| """Build static English reading pages from the research Markdown documents.""" |
| import html |
| import json |
| import re |
| import subprocess |
| from pathlib import Path |
| from urllib.parse import urlsplit |
|
|
| from bs4 import BeautifulSoup |
| from markdown_it import MarkdownIt |
|
|
| ROOT = Path(__file__).resolve().parent.parent |
| DOCS = ROOT / "outputs/cliff_realistic_dense_v1" |
| SITE = ROOT / "deployments/cloudflare/site" |
| DEST = SITE / "materials" |
| HF = "https://huggingface.co/datasets/0721alex/a-plan-is-not-a-path" |
| PACK = HF + "/resolve/main/outputs/english_research_pack.zip?download=true" |
| PAGES = [ |
| ("submission", "submission_en.md", "Application", "The question and the study", "Personal experience, a systematic evaluation and proposals for better LLM assistance.", "01"), |
| ("methods", "methods_and_results_en.md", "Methods & results", "Methods and complete results", "The pinned checkpoint, full experimental settings, every cash level and reproduction commands.", "02"), |
| ("theory", "theory_and_improvement_en.md", "Theory & improvements", "From diagnosis to better planning", "The economic mechanism and specific proposals for data, supervision and institutional support.", "03"), |
| ("sources", "source_and_development_en.md", "Sources & development", "Sources and development history", "Published references, synthetic assumptions and every stage of the experiment's development.", "04"), |
| ("data", "data_guide_en.md", "Data guide", "Inside the evaluation data", "A guide to the original records, reading copies, fields and reproduction workflow.", "05"), |
| ("example", "example_task_en.md", "Example task", "The task, in full", "Read all six notices for one saved request, with the exact amounts, dates and funding conditions.", "06"), |
| ("essay", "essay_en.md", "Full essay", "The full research essay", "The extended account, interactive-study captions and complete academic references.", "07"), |
| ] |
| BY_FILE = {p[1]: p[0] for p in PAGES} |
| MARKDOWN = MarkdownIt("commonmark", {"html": True}).enable("table") |
|
|
|
|
| def clean_source(text): |
| return text.split("<!-- /edition -->", 1)[-1].lstrip() |
|
|
|
|
| def render_math(text): |
| expressions = [] |
| def take(match): |
| expr = match.group(1) or match.group(2) |
| expressions.append({"tex": expr.strip(), "display": match.group(1) is not None}) |
| return f"<div>MATHTOKEN{len(expressions)-1}</div>" if match.group(1) else f"MATHTOKEN{len(expressions)-1}" |
| text = re.sub(r"\$\$(.*?)\$\$|\$([^$\n]+)\$", take, text, flags=re.S) |
| if not expressions: |
| return text |
| js = "const fs=require('fs'), k=require('./code/vendor/katex.cjs');process.stdout.write(JSON.stringify(JSON.parse(fs.readFileSync(0,'utf8')).map(e=>k.renderToString(e.tex,{output:'mathml',displayMode:e.display,throwOnError:true}))));" |
| result = subprocess.run(["node", "-e", js], input=json.dumps(expressions), text=True, capture_output=True, cwd=ROOT, check=True) |
| for i, rendered in reversed(list(enumerate(json.loads(result.stdout)))): |
| text = text.replace(f"MATHTOKEN{i}", rendered) |
| return text |
|
|
|
|
| def fix_link(href): |
| if href.startswith(("https:", "http:", "#", "mailto:")): |
| return href |
| part = urlsplit(href) |
| name = Path(part.path).name |
| if name in BY_FILE: |
| return BY_FILE[name] + ".html" + ("#" + part.fragment if part.fragment else "") |
| if name == "report.pdf": |
| return "../assets/report.pdf" |
| if part.path.startswith("figures_en/"): |
| return "../assets/figures/" + Path(name).stem + ".svg" |
| local = (DOCS / part.path).resolve() |
| relative = local.relative_to(ROOT).as_posix() |
| kind = "tree" if local.is_dir() else "blob" if local.suffix in {".md", ".py", ".txt"} else "resolve" |
| return HF + "/" + kind + "/main/" + relative + ("#" + part.fragment if part.fragment else "") |
|
|
|
|
| def body_and_toc(text): |
| text = re.sub(r"^# .+\n", "", clean_source(text), count=1) |
