site2voice 0.2.1__tar.gz

site2voice-0.2.1/CHANGELOG.md

@@ -0,0 +1,18 @@
+# Changelog
+
+## 0.2.1
+
+- Added the `pypi` GitHub environment claim to the Trusted Publishing workflow.
+
+## 0.2.0
+
+- Added `site2voice bench` for before/after voice-alignment scoring.
+- Added style fingerprints for heading shape, paragraph rhythm, CTA shape, CTA
+  verbs, and lexical variety.
+- Added copy-safety and claim-boundary gates.
+- Added synthetic editorial fixtures and benchmark report.
+- Added source-candidate, benchmark, awesome-eligibility, and harness docs.
+
+## 0.1.0
+
+- Initial CLI for generating `VOICE.md` from URL or local HTML copy.

site2voice-0.2.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Sihyeon Jeon
+
+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.

site2voice-0.2.1/MANIFEST.in

@@ -0,0 +1,3 @@
+include CHANGELOG.md
+recursive-include docs *.md
+recursive-include examples *.html *.md

site2voice-0.2.1/PKG-INFO

@@ -0,0 +1,134 @@
+Metadata-Version: 2.4
+Name: site2voice
+Version: 0.2.1
+Summary: Generate AI-agent VOICE.md files from website copy and CTAs.
+Author: Sihyeon Jeon
+License-Expression: MIT
+Keywords: agents,voice,copywriting,markdown,branding
+Classifier: Programming Language :: Python :: 3
+Classifier: Environment :: Console
+Classifier: Topic :: Text Processing :: Markup :: HTML
+Classifier: Topic :: Software Development :: Documentation
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Dynamic: license-file
+
+# site2voice
+
+**Generate `VOICE.md` from any website.**
+
+`site2voice` reads website copy and writes a small Markdown brief that tells an
+AI coding agent how the site sounds: headings, CTAs, navigation labels, sentence
+shape, repeated vocabulary, and claim boundaries.
+
+```bash
+pipx install site2voice
+
+site2voice https://example.com --out VOICE.md
+```
+
+From a repo clone, run the included benchmark fixture:
+
+```bash
+site2voice examples/saas-home.html --format json
+site2voice bench examples/editorial-home.html examples/before-copy.md examples/after-copy.md
+```
+
+## Why
+
+`DESIGN.md` helps agents stop guessing visual style. `VOICE.md` helps them stop
+guessing copy style.
+
+Drop the generated file into a project and tell the agent:
+
+```text
+Use @VOICE.md for landing-page copy, headings, CTAs, and UI microcopy.
+```
+
+## Output
+
+```md
+# VOICE.md
+
+## Voice Summary
+
+- Overall tone: explanatory, action-oriented, trust-forward.
+- Sentence shape: about 20.4 words per sentence.
+- Main vocabulary: `teams`, `security`, `pricing`, `launch`.
+- Common CTAs: `Start free`, `Book a demo`, `See pricing`.
+
+## Agent Rules
+
+- Start with a concrete user outcome before describing implementation details.
+- Prefer short active sentences and visible verbs from the CTA list.
+- Do not invent compliance, security, customer, or performance claims.
+```
+
+The real output also includes a small style fingerprint for heading length,
+paragraph rhythm, CTA shape, CTA verbs, and lexical variety.
+
+## What It Does
+
+- Reads a URL or local HTML file.
+- Extracts title, meta description, headings, links, buttons, and paragraphs.
+- Finds CTA candidates from short action-led links/buttons.
+- Measures average sentence length.
+- Extracts a compact style fingerprint: heading shape, paragraph rhythm,
+  CTA shape, CTA verbs, and lexical variety.
+- Builds a repeated-vocabulary lexicon.
+- Writes Markdown or JSON.
+- Benchmarks candidate copy against a source voice profile.
+- Gates against unsupported claims and copied spans.
+- Uses only the Python standard library.
+
+## Benchmark
+
+`site2voice bench` compares candidate copy against measurable source signals:
+sentence length, vocabulary overlap, CTA shape, tone labels, heading shape,
+claim boundaries, and copy safety.
+
+```bash
+site2voice bench examples/editorial-home.html \
+  examples/before-copy.md \
+  examples/after-copy.md \
+  --out examples/editorial-benchmark.md
+```
+
+| Candidate | Result | Overall | Lexicon | Copy safety |
+| --- | --- | ---: | ---: | ---: |
+| `after-copy` | PASS | 83.8 | 70.0 | 93.2 |
+| `before-copy` | FAIL | 36.6 | 0.0 | 100.0 |
+
+The benchmark rewards measurable voice alignment without rewarding verbatim
+copying.
+
+## What It Is Not
+
+- Not an official brand guideline.
+- Not a DESIGN.md visual-token extractor.
+- Not a crawler for private pages or authenticated apps.
+- Not an LLM prompt that copies a site's prose.
+
+## Develop
+
+```bash
+python3 -m pip install -e .
+make test
+make bench
+site2voice examples/saas-home.html --out examples/saas-VOICE.md
+```
+
+## Links
+
+- [Research](docs/research.md)
+- [Benchmark](docs/benchmark.md)
+- [Voice patterns](docs/voice-patterns.md)
+- [Source candidates](docs/source-candidates.md)
+- [Awesome eligibility](docs/awesome-eligibility.md)
+- [Harness](docs/harness.md)
+- [Launch kit](docs/launch-kit.md)
+
+## License
+
+MIT

