smallwords 0.1.0__tar.gz

smallwords-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Chris McComb
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

smallwords-0.1.0/MANIFEST.in

@@ -0,0 +1,11 @@
+include LICENSE
+include README.md
+include RELEASING.md
+graft docs
+graft examples
+graft scripts
+graft tests
+prune docs/_build
+global-exclude __pycache__
+global-exclude *.py[cod]
+global-exclude .DS_Store

smallwords-0.1.0/PKG-INFO

@@ -0,0 +1,175 @@
+Metadata-Version: 2.4
+Name: smallwords
+Version: 0.1.0
+Summary: Controlled-vocabulary prompts plus portable GBNF and JSON Schema resources for small-word English generation.
+Author: Christopher McComb
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/cmccomb/smallwords
+Project-URL: Documentation, https://cmccomb.github.io/smallwords/
+Project-URL: Repository, https://github.com/cmccomb/smallwords.git
+Project-URL: Issues, https://github.com/cmccomb/smallwords/issues
+Project-URL: Changelog, https://github.com/cmccomb/smallwords/releases
+Keywords: grammar,gbnf,json-schema,structured-output,controlled-language,simple-english
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+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: Operating System :: OS Independent
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Topic :: Text Processing :: Linguistic
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == "dev"
+Requires-Dist: pydata-sphinx-theme>=0.16; extra == "dev"
+Requires-Dist: pytest>=8.0; extra == "dev"
+Requires-Dist: pytest-cov>=7.1; extra == "dev"
+Requires-Dist: ruff>=0.11; extra == "dev"
+Requires-Dist: sphinx<9,>=8.1; extra == "dev"
+Requires-Dist: twine>=6.1; extra == "dev"
+Dynamic: license-file
+
+# smallwords
+
+[![CI](https://github.com/cmccomb/smallwords/actions/workflows/ci.yml/badge.svg)](https://github.com/cmccomb/smallwords/actions/workflows/ci.yml)
+[![PyPI version](https://img.shields.io/pypi/v/smallwords)](https://pypi.org/project/smallwords/)
+[![Python versions](https://img.shields.io/pypi/pyversions/smallwords)](https://pypi.org/project/smallwords/)
+[![License](https://img.shields.io/github/license/cmccomb/smallwords)](https://github.com/cmccomb/smallwords/blob/main/LICENSE)
+[![Docs](https://img.shields.io/badge/docs-GitHub%20Pages-4c1)](https://cmccomb.github.io/smallwords/)
+
+`smallwords` is a tiny Python package for controlled-vocabulary prompting plus
+portable output resources. It keeps one wordlist at the center of the workflow
+so prompt text, GBNF, JSON Schema, and post-generation validation all stay in
+sync.
+
+The package ships with a small set of bundled wordlists: direct source-backed
+lists such as `moby_898`, `basic_850`, and `special_english_1475`, plus a
+couple of intentionally themed remixes. By default, the built-ins also allow
+slight family variants such as `go`, `goes`, and `going`.
+
+It supports Python 3.10 and newer.
+
+The hosted API-and-examples docs live at
+[`cmccomb.github.io/smallwords`](https://cmccomb.github.io/smallwords/).
+
+## Installation
+
+```bash
+pip install smallwords
+```
+
+For local development, create and activate a virtualenv first:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+python -m pip install -e ".[dev]"
+```
+
+## Quick Start
+
+```python
+from smallwords import OutputResources, OutputShape, allow_input_words, is_compliant
+from smallwords.prompts import build_prompt
+
+shape = OutputShape(max_words_per_line=24, max_lines=1)
+spec = allow_input_words("basic_850", "How does a bridge work?")
+resources = OutputResources.from_wordlist(spec, shape=shape)
+prompt = build_prompt("explain", "How does a bridge work?", wordlist=spec)
