Publications
Pre-experimental release materials, drafted 2026-05-25, not yet posted publicly. Use
RELEASE_CHECKLIST.mdto coordinate the publication wave when ready to ship.
| Artifact | What | Where it goes |
|---|---|---|
PAPER_v0.md |
Longform methodology paper (~6,500 words) — central document | arXiv (eventually) or just as the canonical writeup on the repo |
BLOG_POST.md |
Blog post (~2,400 words) in HuggingFace Blog markdown format | HuggingFace blog PR + personal blog / Substack / Medium |
HF_DISCUSSION_POST.md |
Repo Community-tab discussion announcing the release | This repo's Discussions tab |
TWITTER_THREAD.md |
13-tweet thread, 5-tweet short version, LinkedIn variant | X / Twitter / LinkedIn |
RELEASE_CHECKLIST.md |
Pre-flight checklist + sequencing recommendation + risk register | Internal coordination |
/CITATION.cff |
Citation File Format — HF/GitHub renders a "Cite this repository" UI from this | Repo root |
/CITATION.bib |
BibTeX equivalent | Repo root |
What this collection is and isn't
It is: a complete, self-consistent draft of a pre-experimental release announcing the methodology, integration architecture, OPSD/SDPO framing, the novel TR-DPO channel, and the spike-001/spike-005 results. Every claim is either upstream-citation-backed or empirically validated by the spikes.
It isn't: post-experimental. There are no training results yet. Spike 002–004 (~$500 GPU + a few weeks of wallclock) are the gate to a v0.1 release that adds empirical training validation.
Honest framing reused throughout
All four publication-facing documents (PAPER_v0.md, BLOG_POST.md, HF_DISCUSSION_POST.md, TWITTER_THREAD.md) include explicit "what I'm NOT claiming" sections. That framing is the publication's defense against overclaim — the work being released is methodology, integration architecture, and economic feasibility for the novel channel, not "this method works."
If anything in those documents reads as if it claims more than that, edit before posting.
Sequencing TL;DR
- HF Discussion post (lowest stakes; pre-announces the methodology)
- Blog post (anchor narrative)
- X / LinkedIn (after blog URL exists)
- arXiv (defer until v0.1 with empirical results — see
RELEASE_CHECKLIST.md)
Three-day gap between (1) and (2) lets early-feedback iterations land before the bigger announcement.