Measuring Proof Burden in Public Bounty Listings: A RentAHuman Case Study

Accepted at ACM HCOMP 2026 and available on arXiv. Introduces 'proof burden' as a measurable task-design property and audits 779 public bounty listings, showing that 56.2% meet the paper's author-defined severe threshold and span 154 distinct requirement combinations. It raises — explicitly as a post-hoc hypothesis for future testing — the possibility that agent- and bot-labeled requesters more often demand physical-world action and recurring monitoring.

Authors: Iman YeckehZaare · Venue/status: 2026 ACM Conference on Human-AI Complementarity and Alignment (HCOMP 2026) · Year: 2026 · DOI: 10.1145/3834580.3838744 · arXiv:2608.18547 · Camera-ready PDF · Supplement · EasyChair package · Interactive reader snapshot

Accepted at ACM HCOMP 2026 and available on arXiv. The paper codes 13 proof-of-completion requirements across 779 eligible public bounty listings using two independent coders and blinded adjudication. It reports that 56.2% meet its author-defined severe threshold and that those listings span 154 distinct requirement combinations. The requester-label comparison is explicitly post hoc and cluster-sensitive: a hypothesis for a preplanned follow-up, not a confirmed group difference. Location proof itself was less common among agent-or-bot-labeled listings (2.8% versus 11.7%).

Version boundary: this paper was accepted after peer review and will be presented at its venue. The linked manuscript is the author's camera-ready version, also posted on arXiv; it is not the final ACM version of record.

The rendered manuscript page adds status, visual summary, review boundary, citation metadata, and contribution notes. Key links: home, systems, papers, manuscripts, Google Scholar, GitHub, LinkedIn, ORCID, MIT profile, and CV PDF.