Skip to content
← JoywriteRead the posts

arXivBangers

Find the bangers.
Improve the loop.

Community-driven experiments in self-improving research distribution.

Explore an experiment Browse the papers

rank, write, post, repeat.

The working system

The post is the output.
The method is the experiment.

Ranking and writing are separate methods. Audits and comparisons help decide what to change next.

01

Rank

Judge titles and abstracts for reader value. V3 breaks exact editorial ties.

editorial-v1
02

Write

Explain why a selected paper is worth the reader's attention.

reader-value-v1
03

Post

Publish to X and record reach and engagement for later comparison.

@arXivBangers ↗
04

Learn

Audit the work. Compare another method. Review the change.

What makes a method inspectable?

Methods record their authors, model settings and source fingerprints. Fixtures check the input and output contract.

Offline audience-response experiments test other ways to rank papers. The current posting loop uses the methods shown above.

How the current selection works

V3 is a model trained on observed alphaXiv selections. The live experiment compares the editorial nominee with V3's top eligible paper scoring at least 80 editorially. When the nominees differ, each has a 50% chance. Agreements and fallbacks are recorded separately.

Imports run hourly. Upcoming choices can change as papers arrive, then lock 30 minutes before posting. Missing V3 scores or equal V3 scores resolve editorial ties newest-first. Stale grades and drafts put papers on hold.

Community experiments

Test your
hypotheses.

The goal is measurable growth in research distribution. Propose a hypothesis or substitute a ranking or writing method. Every comparison uses the same three measures.Community submissions are coming soon.

Fixed growth measures

  • Follower growth rate
  • Average post engagement
  • Reach
What would you change?
Example hypothesis

Could a reproduction-first ranker make research travel further and attract new followers faster?

Keep fixed

The paper pool, evidence, and writer.

Change

How the ranker values reproducible results.

Shared evidence

Keep metric definitions and evaluation windows fixed. Compare every candidate with the current baseline over repeated trials. Record the ranking and writing methods behind each post, and account for paper choice, timing and audience differences.

Results and methods are reviewed before promotion.

Each comparison feeds the next hypothesis. Reviewed improvements become the next baseline.

Opening the loop

The next good idea
can come from you.

Anyone will be able to propose hypotheses and alternative ranking or writing methods. arXivBangers ranks papers, writes posts, and records how they perform so the community can test what improves research distribution.

Results guide the next hypothesis. Every method is judged by the same follower growth rate, average post engagement and reach criteria.

We're preparing a public, open source release.

Follow arXivBangers