as a flat 1.6 mcg, which the unit check allows.
+Putting all seven models through one gate caught one more, because when they were wrong they
+mostly agreed.
+
+The benchmark also found a bug. When LangExtract's chunker cut "29.97" after "29.", groundgate
+read the span as 29, which its own spec forbids. It's fixed, and three wrong altimeter settings no
+longer get through.
+
+Tables and planted-error results: [bench/RESULTS.md](https://github.com/mohanraj00/groundgate/blob/main/bench/RESULTS.md). Method and caveats:
+[bench/README.md](https://github.com/mohanraj00/groundgate/blob/main/bench/README.md).
+
+## Quickstart
+
+```bash
+pip install groundgate
+```
+
+```python
+import groundgate as gg
+
+text = "The IRA contribution limit is $7,000. The deduction phases out above $79,000."
+schema = {
+    "fields": {
+        "ira_limit": {"type": "integer", "unit": "USD"},
+        "phaseout_start": {"type": "integer", "unit": "USD"},
+    }
+}
+
+
+def cite(field, value, quote):
+    """A candidate citing the first place `quote` appears, as UTF-8 byte offsets."""
+    start = text.encode().index(quote.encode())
+    span = {"start": start, "end": start + len(quote.encode()), "text": quote}
+    return {"field": field, "value": value, "unit": "USD", "evidence": span}
+
+
+candidates = [
+    cite("ira_limit", "7000", "$7,000"),
+    cite("ira_limit", "70000", "$7,000"),  # a digit too many
+    cite("phaseout_start", "79000", "$79,000"),  # "above" is not "at"
+]
+receipt = gg.admit(text, schema, candidates)
+for d in receipt.decisions:
+    print(f"{d.outcome:<19} {d.field:<15} {d.value:<6} {' '.join(d.codes)}".rstrip())
+
+print(gg.verify(receipt.to_dict(), text, schema, candidates).ok)
+```
+
+```text
+needs_verification  phaseout_start  79000  QUALIFIED_VALUE
+admitted            ira_limit       7000
+rejected            ira_limit       70000  VALUE_NOT_IN_EVIDENCE
+True
+```
+
+The core has no dependencies. `groundgate[pdf]` adds pdfminer.six for PDF text, and
+`groundgate[langextract]` installs LangExtract. [docs/guide.md](https://github.com/mohanraj00/groundgate/blob/main/docs/guide.md) covers schemas,
+policies, candidates and the Python API.
+
+### With LangExtract
+
+```python
+import langextract as lx
+from groundgate.adapters.langextract import admit_document
+
+result = lx.extract(text_or_documents=text, prompt_description=prompt, examples=examples)
+receipt = admit_document(result, schema)
+```
+
+Each extraction's `value` and `unit` attributes are checked at the place LangExtract aligned its
+`extraction_text`. An extraction LangExtract could not align is rejected `NO_EVIDENCE`.
+
+### From the command line
+
+```bash
+groundgate extract p590a.pdf --pages 1-2 -o doc.txt --layout layout.json
+groundgate admit   doc.txt schema.json candidates.json -o receipt.json
+groundgate verify  receipt.json doc.txt schema.json candidates.json
+groundgate report  receipt.json doc.txt schema.json candidates.json --layout layout.json -o report.html
+```
+
+`extract` turns a PDF, HTML, XML or text file into the NFC text everything else reads, and
+records the page and box of every word. Text a reader can't see (proof marks drawn off the page,
+hidden HTML) is dropped, and superscripts are marked so that 10⁹ reads `10^9`, never `109`. Pass
+`-` as the document to read it from a pipe. `report` refuses to render a receipt that doesn't
+re-derive from its inputs. [examples/irs-590a](https://github.com/mohanraj00/groundgate/blob/main/examples/irs-590a) runs the whole pipeline on an
+IRS publication with cached model output, so it reproduces without an API key.
+
+## What it checks
+
+Checks run in a fixed order. The first failure rejects the fact.