| soup = BeautifulSoup(MARKDOWN.render(render_math(text)), "html.parser") |
| toc = [] |
| for heading in soup.select("h2, h3"): |
| ident = re.sub(r"[^a-z0-9]+", "-", heading.get_text().lower()).strip("-") |
| heading["id"] = heading.get("id", ident) |
| if heading.name == "h2": |
| toc.append((heading["id"], heading.get_text())) |
| for tag in soup.select("a[href], img[src]"): |
| attr = "href" if tag.name == "a" else "src" |
| tag[attr] = fix_link(tag[attr]) |
| for table in soup.select("table"): |
| wrap = soup.new_tag("div", attrs={"class": "table-scroll", "tabindex": "0", "aria-label": "Scrollable data table"}) |
| table.wrap(wrap) |
| for image in soup.select("img"): |
| image["loading"] = "lazy" |
| figure = soup.new_tag("figure") |
| image.wrap(figure) |
| if image.get("alt"): |
| caption = soup.new_tag("figcaption") |
| caption.string = image["alt"] |
| figure.append(caption) |
| return str(soup), toc |
|
|
|
|
| def shell(title, content, active="", description="Research documents and original experimental records."): |
| links = "".join(f'<a href="{slug}.html" {"aria-current=page" if slug==active else ""}><span>{number}</span>{label}</a>' for slug, _, label, _, _, number in PAGES) |
| return f'''<!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><meta name="description" content="{html.escape(description, quote=True)}"><title>{html.escape(title)} · A Plan Is Not a Path</title><link rel="stylesheet" href="library.css"><script src="library.js" defer></script></head><body> |
| <a class="skip" href="#main">Skip to content</a><div class="reading-progress" aria-hidden="true"></div> |
| <header class="library-header"><a class="brand" href="index.html"><svg viewBox="0 0 28 32" aria-hidden="true"><path d="M3 28V5h10v11m5 0h7v12" fill="none" stroke="currentColor" stroke-width="1.8"/><circle cx="3" cy="28" r="2.7" fill="currentColor"/><circle cx="25" cy="28" r="2.7" fill="currentColor"/></svg>A Plan Is Not a Path<span>Research library</span></a><nav aria-label="Research links"><a href="../?lang=en">Interactive study ↗</a><a href="{HF}">Hugging Face ↗</a><a class="paper-link" href="../assets/report.pdf">Read the paper ↗</a></nav></header> |
| {content} |
| <footer class="library-footer"><a class="footer-brand" href="index.html">A Plan Is Not a Path</a><p>One study of cash flow in LLM planning.<br>All 504 outcomes remain available.</p><div><a href="{PACK}">Research pack ↓</a><a href="{HF}/tree/main">Original data & code ↗</a></div></footer> |
| <template id="library-menu">{links}</template></body></html>''' |
|
|
|
|
| def main(): |
| DEST.mkdir(parents=True, exist_ok=True) |
| cards = [] |
| for i, (slug, filename, label, title, description, number) in enumerate(PAGES): |
| body, toc = body_and_toc((DOCS / filename).read_text()) |
| contents = "".join(f'<a href="#{ident}">{html.escape(text)}</a>' for ident, text in toc) |
| previous = PAGES[i-1] if i else None |
| following = PAGES[i+1] if i+1 < len(PAGES) else None |
| neighbours = (f'<a href="{previous[0]}.html"><small>Previous</small>← {previous[2]}</a>' if previous else '<a href="index.html"><small>Collection</small>← Research library</a>') |
| neighbours += f'<a href="{following[0]}.html"><small>Continue reading</small>{following[2]} →</a>' if following else '<a href="../?lang=en"><small>Explore</small>Interactive study ↗</a>' |
| menu = "".join(f'<a href="{p[0]}.html" {"aria-current=page" if p[0]==slug else ""}><span>{p[5]}</span>{p[2]}</a>' for p in PAGES) |
| content = f'''<main id="main" class="document-layout"><aside class="document-menu"><a class="all-materials" href="index.html">← All materials</a><p class="eyebrow">In this collection</p><nav aria-label="Documents">{menu}</nav><a class="download-side" href="{PACK}">Download the collection ↓</a></aside><div class="document-main"><div class="doc-heading"><p class="eyebrow">Research document / {number}</p><h1>{title}</h1><p class="standfirst">{description}</p><div class="doc-actions"><a href="{HF}/blob/main/outputs/cliff_realistic_dense_v1/{filename}">View on Hugging Face ↗</a><a href="{HF}/resolve/main/outputs/cliff_realistic_dense_v1/{filename}?download=true">Markdown ↓</a><button type="button" onclick="window.print()">Print</button></div></div><details class="mobile-toc"><summary>In this document</summary><nav>{contents}</nav></details><article class="prose">{body}</article><nav class="next-reading" aria-label="Continue reading">{neighbours}</nav></div><aside class="page-toc"><p class="eyebrow">In this document</p><nav>{contents}</nav></aside></main>''' |
