Double-blind done right
What to anonymize, how to cite yourself, and the metadata that betrays you.
Double-blind done right
Double-blind review means reviewers do not know who wrote the paper. Getting it wrong is one of the few formatting mistakes that can get a paper rejected without being read, so treat anonymization as a checklist, not a vibe.
What must go
| Item | What to do |
|---|---|
| Author names and affiliations | Remove or replace with “Anonymous Author(s)” |
| Acknowledgments | Delete for submission (funders and colleagues identify you) |
| Repo and project links | Replace with an anonymized link, see below |
| Institution-specific details | “our university’s cluster” beats “the Stanford GPU cluster” |
| “Our prior work” phrasing | Rewrite in third person, see below |
| Logos, watermarks, headers | Strip anything with a group or lab name |
Many classes automate the first row. With ACM’s class it is one option:
\documentclass[sigconf,review,anonymous]{acmart}Other venues ship a submission variant of their template, IEEEtran venues often just tell you to omit the author block.
Citing your own work
The rule is simple: cite yourself exactly as you would cite a stranger.
Wrong: “In our previous work [7], we showed…” Right: “Lovelace et al. [7] showed…”
Do not omit your own relevant papers, reviewers expect the obvious related work to be cited, and a suspicious gap can de-anonymize you as effectively as a name. The only exception is genuinely unpublished work, which you can upload as anonymous supplementary material instead.
Anonymous artifacts
Reviewers increasingly expect code and data. GitHub links are radioactive: the username is right there. Options that work:
- Services built for this (anonymous.4open.science is the field standard)
- A fresh repository under a throwaway account with scrubbed commit history
- Supplementary zip uploaded with the submission, with author names removed from file headers and notebooks
Check the zip before uploading: license files, AUTHORS, notebook metadata, and hardcoded home-directory paths (/home/ada/...) all leak names.
The accidents that betray you
The paper text is anonymous, but the PDF may not be:
- PDF metadata. The Author field often auto-fills from your system. Check with
pdfinfoor your viewer’s document properties, and clear it via\hypersetup{pdfauthor={}}. - File names.
lovelace-sigconf-final.pdfundoes everything. Name itpaper.pdfor the submission number. - Figure sources. Plots exported from tools can embed usernames in metadata or in a visible path in a screenshot.
- Preprints. Posting to arXiv before submitting is allowed at many venues and forbidden at others. The CFP says which, read it, this is part of choosing the venue.
Camera-ready flips it back
After acceptance, everything returns: names, affiliations, acknowledgments, the real repo link, and the first-person phrasing if you like it. Budget an hour for the de-anonymization pass, it touches more places than you remember, and the camera-ready checklist is the moment to catch stragglers.