Why I Withdrew an Accepted COMPSAC Paper¶
"Who Audits the Auditor?" — a blockchain-anchored fraud detection pipeline, F1 = 0.895 — was accepted at IEEE COMPSAC 2026. I withdrew it.
Not because of a flaw. Because of the registration fee.
The arithmetic¶
IEEE conferences require at least one author to register at full rate for the paper to appear in the proceedings. For a self-funded researcher without travel support, that is a four-figure cost for a single workshop-adjacent publication.
I had four other 2026 papers to bring to camera-ready. The same money covers compute for months of experiments. The paper does not become more true by being in the proceedings.
So it stays where it already is: arXiv:2604.22096, free to read, permanently.
What is actually lost¶
Honest accounting — withdrawing costs something real:
- Indexing. IEEE Xplore is where reviewers in this subfield look. arXiv is not.
- The acceptance signal. Peer review passed. Nothing durable records that.
- Citations. Proceedings papers get cited more than preprints, regardless of quality.
I do not think "just put it on arXiv" is a complete substitute. It is a compromise, and pretending otherwise helps no one.
Why say it publicly¶
The convention is to quietly drop the paper and never mention it. That convention makes the cost invisible — it looks like these papers were never written, rather than written, accepted, and priced out.
The publication page for this work says exactly this: accepted, withdrawn, preprint, seeking resubmission. It reads worse than a clean acceptance line. It is also what happened.
If a venue's business model means accepted work does not appear because the author cannot pay, that is a property of the venue worth stating out loud — especially by people for whom it is merely inconvenient rather than disqualifying.
Where it goes¶
The work is not dead. It is being prepared for a venue whose cost structure I can actually meet. If it lands, the CMS entry flips from preprint back to conference and this post gets a postscript.
Until then it is a preprint, and the number is still 0.895.