CDC Report Provenance

The exemplar the long-horizon-prompting skill is built on: OpenAI's published prompt and candidate proof for the Cycle Double Cover Conjecture, produced by GPT-5.6 Sol Ultra running a 64-subagent orchestration.

Files

File	Source URL	Bytes	SHA-256
cdc_prompt.pdf	https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd98d31/cdc_prompt.pdf	123015	0e48deee28caba82ee5b4191d4c5c6ec4d62e5d27890fa7f0d2c8868f8b758f3
cdc_proof.pdf	https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd98d31/cdc_proof.pdf	325198	b4797f5053d9067329b3dcfcbf913f8bb40d13467453b1300f6d78d08460fc13

cdc_prompt.txt and cdc_proof.txt are text extractions (pypdf) kept for grep-ability and diffing. The PDFs are the authoritative artifacts.

Retrieval

• Retrieved: 2026-07-11 (this run), via scripts/fetch_report.sh.
• PDF internal CreationDate: D:20260710 for both files, consistent with the documented 2026-07-10 publication date.
• Producer: pdfTeX-1.40.28, LaTeX with hyperref (both).
• Prompt PDF: 2 pages. Proof PDF: 3 pages.

Cross-check against the repo skill reference

scripts/verify_report.py re-extracts the prompt text and asserts that every block-quote in skills/long-horizon-prompting/references/cdc-prompt-annotated.md appears verbatim in the published prompt. Result at retrieval time: 0 drifted fragments. This confirms the skill's annotated reference is faithful to the primary source, not paraphrased.

Caveats (carried from the skill, do not drop)

• The candidate proof had no independent peer review, no Lean/Coq formalization, and no arXiv posting at publication time. The validated artifact of interest is the prompt structure, not the theorem.
• No public ablation isolates which prompt elements carried the result. Per-element evidence in the skill comes from independent academic work (see references/research-evidence.md), not from this single run.
• The proof's own statement of AI use: "The proof in this note is entirely due to GPT 5.6 Sol Ultra and the writeup with Codex (with GPT 5.6 Sol)."