HIVE / VERIFIED DISTRIBUTION / 2026-11-16
Verified model delivery. Visible costs.
Hive-governed public and workspace-private catalogs, authenticated HTTPS peers and a verified content cache. Actual model-file transfers, measured separately from inference.
15 successful operations. Three repeats per route. One Mac and one Linux VPS; co-located Mac processes are not two physical devices.
Verified-file preparation time

Where the model payload traveled

What these results establish
The WAN peer took 25.09 s versus 19.99 s for direct HTTPS. The two-source path took 11.29 s, with 52.2% less model payload over WAN because part came from a co-located source. Verified cache reuse took 0.70 s. These are observed medians, not universal acceleration guarantees.
Public metadata can be browsed without exposing private peer inventories. Workspace membership and current authorization are checked independently for every chunk. File and chunk hashes verify exact bytes. Credentials and private endpoints are excluded from this report.
Energy coverage, without invented totals
All 15 final Mac sensor intervals measured while discharging, with complete recorded coverage. Median gross Mac energy: 649.1 J direct, 793.7 J WAN peer, 237.7 J co-located peer, 307.0 J two sources, 26.4 J cache. Sensor windows include observer overhead; gross energy includes other apps. VPS/router energy and calibration uncertainty unavailable. No incremental application or ecosystem energy saving established.

Mac ARM64 standalone companion
Model Peer 0.1.0-preview.1 (build 2): bundled runtime, ad-hoc signature; no Developer ID or notarization. Storage/transfer CLI, not a new mobile or inference engine release.
Download build 2SHA-256: a3dac1bc18cb10dc91c9657edd50b233a3136426846b7e90375764d4abc8029e
Inspired by Pirate Face; independently authored HTTPS protocol. No BitTorrent/DHT interoperability, automatic public seeding, pooled inference RAM or mobile P2P qualification.
Pirate Face · Raw sanitized measurements · Fabrizio Degni