site2voice-0.2.1/README.md

@@ -0,0 +1,118 @@
+# site2voice
+
+**Generate `VOICE.md` from any website.**
+
+`site2voice` reads website copy and writes a small Markdown brief that tells an
+AI coding agent how the site sounds: headings, CTAs, navigation labels, sentence
+shape, repeated vocabulary, and claim boundaries.
+
+```bash
+pipx install site2voice
+
+site2voice https://example.com --out VOICE.md
+```
+
+From a repo clone, run the included benchmark fixture:
+
+```bash
+site2voice examples/saas-home.html --format json
+site2voice bench examples/editorial-home.html examples/before-copy.md examples/after-copy.md
+```
+
+## Why
+
+`DESIGN.md` helps agents stop guessing visual style. `VOICE.md` helps them stop
+guessing copy style.
+
+Drop the generated file into a project and tell the agent:
+
+```text
+Use @VOICE.md for landing-page copy, headings, CTAs, and UI microcopy.
+```
+
+## Output
+
+```md
+# VOICE.md
+
+## Voice Summary
+
+- Overall tone: explanatory, action-oriented, trust-forward.
+- Sentence shape: about 20.4 words per sentence.
+- Main vocabulary: `teams`, `security`, `pricing`, `launch`.
+- Common CTAs: `Start free`, `Book a demo`, `See pricing`.
+
+## Agent Rules
+
+- Start with a concrete user outcome before describing implementation details.
+- Prefer short active sentences and visible verbs from the CTA list.
+- Do not invent compliance, security, customer, or performance claims.
+```
+
+The real output also includes a small style fingerprint for heading length,
+paragraph rhythm, CTA shape, CTA verbs, and lexical variety.
+
+## What It Does
+
+- Reads a URL or local HTML file.
+- Extracts title, meta description, headings, links, buttons, and paragraphs.
+- Finds CTA candidates from short action-led links/buttons.
+- Measures average sentence length.
+- Extracts a compact style fingerprint: heading shape, paragraph rhythm,
+  CTA shape, CTA verbs, and lexical variety.
+- Builds a repeated-vocabulary lexicon.
+- Writes Markdown or JSON.
+- Benchmarks candidate copy against a source voice profile.
+- Gates against unsupported claims and copied spans.
+- Uses only the Python standard library.
+
+## Benchmark
+
+`site2voice bench` compares candidate copy against measurable source signals:
+sentence length, vocabulary overlap, CTA shape, tone labels, heading shape,
+claim boundaries, and copy safety.
+
+```bash
+site2voice bench examples/editorial-home.html \
+  examples/before-copy.md \
+  examples/after-copy.md \
+  --out examples/editorial-benchmark.md
+```
+
+| Candidate | Result | Overall | Lexicon | Copy safety |
+| --- | --- | ---: | ---: | ---: |
+| `after-copy` | PASS | 83.8 | 70.0 | 93.2 |
+| `before-copy` | FAIL | 36.6 | 0.0 | 100.0 |
+
+The benchmark rewards measurable voice alignment without rewarding verbatim
+copying.
+
+## What It Is Not
+
+- Not an official brand guideline.
+- Not a DESIGN.md visual-token extractor.
+- Not a crawler for private pages or authenticated apps.
+- Not an LLM prompt that copies a site's prose.
+
+## Develop
+
+```bash
+python3 -m pip install -e .
+make test
+make bench
+site2voice examples/saas-home.html --out examples/saas-VOICE.md
+```
+
+## Links
+
+- [Research](docs/research.md)
+- [Benchmark](docs/benchmark.md)
+- [Voice patterns](docs/voice-patterns.md)
+- [Source candidates](docs/source-candidates.md)
+- [Awesome eligibility](docs/awesome-eligibility.md)
+- [Harness](docs/harness.md)
+- [Launch kit](docs/launch-kit.md)
+
+## License
+
+MIT