+schema = resources.json_schema(key="answer", title="bridge_explanation")
+
+text = "A bridge is a structure that helps people and things move across a river or a deep place."
+ok = is_compliant(text, spec)
+```
+
+The contrast is the point. `build_prompt(...)` is the soft instruction layer.
+`OutputResources` gives you the matching hard constraints in both GBNF and JSON
+Schema form. `is_compliant(...)` is the lightweight offline check.
+
+If you want the model to be able to repeat topic or question terms such as
+`bridge`, `neighbor`, or `order`, use `allow_input_words(...)` once and pass
+that derived spec into the prompt, resources, and validation helpers together.
+
+## Built-In Wordlists
+
+- `moby_898`: the full normalized alpha-only Moby Words II frequency list
+- `basic_850`: Charles Ogden's Basic English 850
+- `special_english_1475`: Voice of America Special English
+- `caveman_898`: a size-neutral surface-only `moby_898` remix with caveman adjustments
+- `pirate_898`: a size-neutral `moby_898` remix with pirate adjustments
+
+The bundled text files live in `src/smallwords/data/`. `moby_898`,
+`basic_850`, and `special_english_1475` are direct source-backed lists.
+`caveman_898` and `pirate_898` are derived size-neutral remixes built on top of
+`moby_898`.
+
+The themed remixes live in `src/smallwords/themes/caveman.py` and
+`src/smallwords/themes/pirate.py`. If you want to build your own, use
+`remix_wordlist(...)` with a base list plus curated additions and removals.
+
+## Contrastive Example
+
+This is the clearest way to see what `smallwords` is trying to do. Both blocks
+below are genuine local Qwen outputs from April 5, 2026. The first uses a plain
+prompt. The second uses the same base prompt plus an explicit `basic_850`
+vocabulary list, the topic word `bridge`, and the generated GBNF.
+
+A plain prompt stays fairly natural:
+
+> A bridge connects two points, usually across a body of water or a gap,
+> allowing people and vehicles to cross safely.
+
+A constrained `basic_850 + topic words` run stays simpler while still sounding
+reasonably natural:
+
+> A bridge is a structure that helps people and things move across a river or a
+> deep place.
+
+These runs use `llama-server` from `llama.cpp` and
+[`Qwen/Qwen3-8B-GGUF`](https://huggingface.co/Qwen/Qwen3-8B-GGUF)
+via
+[`bartowski/Qwen_Qwen3-8B-GGUF`](https://huggingface.co/bartowski/Qwen_Qwen3-8B-GGUF).
+
+Reproduce that comparison from a clone of the repository with an activated
+virtualenv:
+
+```bash
+llama-server -hf bartowski/Qwen_Qwen3-8B-GGUF:q4_k_m --host 127.0.0.1 --port 8080 --reasoning-budget 0 --log-disable
+python examples/readme_bridge_contrast.py
+```
+
+## Examples
+
+See the repository's
+[`examples/README.md`](https://github.com/cmccomb/smallwords/blob/main/examples/README.md)
+for the runnable examples. The current example set is live-model based:
+the README bridge contrast, a focused pirate greeting, and a focused technical
+rewrite all call a live `llama-server` model with a prompt plus generated
+grammar.
+
+## Development
+
+Run these commands from an activated virtualenv:
+
+```bash
+python -m pip install -e ".[dev]"
+python -m ruff check .
+python -m ruff format --check .
+python -m pytest
+python scripts/check_documentation.py
+python -m sphinx -W --keep-going -b html docs docs/_build/html
+python -m build
+python -m twine check --strict dist/*
+```
+
+CI runs linting, tests, the documentation policy check, a `>=90%` coverage
+gate, a Sphinx docs build, and a package build on GitHub Actions.
+
+For release steps and Trusted Publishing setup, see
+[`RELEASING.md`](https://github.com/cmccomb/smallwords/blob/main/RELEASING.md).

