Building with Claude: Open Code, Closed PDF — Sharing Research Without Getting Scooped

I wanted to add two research papers to my site: a published conference paper, and my still-unpublished undergraduate thesis. I assumed “put the PDF up” for both. Each one turned out to need a different decision, and thinking them through changed how I see the trade-off between visibility and protection.

The problem we were solving

Two papers, two situations:

• A published IEEE conference paper on Banglish spell correction. Done, peer reviewed, out in the world.
• An unpublished thesis comparing imputation and clustering pipelines on incomplete suicide data. Defended, but no formal paper yet.

My default for both was the same: host the PDF, add a download button. For both, that default was wrong — for opposite reasons.

What Claude suggested

For the published paper, the issue was copyright. The camera-ready PDF carries the IEEE header and IEEE’s copyright. Self-hosting that exact file isn’t really mine to do — publishers let you share an accepted manuscript or link to the official page, but not redistribute their typeset version. So instead of a hosted file, the post links to IEEE Xplore as the source. Visibility without violating anyone’s rights.

For the unpublished thesis, my worry was the opposite: if I post the whole thing, can someone steal it before I publish a formal paper? Claude’s reframing was the useful part. The work’s code and dataset were already public on GitHub under open licenses — so hiding the writeup protected almost nothing. And a public, dated, attributed document doesn’t expose you to theft; it establishes authorship. Priority is protected by publishing first with your name on it, not by hiding. A thesis is also a legitimate scholarly output that doesn’t block a later journal submission.

We still chose the cautious middle path — show the abstract and link the GitHub repo, hold the full PDF until a formal paper lands:

type: paper
abstract: >-
  An exploratory study on 1,617 newspaper-reported suicide cases...
link: https://github.com/akashmony01/imputation-clustering-comparison

So: published paper, link out (copyright). Unpublished thesis, share the substance, hold the PDF (caution, not fear).

What I got wrong

I treated “share research” as one decision with one answer. It’s at least two different decisions — can I legally share this? (copyright) and should I share this yet? (strategy) — and they pull in different directions for different papers. I’d have either broken IEEE’s copyright on one or needlessly hidden my own work on the other.

I also had the protection logic backwards on the thesis. I thought visibility was the risk. For authorship, visibility is the defense: the public, timestamped, attributed version is the thing that proves the work is yours.

What Claude got wrong

Claude initially just hosted both PDFs and wired up download buttons — it built the straightforward version first and only raised the IEEE copyright problem afterward. That flag should have come before publishing the file, not after. When something touches copyright or “is this even mine to post,” that’s a question to ask up front, not a footnote once it’s already live.

What we solved

• The published paper links to IEEE Xplore — visible, properly credited, no copyright issue.
• The unpublished thesis shows its abstract and links its open code and data, with the full writeup held back until formal publication — a deliberate choice, not a fearful one.
• I now have a cleaner mental model for putting research online.

Key lessons learned

1. “Can I share this?” and “should I share this?” are different questions. One is about copyright, the other about strategy. Answer both, separately.
2. Don’t self-host a publisher’s PDF. Link to the official page; share an accepted manuscript if you need the text. The typeset version isn’t yours to redistribute.
3. Visibility protects authorship — it doesn’t endanger it. A public, dated, attributed document is evidence the work is yours. Hiding it protects nothing, especially when the code is already open.
4. Flag the legal/ethical question before you ship, not after. If something touches copyright or ownership, raise it up front.