site2voice-0.2.1/docs/awesome-eligibility.md

@@ -0,0 +1,51 @@
+# Awesome Eligibility
+
+Current status: **not ready to submit broadly**.
+
+The project now has a public repo, a release, examples, tests, and a benchmark.
+That is enough for lightweight launch posts, but not enough for high-signal
+awesome-list submissions. Awesome maintainers usually expect visible usage,
+clear novelty, and a stable public surface.
+
+## Best Fit
+
+| List family | Fit | Why |
+| --- | --- | --- |
+| Vibe coding / context engineering | High later | `site2voice` creates agent context files and includes an eval path. |
+| Prompt engineering tools | Medium later | Useful for prompt/context prep, but not a prompt library. |
+| AI for design | Medium-low | Related to design handoff, but voice/copy is not visual design. |
+| Claude Code tools | Not yet | Some lists require high star counts or mature adoption. |
+
+## Submit When
+
+- v0.2+ release is public and installable.
+- README shows one command, one result, and one before/after benchmark.
+- At least one real-world demo or user workflow exists.
+- No generated third-party prose is committed.
+- The submission can be phrased as a context-engineering tool, not a brand
+  imitation tool.
+
+## Candidate Positioning
+
+Use this description:
+
+> Generate compact `VOICE.md` context files from website copy, then benchmark
+> whether AI-written copy matches the source voice without copying spans or
+> inventing claims.
+
+Avoid:
+
+- asking for stars;
+- submitting to unrelated design lists;
+- claiming official brand replication;
+- using EYESMAG, Highsnobiety, Stripe, or Linear outputs as public examples
+  unless only derived metrics are shown.
+
+## References
+
+- [no-fluff/awesome-vibe-coding](https://github.com/no-fluff/awesome-vibe-coding)
+- [taskade/awesome-vibe-coding](https://github.com/taskade/awesome-vibe-coding)
+- [subinium/awesome-claude-code](https://github.com/subinium/awesome-claude-code)
+- [promptslab/awesome-prompt-engineering](https://github.com/promptslab/awesome-prompt-engineering)
+- [allanjsx/awesome-ai-for-design](https://github.com/allanjsx/awesome-ai-for-design)
+