smallwords-0.1.0/README.md

@@ -0,0 +1,138 @@
+# smallwords
+
+[![CI](https://github.com/cmccomb/smallwords/actions/workflows/ci.yml/badge.svg)](https://github.com/cmccomb/smallwords/actions/workflows/ci.yml)
+[![PyPI version](https://img.shields.io/pypi/v/smallwords)](https://pypi.org/project/smallwords/)
+[![Python versions](https://img.shields.io/pypi/pyversions/smallwords)](https://pypi.org/project/smallwords/)
+[![License](https://img.shields.io/github/license/cmccomb/smallwords)](https://github.com/cmccomb/smallwords/blob/main/LICENSE)
+[![Docs](https://img.shields.io/badge/docs-GitHub%20Pages-4c1)](https://cmccomb.github.io/smallwords/)
+
+`smallwords` is a tiny Python package for controlled-vocabulary prompting plus
+portable output resources. It keeps one wordlist at the center of the workflow
+so prompt text, GBNF, JSON Schema, and post-generation validation all stay in
+sync.
+
+The package ships with a small set of bundled wordlists: direct source-backed
+lists such as `moby_898`, `basic_850`, and `special_english_1475`, plus a
+couple of intentionally themed remixes. By default, the built-ins also allow
+slight family variants such as `go`, `goes`, and `going`.
+
+It supports Python 3.10 and newer.
+
+The hosted API-and-examples docs live at
+[`cmccomb.github.io/smallwords`](https://cmccomb.github.io/smallwords/).
+
+## Installation
+
+```bash
+pip install smallwords
+```
+
+For local development, create and activate a virtualenv first:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+python -m pip install -e ".[dev]"
+```
+
+## Quick Start
+
+```python
+from smallwords import OutputResources, OutputShape, allow_input_words, is_compliant
+from smallwords.prompts import build_prompt
+
+shape = OutputShape(max_words_per_line=24, max_lines=1)
+spec = allow_input_words("basic_850", "How does a bridge work?")
+resources = OutputResources.from_wordlist(spec, shape=shape)
+prompt = build_prompt("explain", "How does a bridge work?", wordlist=spec)
+schema = resources.json_schema(key="answer", title="bridge_explanation")
+
+text = "A bridge is a structure that helps people and things move across a river or a deep place."
+ok = is_compliant(text, spec)
+```
+
+The contrast is the point. `build_prompt(...)` is the soft instruction layer.
+`OutputResources` gives you the matching hard constraints in both GBNF and JSON
+Schema form. `is_compliant(...)` is the lightweight offline check.
+
+If you want the model to be able to repeat topic or question terms such as
+`bridge`, `neighbor`, or `order`, use `allow_input_words(...)` once and pass
+that derived spec into the prompt, resources, and validation helpers together.
+
+## Built-In Wordlists
+
+- `moby_898`: the full normalized alpha-only Moby Words II frequency list
+- `basic_850`: Charles Ogden's Basic English 850
+- `special_english_1475`: Voice of America Special English
+- `caveman_898`: a size-neutral surface-only `moby_898` remix with caveman adjustments
+- `pirate_898`: a size-neutral `moby_898` remix with pirate adjustments
+
+The bundled text files live in `src/smallwords/data/`. `moby_898`,
+`basic_850`, and `special_english_1475` are direct source-backed lists.
+`caveman_898` and `pirate_898` are derived size-neutral remixes built on top of
+`moby_898`.
+
+The themed remixes live in `src/smallwords/themes/caveman.py` and
+`src/smallwords/themes/pirate.py`. If you want to build your own, use
+`remix_wordlist(...)` with a base list plus curated additions and removals.
+
+## Contrastive Example
+
+This is the clearest way to see what `smallwords` is trying to do. Both blocks
+below are genuine local Qwen outputs from April 5, 2026. The first uses a plain
+prompt. The second uses the same base prompt plus an explicit `basic_850`
+vocabulary list, the topic word `bridge`, and the generated GBNF.
+
+A plain prompt stays fairly natural:
+
+> A bridge connects two points, usually across a body of water or a gap,
+> allowing people and vehicles to cross safely.
+
+A constrained `basic_850 + topic words` run stays simpler while still sounding
+reasonably natural:
+
+> A bridge is a structure that helps people and things move across a river or a
+> deep place.
+
+These runs use `llama-server` from `llama.cpp` and
+[`Qwen/Qwen3-8B-GGUF`](https://huggingface.co/Qwen/Qwen3-8B-GGUF)
+via
+[`bartowski/Qwen_Qwen3-8B-GGUF`](https://huggingface.co/bartowski/Qwen_Qwen3-8B-GGUF).
+
+Reproduce that comparison from a clone of the repository with an activated
+virtualenv:
+
+```bash
+llama-server -hf bartowski/Qwen_Qwen3-8B-GGUF:q4_k_m --host 127.0.0.1 --port 8080 --reasoning-budget 0 --log-disable
+python examples/readme_bridge_contrast.py
+```
+
+## Examples
+
+See the repository's
+[`examples/README.md`](https://github.com/cmccomb/smallwords/blob/main/examples/README.md)
+for the runnable examples. The current example set is live-model based:
+the README bridge contrast, a focused pirate greeting, and a focused technical
+rewrite all call a live `llama-server` model with a prompt plus generated
+grammar.
+
+## Development
+
+Run these commands from an activated virtualenv:
+
+```bash
+python -m pip install -e ".[dev]"
+python -m ruff check .
+python -m ruff format --check .
+python -m pytest
+python scripts/check_documentation.py
+python -m sphinx -W --keep-going -b html docs docs/_build/html
+python -m build
+python -m twine check --strict dist/*
+```
+
+CI runs linting, tests, the documentation policy check, a `>=90%` coverage
+gate, a Sphinx docs build, and a package build on GitHub Actions.
+
+For release steps and Trusted Publishing setup, see
+[`RELEASING.md`](https://github.com/cmccomb/smallwords/blob/main/RELEASING.md).

