AI
When Privacy Moves ML-Mediated Decisions On Device: Information and Incentive Misalignment in Auctions
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 26/09/2026, 20:00:00 UTC
Briefing created: 29/09/2026, 12:16:09 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 29/09/2026, 06:18:24 UTC.
Briefings
This briefing examines an on-device auction simulation involving 36 campaigns and 50 devices where ML-mediated decisions create information-structure failures. The study reveals that proportional pacing overspends budget significantly under one tick of staleness (17.77%) and 50 ticks (1,669.31%), even with a visible-budget guard that achieves zero-lag compliance at the score-unit level. A second finding highlights a severe incentive misalignment where 98.23% of rival auctions admit profitable deviations at one tick, isolating the runner-up's multiplier in the winner's price. The research uses dimensionless integer score units without currency semantics and derives a finite-window expected excess-debit bound under conditional charge caps.
Key points
- Proportional pacing overspends 17.77% after one tick of staleness and 1,669.31% after 50 ticks under original 20-times budget pressure.
- 98.23% of rival auctions at one tick admit a profitable deviation when ML/pacing score transformation changes payment units.
Context and limitations
Data covers 30 paired demand paths across 36 campaigns and 50 devices with dimensionless integer score units. No currency semantics are claimed; budget pressure varies from 20-times to 2-times. Results are from an on-device simulation with no real-world deployment context.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 29/09/2026, 12:16:09 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeic7puj3g3qs2gsuqywme56q57rihxl2ltofkgxoffqsgdnlni6zxu