site2voice-0.2.1/docs/benchmark.md

@@ -0,0 +1,47 @@
+# Benchmark
+
+`site2voice bench` checks whether candidate copy follows a source voice profile.
+It is deterministic and uses only local text statistics.
+
+```bash
+site2voice bench examples/editorial-home.html \
+  examples/before-copy.md \
+  examples/after-copy.md \
+  --out examples/editorial-benchmark.md
+```
+
+## Metrics
+
+| Metric | What it checks |
+| --- | --- |
+| Sentence fit | Candidate average sentence length vs source average. |
+| Lexicon fit | Whether candidate reuses source vocabulary without requiring exact prose. |
+| CTA fit | Whether short action-led lines match observed CTA vocabulary. |
+| Tone fit | Overlap between deterministic tone labels. |
+| Heading fit | Whether heading length/shape is close to the source. |
+| Claim safety | Penalizes unsupported security, performance, customer, pricing, and AI claims. |
+| Copy safety | Penalizes long shared spans and high 5-gram overlap with source snippets. |
+
+The reference profile also keeps a style fingerprint: average heading length,
+average paragraph length, average CTA length, CTA verbs, lexical variety, and
+punctuation counts.
+
+## Gates
+
+A candidate passes when:
+
+- overall score is at least 75;
+- copy safety is at least 85;
+- claim safety is at least 75.
+
+This matters because direct copying can look stylistically aligned while being
+the wrong behavior. Copy safety is a gate, not just a metric.
+
+## Example
+
+| Candidate | Result | Overall | Sentence | Lexicon | CTA | Tone | Heading | Claim safety | Copy safety |
+| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| `after-copy` | PASS | 83.8 | 95.8 | 70.0 | 50.0 | 100.0 | 93.8 | 100.0 | 93.2 |
+| `before-copy` | FAIL | 36.6 | 57.5 | 0.0 | 0.0 | 0.0 | 62.5 | 100.0 | 100.0 |
+
+See [editorial-benchmark.md](../examples/editorial-benchmark.md).

site2voice-0.2.1/docs/harness.md

@@ -0,0 +1,34 @@
+# Harness
+
+`site2voice` is built as a small agent-orchestration harness:
+
+1. **Scout** finds relevant public sources and decides whether they are safe as
+   live examples or should be represented by synthetic fixtures.
+2. **Profiler** runs `site2voice SOURCE --format json` and extracts measurable
+   voice signals: sentence shape, heading shape, CTA verbs, lexical variety,
+   and repeated vocabulary.
+3. **Writer** uses the generated `VOICE.md` as context to write new candidate
+   copy.
+4. **Evaluator** runs `site2voice bench SOURCE before.md after.md` and checks
+   alignment, claim boundaries, and copy safety.
+5. **Publisher** updates examples, docs, releases, and launch notes.
+
+## Claude Operator Review
+
+Claude operator review agreed with the direction and recommended:
+
+- deterministic style fingerprints instead of vague tone claims;
+- a scorer before additional promotion;
+- explicit copy-safety and unsupported-claim gates;
+- source candidates as live examples, not committed third-party prose.
+
+## Current Team Findings
+
+- Editorial/magazine sources are useful for tone, but checked-in examples should
+  be synthetic.
+- `site2voice` should be submitted to awesome lists only after release/demo
+  evidence is stronger.
+- Best awesome-list fit found so far: vibe-coding/context-engineering lists, not
+  visual design lists.
+
+See [awesome-eligibility.md](awesome-eligibility.md) for submission gates.

site2voice-0.2.1/docs/launch-kit.md

@@ -0,0 +1,38 @@
+# Launch Kit
+
+## One-Line Pitch
+
+`site2voice` turns any website into a small `VOICE.md` file for AI coding
+agents.
+
+## Short Post
+
+I made `site2voice`, a tiny CLI that generates agent-readable `VOICE.md` files
+from website copy.
+
+It extracts headings, CTA text, navigation labels, sentence length, and repeated
+vocabulary, then writes a concise voice brief for Claude Code, Cursor, Codex, or
+any other coding agent.
+
+```bash
+pipx install site2voice
+site2voice https://example.com --out VOICE.md
+
+# from a repo clone
+site2voice bench examples/editorial-home.html examples/before-copy.md examples/after-copy.md
+```
+
+No LLM dependency. No browser dependency. The goal is simple: give agents voice
+and positioning context before they write new UI copy.
+
+The benchmark path shows whether the generated copy actually moved closer to
+the source voice, without rewarding copied spans.
+
+Repo: https://github.com/SihyeonJeon/site2voice
+
+## Avoid
+
+- Do not claim official brand approval.
+- Do not paste long generated excerpts from third-party sites.
+- Do not ask for stars.
+- Do not spam unrelated design repositories.