smallwords-0.1.0/RELEASING.md

@@ -0,0 +1,39 @@
+# Releasing `smallwords`
+
+This repository is set up for PyPI Trusted Publishing through GitHub Actions.
+
+Before the first release:
+
+1. Create the `smallwords` project on PyPI.
+2. Configure a Trusted Publisher on PyPI for this GitHub repository.
+3. Add the `pypi` GitHub Actions environment if you want environment-level protection rules.
+
+Recommended release flow:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+python -m pip install -e ".[dev]"
+python -m ruff check .
+python -m ruff format --check .
+python -m pytest
+python scripts/check_documentation.py
+python -m sphinx -W --keep-going -b html docs docs/_build/html
+python -m build
+python -m twine check --strict dist/*
+python -m venv /tmp/smallwords-release-check
+/tmp/smallwords-release-check/bin/python -m pip install --upgrade pip
+/tmp/smallwords-release-check/bin/python -m pip install dist/*.whl
+/tmp/smallwords-release-check/bin/python -c "import smallwords; print(smallwords.__version__)"
+```
+
+Then:
+
+1. Update `version` in `pyproject.toml`.
+2. Commit the release.
+3. Create and push a Git tag.
+4. Publish a GitHub release from that tag.
+
+The `Publish` workflow builds the distributions, validates them with `twine`,
+and uploads them to PyPI through Trusted Publishing. The separate `Docs`
+workflow builds the Sphinx site and deploys it to GitHub Pages from `main`.

