{
  "checked_at": "2026-09-07",
  "claims": [
    {"id":"context-occupants","claim":"System prompts, messages, tool results, documents, images, tool definitions, and generated output can count toward a model context window.","source":"https://platform.claude.com/docs/en/build-with-claude/context-windows","publisher":"Anthropic","source_type":"primary documentation"},
    {"id":"english-token-rule","claim":"A useful English rule of thumb is about four characters or three-quarters of a word per token; exact counts vary by model and encoding.","source":"https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them","publisher":"OpenAI","source_type":"primary documentation"},
    {"id":"tokenizer-variation","claim":"Tokenizer pipelines can differ in normalization, pre-tokenization, word/subword modeling, and post-processing.","source":"https://huggingface.co/docs/tokenizers/main/api/tokenizer","publisher":"Hugging Face","source_type":"primary documentation"},
    {"id":"capacity-not-quality","claim":"A larger context window does not by itself guarantee better use of distant context.","source":"https://platform.claude.com/docs/en/build-with-claude/context-windows","publisher":"Anthropic","source_type":"primary documentation"}
  ],
  "planning_assumptions": {"pdf_report_tokens_per_page":667,"prose_tokens_per_word":1.333,"code_tokens_per_line":10,"transcript_tokens_per_minute":200,"agent_tokens_per_turn":2500,"note":"Editable planning defaults, not provider specifications. Use a target-model tokenizer or token-counting API for exact input counts."}
}