site2voice-0.2.1/docs/research.md

@@ -0,0 +1,69 @@
+# Research
+
+## What Exists
+
+DESIGN.md-style repos and services give AI coding agents visual design context:
+colors, typography, spacing, layout patterns, and component guidance.
+
+Observed categories:
+
+- DESIGN.md libraries for ready-made visual systems.
+- URL-to-DESIGN.md generators for design tokens and visual patterns.
+- Agent memory files such as `AGENTS.md`, `CLAUDE.md`, and `DESIGN.md`.
+
+## Gap
+
+Visual style is only half of what an agent needs. When generating landing pages,
+docs, onboarding flows, and UI microcopy, the agent also needs:
+
+- message hierarchy;
+- CTA language;
+- navigation vocabulary;
+- sentence length;
+- claim boundaries;
+- words to reuse and words to avoid.
+
+`site2voice` targets that gap with a deterministic `VOICE.md` generator.
+
+## New Evidence Path
+
+The project now includes `site2voice bench`, which measures whether candidate
+copy follows a source voice profile. This turns `VOICE.md` from a descriptive
+artifact into a testable workflow:
+
+```bash
+site2voice SOURCE --out VOICE.md
+site2voice bench SOURCE before.md after.md
+```
+
+The benchmark scores sentence shape, vocabulary overlap, CTA shape, heading
+shape, tone labels, claim boundaries, and copy safety.
+
+## Product Bet
+
+The README should be short:
+
+```bash
+site2voice https://example.com --out VOICE.md
+```
+
+The output should answer:
+
+- What does this site sound like?
+- What CTAs does it use?
+- What vocabulary should an agent reuse?
+- What claims must the agent avoid inventing?
+
+## Boundaries
+
+- No LLM dependency.
+- No screenshot or brand asset copying.
+- Do not paste long source paragraphs into the generated file.
+- Treat the output as a writing/style brief, not legal brand guidance.
+
+## References
+
+- [DESIGN.md library](https://designmd.app/)
+- [Better Stack guide to DESIGN.md](https://betterstack.com/community/guides/ai/design-md-ai/)
+- [DESIGN.MD by Parallect coverage](https://chatgate.ai/post/design-md-by-parallect)
+- [DesignMD URL-to-DESIGN.md generator](https://www.designmd.cc/)

site2voice-0.2.1/docs/source-candidates.md

@@ -0,0 +1,22 @@
+# Source Candidates
+
+Use these as live examples for local analysis. Do not commit generated outputs
+from third-party sites unless you remove source snippets and preserve only
+derived metrics.
+
+| Source | Why it is useful | Use safely |
+| --- | --- | --- |
+| [EYESMAG](https://www.eyesmag.com/) | Korean fashion/lifestyle web-magazine tone, mixed brand names, short trend labels. | Prefer live local runs or synthetic Korean fixtures. |
+| [Highsnobiety](https://www.highsnobiety.com/) | Culture, fashion, commerce, and editorial headline cadence. | Use homepage/category pages; do not republish article body output. |
+| [Hypebeast](https://hypebeast.com/) | Trend-driven streetwear/product vocabulary and title patterns. | Use as a live URL example; JS-heavy pages may need future browser support. |
+| [Monocle](https://monocle.com/) | Polished global affairs, design, travel, and city-guide register. | Good live source for nav/category/CTA patterns. |
+| [Wallpaper](https://www.wallpaper.com/) | Design, interiors, architecture, and product-focused editorial voice. | Use derived metrics, not copied article snippets. |
+| [Dazed](https://www.dazeddigital.com/) | Youth culture, fashion, music, and art phrasing. | Prefer synthetic fixtures for public examples. |
+| [Stripe](https://stripe.com/) | Developer/business SaaS claims, technical nouns, conversion CTAs. | Good contrast against editorial fixtures. |
+| [Linear](https://linear.app/) | Modern product-team copy, short headings, system-oriented vocabulary. | Good live source, but homepage content changes often. |
+| [Vercel](https://vercel.com/) | Developer platform voice, performance claims, product CTA density. | Good for claim-boundary tests. |
+
+## Fixture Policy
+
+Checked-in examples should be synthetic. Live URL extraction is for local
+analysis and should not redistribute third-party prose.