+
+| Code | Rejects when |
+|---|---|
+| `CANDIDATE_INVALID` | the candidate is malformed |
+| `FIELD_UNKNOWN` | the field is not in the schema |
+| `NULL_STRING_LITERAL` | the value is `"null"`, `"none"`, `"n/a"` |
+| `TYPE_INVALID` | the value does not parse as the field's type |
+| `RANGE_INVALID` | the value is outside the field's bounds |
+| `UNIT_INVALID` | the unit is not the field's unit |
+| `NO_EVIDENCE` | no evidence is cited |
+| `SPAN_INVALID` | the span is outside the document or splits a character |
+| `VALUE_NOT_IN_EVIDENCE` | no number in the span equals the value |
+| `UNIT_NOT_IN_EVIDENCE` | the value is there, but not with the field's unit |
+
+A fact that passes every check can still be flagged for a person:
+
+| Code | Flags when |
+|---|---|
+| `NON_VERBATIM_EVIDENCE` | the quote differs from the text at the span |
+| `QUALIFIED_VALUE` | "up to", "approximately", "or more" changes the value, and the schema didn't declare it |
+| `SCALE_WORD` | "million", "lakh" and similar follow the value |
+| `LOW_CONFIDENCE` | the extractor's confidence is below the policy minimum |
+| `CONFLICTING_CANDIDATES` | another proposal for the same field has a different value |
+
+Number matching is collision-safe. The span `500 mg` inside `1,500 mg` never reads as 500, a span
+that stops at `29.` inside `29.97` never reads as 29, and a malformed number like the `$252,0000`
+printed in IRS Publication 590-A never equals 252,000 or 2,520,000. When a span misses the value
+but its quote occurs exactly once elsewhere with the right value and unit, groundgate moves the
+evidence there and records `EVIDENCE_REANCHORED`.
+
+The rules are in [SPEC.md](https://github.com/mohanraj00/groundgate/blob/main/SPEC.md). [conformance/](https://github.com/mohanraj00/groundgate/blob/main/conformance) holds 16 language-neutral
+vectors that pin every code, so another implementation can prove it agrees.
+
+## Receipts
+
+```json
+{"groundgate": "0.1",
+ "document": {"id": "irs-p590a-2025-pages-1-2", "sha256": "sha256:387c5991b989..."},
+ "schema_sha256": "sha256:...", "policy_sha256": "sha256:...",
+ "decisions": [{"candidate_id": "Gemini_3.6_Flash_Medium/0", "field": "ira_limit_2025",
+                "outcome": "admitted", "codes": [], "value": "7000", "unit": "USD",
+                "evidence": {"start": 2634, "end": 2640}, "candidate_sha256": "sha256:37e4fbb8..."},
+               "..."],
+ "coverage": [{"field": "roth_phaseout_joint_2026_end", "code": "REQUIRED_FIELD_MISSING"}],
+ "summary": {"admitted": 50, "needs_verification": 0, "rejected": 2},
+ "receipt_sha256": "sha256:..."}
+```
+
+Hashes are SHA-256 over RFC 8785 canonical JSON, so a receipt means the same thing in any
+language. `groundgate verify` re-runs every decision from the inputs and fails on any difference:
+a changed document, schema, policy, candidate or outcome.
+
+## What it does not do
+
+- **It does not judge meaning.** If the model reports a number that really is in the sentence but
+  belongs to another field, every span check passes: a model that reads an age-50 limit of $8,000
+  as the IRA limit cites a real "$8,000" with the right unit. A second proposer plus
+  `CONFLICTING_CANDIDATES` catches it only when the models disagree, and in the benchmark they
+  mostly agreed: six of the eight escapes were this case.
+- **No dates, arrays of records, or cross-document checks** in spec v0.1.
+- **No OCR.** Scanned PDFs need a text layer first (for example `ocrmypdf`).
+- **No model calls.** groundgate never asks an LLM whether an LLM was right.
+
+## Status
+
+Alpha. groundgate 0.1.0 implements spec v0.1, and a receipt names the spec version it was decided
+under. The next spec version starts from what the benchmark found: values that depend on a
+condition (a dose per indication, a limit per year) and weight-based units. See
+[CHANGELOG.md](https://github.com/mohanraj00/groundgate/blob/main/CHANGELOG.md).
+
+## License
+
+Apache-2.0.