| (DEST / (slug + ".html")).write_text(shell(title, content, slug, description)) |
| cards.append(f'<a class="material-card" href="{slug}.html"><span class="card-number">{number}</span><span class="card-arrow">↗</span><h3>{label}</h3><p>{description}</p><span class="card-format">Read online <span>Markdown available</span></span></a>') |
| content = f'''<main id="main" class="collection"><section class="collection-hero"><div><p class="eyebrow">A Plan Is Not a Path / Research collection</p><h1>Research<br><em>library.</em></h1><p class="standfirst">The question, the evidence and the work behind it.<br>Research documents alongside the original experimental records.</p><div class="hero-actions"><a class="primary" href="submission.html">Start with the application <span>↗</span></a><a href="{PACK}">Download the research pack ↓</a></div></div><aside class="collection-evidence"><p class="eyebrow">The finding / Qwen3.5-4B</p><p class="evidence-question">When an LLM says its plan is feasible, can it pay every bill?</p><div class="evidence-pair"><div><strong>21<span>/24</span></strong><p>promised feasibility</p></div><div><strong>0<span>/24</span></strong><p>completed the year</p></div></div><p class="evidence-note">At RMB 50,000 in starting cash.<br>Same twenty-four responses. Full-year execution.</p><a href="methods.html#complete-results">Read the complete results ↗</a></aside></section><section class="library-documents" aria-labelledby="documents-title"><div class="section-heading"><p class="eyebrow">Reading sequence</p><h2 id="documents-title">Inside the study</h2><p>Begin with the application. Use the methods and source record to examine how the result was obtained.</p></div><div class="material-grid">{''.join(cards)}</div></section><section class="data-shelf"><div><p class="eyebrow">Original experiment</p><h2>504 responses.<br>Every response is available.</h2><p>The evaluated prompts and original model prose are Chinese. English prompt copies and outcome records make all 504 tasks and scored results accessible; the original responses remain available alongside them.</p></div><div class="data-links"><a href="{HF}/resolve/main/data/cliff_realistic_dense_v1/english_reading/requests.en.jsonl"><strong>504 prompt reading copies</strong><span>JSONL · same request IDs ↓</span></a><a href="{HF}/resolve/main/data/cliff_realistic_dense_v1/english_reading/outcomes.en.jsonl"><strong>504 outcome records</strong><span>JSONL · saved judgments and execution results ↓</span></a><a href="{HF}/tree/main/results"><strong>Original responses and scores</strong><span>Full final answers, thinking and runtime ↗</span></a><a href="{HF}/tree/main/code"><strong>Reproduction code</strong><span>Generation, execution and reference search ↗</span></a></div></section><section class="paper-shelf"><div><p class="eyebrow">The short paper</p><h2>The study in five pages.</h2><p>Methods, three figures, complete findings and development context in one report.</p><a class="primary" href="../assets/report.pdf">Open the paper <span>↗</span></a></div><a href="../assets/report.pdf"><img src="../assets/figures/figure1_absolute_cash_claim_execution.svg" alt="The paper's first figure: promises, structural validity and execution across starting cash."></a></section></main>''' |
| (DEST / "index.html").write_text(shell("Research library", content)) |
| print(f"Built {len(PAGES)} English documents and the research-library index.") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