site2voice-0.2.1/docs/voice-patterns.md

@@ -0,0 +1,56 @@
+# Voice Patterns
+
+`site2voice` does not try to copy a brand or magazine. It extracts measurable
+signals that help an agent write new copy in a nearby register without pasting
+source prose.
+
+## Editorial / Magazine
+
+What to measure:
+
+- short section labels mixed with longer descriptive paragraphs;
+- named entities, places, seasons, materials, products, and collaborators;
+- restrained adjectives before concrete nouns;
+- headline length and whether titles read like indexes, blurbs, or news alerts;
+- CTA words such as `read`, `explore`, `view`, `subscribe`, or `join`.
+
+Good source families:
+
+- EYESMAG-style Korean fashion and lifestyle indexing;
+- Highsnobiety/Hypebeast-style culture and product trend coverage;
+- Monocle/Wallpaper-style design, city, travel, and object language;
+- Dazed/i-D-style youth culture and fashion framing.
+
+Use synthetic fixtures for checked-in examples. Use live URLs only for local
+analysis or private benchmark runs.
+
+## Product / Developer Sites
+
+What to measure:
+
+- outcome-first headings;
+- short conversion CTAs;
+- technical nouns and integration terms;
+- proof language, customer claims, security claims, and performance claims;
+- whether copy starts with user value or implementation detail.
+
+Good source families:
+
+- Stripe, Vercel, Linear, GitHub, and other public product pages.
+
+These sources are useful contrast sets. They expose whether candidate copy is
+accidentally drifting from editorial voice into SaaS launch-page voice.
+
+## Agent Handoff
+
+Give the agent the generated `VOICE.md`, then score the result:
+
+```bash
+site2voice https://example.com --out VOICE.md --max-snippets 0
+site2voice bench https://example.com before.md after.md
+```
+
+The first command gives the agent a voice brief. The second command verifies
+whether the resulting copy moved closer to the source profile while staying
+inside copy-safety and claim-safety gates.
+

site2voice-0.2.1/examples/after-copy.md

@@ -0,0 +1,9 @@
+# The slow return of the useful object
+
+Current Index follows the small studios, shopkeepers, and collectors turning everyday materials into a quieter kind of record.
+
+## A guide shaped by rooms, streets, and habit
+
+Read the feature
+Explore rooms
+Open the guide

site2voice-0.2.1/examples/before-copy.md

@@ -0,0 +1,9 @@
+# Powerful content for modern brands
+
+Our platform helps teams create better marketing with seamless workflows and next-generation automation.
+
+## Everything you need
+
+Use our all-in-one tools to boost productivity, improve collaboration, and unlock growth faster.
+
+Get started today

site2voice-0.2.1/examples/editorial-benchmark.md

@@ -0,0 +1,26 @@
+# Voice Benchmark
+
+Reference: `examples/editorial-home.html`
+
+## Reference Profile
+
+- Tone: explanatory
+- Average sentence length: 21.4 words
+- Lexicon: `city`, `archive`, `design`, `style`, `join`, `list`, `quiet`, `return`, `studio`, `object`, `current`, `index`
+- CTAs: `Join the list`, `Explore rooms`
+
+## Scores
+
+| Candidate | Result | Overall | Sentence | Lexicon | CTA | Tone | Heading | Claim safety | Copy safety |
+| --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| `after-copy` | **PASS** | 83.8 | 95.8 | 70.0 | 50.0 | 100.0 | 93.8 | 100.0 | 93.2 |
+| `before-copy` | **FAIL** | 36.6 | 57.5 | 0.0 | 0.0 | 0.0 | 62.5 | 100.0 | 100.0 |
+
+## Why This Is Useful
+
+The score is deterministic. It checks measurable voice signals and gates against unsupported claims and copied spans.
+
+## Candidate Evidence
+
+- `after-copy` reused source lexicon: `collectors`, `current`, `follows`, `index`, `object`, `return`, `shopkeepers`; longest shared run: 5 words.
+- `before-copy` reused source lexicon: none; longest shared run: 1 words.