smallwords-0.1.0/docs/conf.py

@@ -0,0 +1,39 @@
+"""Sphinx configuration for the compact project docs site."""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+# Add the source tree so autodoc can import the local package during docs builds.
+ROOT = Path(__file__).resolve().parents[1]
+SRC = ROOT / "src"
+if str(SRC) not in sys.path:
+    sys.path.insert(0, str(SRC))
+
+from smallwords import __version__  # noqa: E402
+
+project = "smallwords"
+author = "Christopher McComb"
+copyright = "2026, Christopher McComb"
+version = __version__
+release = __version__
+
+extensions = [
+    "sphinx.ext.autodoc",
+    "sphinx.ext.napoleon",
+    "sphinx.ext.viewcode",
+]
+
+exclude_patterns = ["_build", "Thumbs.db", ".DS_Store"]
+autodoc_member_order = "bysource"
+autodoc_typehints = "description"
+napoleon_google_docstring = True
+napoleon_numpy_docstring = False
+
+html_theme = "pydata_sphinx_theme"
+html_title = "smallwords"
+html_theme_options = {
+    "github_url": "https://github.com/cmccomb/smallwords",
+    "show_toc_level": 2,
+}

smallwords-0.1.0/docs/index.rst

@@ -0,0 +1,164 @@
+smallwords
+==========
+
+Large language models can follow "write simply" instructions, but that soft
+guidance is often insufficient when a workflow also needs reproducible wording,
+portable output constraints, or offline validation. Prompt-only approaches are
+easy to start and hard to trust. They leave the active vocabulary implicit,
+make failures difficult to diagnose, and force each integration to recreate the
+same constraint logic in a different format.
+
+``smallwords`` addresses that gap by keeping one controlled vocabulary at the
+center of the workflow. From that single specification, the package can build
+prompt text, GBNF, JSON Schema, and validation checks that stay aligned.
+
+Install
+-------
+
+.. code-block:: bash
+
+   pip install smallwords
+
+Quick API
+---------
+
+Most integrations follow the same chain: choose a vocabulary, optionally add
+task words, build the matching prompt and portable resources, then validate the
+result.
+
+.. autofunction:: smallwords.allow_input_words
+
+.. autofunction:: smallwords.get_wordlist
+
+.. autofunction:: smallwords.list_wordlists
+
+.. autoclass:: smallwords.OutputShape
+   :members:
+
+.. autoclass:: smallwords.OutputResources
+   :members:
+
+.. autofunction:: smallwords.prompts.build_prompt
+
+.. autofunction:: smallwords.is_compliant
+
+.. autofunction:: smallwords.out_of_vocab
+
+Built-ins
+---------
+
+The installed catalog currently includes:
+
+- ``moby_898``
+- ``basic_850``
+- ``special_english_1475``
+- ``caveman_898``
+- ``pirate_898``
+
+Use :func:`smallwords.list_wordlists` to inspect the installed catalog. Use
+:func:`smallwords.get_wordlist` when a workflow needs the underlying
+specification object directly.
+
+Examples With Results
+---------------------
+
+The examples below were produced locally on April 5, 2026 with
+``llama-server`` and ``bartowski/Qwen_Qwen3-8B-GGUF:q4_k_m``.
+
+Bridge Contrast
+~~~~~~~~~~~~~~~
+
+This comparison isolates the package's core claim. A plain prompt can already
+produce a reasonable answer. However, it does not expose an explicit response
+contract. The constrained run uses the same task, adds the active vocabulary to
+the prompt, and applies the matching grammar.
+
+.. code-block:: python
+
+   from smallwords import OutputResources, OutputShape, allow_input_words
+   from smallwords.prompts import build_prompt
+
+   spec = allow_input_words("basic_850", "How does a bridge work?")
+   shape = OutputShape(max_words_per_line=24, max_lines=1)
+   prompt = build_prompt("explain", "How does a bridge work?", wordlist=spec)
+   resources = OutputResources.from_wordlist(spec, shape=shape)
+
+Plain prompt result:
+
+   A bridge connects two points, usually across a body of water or a gap, allowing people and vehicles to cross safely.
+
