# %% import csv import os from pathlib import Path import anthropic from dotenv import load_dotenv load_dotenv("/Users/jhelvy/gh/web/jhelvy_quarto/.env") # %% BLOG_DIR = Path("/Users/jhelvy/gh/web/jhelvy_quarto/blog/2026-05-12-quarto-optimal-claude-output") OUTPUTS_DIR = BLOG_DIR / "experiments" / "outputs" DATA_DIR = BLOG_DIR / "data" CSV_PATH = DATA_DIR / "token_results.csv" OUTPUTS_DIR.mkdir(parents=True, exist_ok=True) DATA_DIR.mkdir(parents=True, exist_ok=True) # %% MODEL = "claude-haiku-4-5" DOCUMENT_SPEC = """ A summary report on electric vehicle charging standards, comparing CCS (Combined Charging System), CHAdeMO, and NACS (North American Charging Standard, formerly Tesla). The report must include: 1. An introduction explaining the landscape 2. A comparison table with columns: Standard, Origin, Max Power (kW), Connector Shape, Adoption Region 3. A pros and cons section for each of the three standards 4. A conclusion with a practical recommendation """.strip() # %% CONDITIONS = [ ( "direct_html", "direct", "html", ( "Produce a COMPLETE, self-contained HTML file (DOCTYPE, with " "embedded CSS, ) for the following document. " "Output ONLY the raw HTML — no markdown fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "html", ), ( "direct_markdown", "direct", "markdown", ( "Produce a plain Markdown document for the following. " "Output ONLY the Markdown — no code fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "md", ), ( "direct_latex", "direct", "pdf", ( "Produce a COMPLETE, compilable LaTeX document (\\documentclass, " "\\usepackage declarations, \\begin{document}...\\end{document}) for " "the following. Output ONLY the LaTeX — no markdown fences, no " "explanation.\n\n" f"{DOCUMENT_SPEC}" ), "tex", ), ( "direct_docx_xml", "direct", "word", ( "Produce the complete Office Open XML content for the " "word/document.xml file inside a .docx package for the following " "document. Include the full XML with proper namespace declarations. " "Output ONLY the XML — no markdown fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "xml", ), ( "quarto_html", "quarto", "html", ( "Produce a Quarto (.qmd) document with 'format: html' in the YAML " "front matter for the following document. Use standard Quarto/Markdown " "syntax only — no raw HTML blocks. " "Output ONLY the .qmd content — no code fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "qmd", ), ( "quarto_pdf", "quarto", "pdf", ( "Produce a Quarto (.qmd) document with 'format: pdf' in the YAML " "front matter for the following document. Use standard Quarto/Markdown " "syntax — no raw LaTeX blocks. " "Output ONLY the .qmd content — no code fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "qmd", ), ( "quarto_word", "quarto", "word", ( "Produce a Quarto (.qmd) document with 'format: docx' in the YAML " "front matter for the following document. Use standard Quarto/Markdown " "syntax. " "Output ONLY the .qmd content — no code fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "qmd", ), ( "quarto_markdown", "quarto", "markdown", ( "Produce a Quarto (.qmd) document with 'format: gfm' in the YAML " "front matter for the following document. Use standard Quarto/Markdown " "syntax. " "Output ONLY the .qmd content — no code fences, no explanation.\n\n" f"{DOCUMENT_SPEC}" ), "qmd", ), ] # %% client = anthropic.Anthropic(api_key=os.environ.get("ANTHROPIC_API_KEY")) # %% CSV_FIELDS = ["condition", "format_type", "output_type", "input_tokens", "output_tokens", "char_count"] rows = [] for condition_id, format_type, output_type, prompt, ext in CONDITIONS: output_file = OUTPUTS_DIR / f"{condition_id}.{ext}" if output_file.exists(): print(f" [skip] {condition_id} — output already exists") continue print(f" [run ] {condition_id} …", end=" ", flush=True) response = client.messages.create( model=MODEL, max_tokens=8192, messages=[{"role": "user", "content": prompt}], ) raw_text = next((b.text for b in response.content if b.type == "text"), "") output_file.write_text(raw_text, encoding="utf-8") row = { "condition": condition_id, "format_type": format_type, "output_type": output_type, "input_tokens": response.usage.input_tokens, "output_tokens": response.usage.output_tokens, "char_count": len(raw_text), } rows.append(row) print(f"in={row['input_tokens']:,} out={row['output_tokens']:,} chars={row['char_count']:,}") # %% existing = {} if CSV_PATH.exists(): with CSV_PATH.open(newline="") as f: for r in csv.DictReader(f): existing[r["condition"]] = r for row in rows: existing[row["condition"]] = row with CSV_PATH.open("w", newline="") as f: writer = csv.DictWriter(f, fieldnames=CSV_FIELDS) writer.writeheader() writer.writerows(existing.values()) print(f"\nResults written to {CSV_PATH}")