+Constrained result:
+
+   A bridge is a structure that helps people and things move across a river or a deep place.
+
+See ``examples/readme_bridge_contrast.py`` for the full prompt-plus-grammar run.
+
+Pirate Greeting
+~~~~~~~~~~~~~~~
+
+This example starts from the built-in ``pirate_898`` list, selects a tiny
+greeting-focused vocabulary, and then applies a matching grammar.
+
+.. code-block:: python
+
+   from smallwords import OutputResources, OutputShape, WordlistSpec, get_wordlist
+   from smallwords.prompts import build_prompt
+
+   base = get_wordlist("pirate_898")
+   spec = WordlistSpec(
+       name="pirate_898_greeting_focus",
+       words=("ahoy", "good", "matey", "meet", "to", "you"),
+       source_name="Selected surface forms from pirate_898 for the pirate greeting example",
+       source_urls=base.source_urls,
+       license_name=base.license_name,
+       allowed_punctuation=(".",),
+       variant_mode="surface_only",
+   )
+   shape = OutputShape(min_words_per_line=6, max_words_per_line=6, max_lines=1)
+   prompt = build_prompt(
+       "answer",
+       "A pirate meets a new friend on a ship. What short friendly greeting should the pirate say?",
+       wordlist=spec,
+   )
+   resources = OutputResources.from_wordlist(spec, shape=shape)
+
+Constrained result:
+
+   Ahoy matey good to meet you.
+
+Technical Rewrite
+~~~~~~~~~~~~~~~~~
+
+This example selects a compact rewrite vocabulary from ``basic_850`` that
+excludes the source terminology altogether. The output shape then forces one
+short ten-word sentence.
+
+.. code-block:: python
+
+   from smallwords import OutputResources, OutputShape, WordlistSpec, get_wordlist
+   from smallwords.prompts import build_prompt
+
+   base = get_wordlist("basic_850")
+   text = (
+       "The thermal controller derates propulsion output after the sensor array "
+       "reports an overtemperature fault."
+   )
+   spec = WordlistSpec(
+       name="basic_850_rewrite_focus",
+       words=("be", "cut", "engine", "heat", "high", "if", "power", "system", "this", "very", "when"),
+       source_name="Selected from basic_850 for the rewrite example",
+       source_urls=base.source_urls,
+       license_name=base.license_name,
+       allowed_punctuation=(".",),
+   )
+   shape = OutputShape(min_words_per_line=10, max_words_per_line=10, max_lines=1)
+   prompt = build_prompt("rewrite", text, wordlist=spec)
+   resources = OutputResources.from_wordlist(spec, shape=shape)
+
+Constrained result:
+
+   When engine heat is very high this system cuts power.
+
+The live scripts in ``examples/`` print the full prompt, grammar, schema, and
+validation details. That fuller output matters because it lets a reader inspect
+not only the answer, but also the exact constraint setup that produced it.

smallwords-0.1.0/examples/README.md

@@ -0,0 +1,35 @@
+# Examples
+
+This directory keeps three live llama.cpp examples:
+
+- `readme_bridge_contrast.py`: the bridge comparison used in the root README
+- `pirate_greeting.py`: a focused pirate greeting built from `pirate_898`
+- `rewrite_technical_passage.py`: a focused technical rewrite built from `basic_850`
+
+Create and activate a virtualenv first:
+
+```bash
+python -m venv .venv
+source .venv/bin/activate
+python -m pip install -e ".[dev]"
+```
+
+Then start a server, for example:
+
+```bash
+llama-server -hf bartowski/Qwen_Qwen3-8B-GGUF:q4_k_m --host 127.0.0.1 --port 8080 --reasoning-budget 0 --log-disable
+```
+
+Run the examples from the project root:
+
+```bash
+python examples/pirate_greeting.py
+python examples/rewrite_technical_passage.py
+python examples/readme_bridge_contrast.py
+```
+
+If your server uses a different address, set `SMALLWORDS_LLAMA_BASE_URL`.
+
+Each script prints the prompt, output shape, matching grammar and schema
+resources, the generated response, and whether the response stayed inside the
+chosen vocabulary.