RevPref is a local-first app hygiene assistant that infers which apps you actually use from observed behavior, then flags rarely used startup and background apps worth disabling. It treats revealed usage as the signal and keeps data on-device.
RevPref
RevPref (GSV-D3) is a local-first app hygiene assistant, situated in GSV's life-systems pace-layer alongside its meaning-across-sapiences and infrastructure concerns: it builds an inventory of a user's installed applications, tracks which ones actually receive foreground time, and surfaces the gap between the two, on the premise that people recall what they use poorly, so the system reads revealed behavior instead of asking and turns that record into explainable suggestions to disable or uninstall. Its scoring logic rests on a small set of asymmetries: frequently opened apps are treated as valuable; apps that launch at OS startup but never reach the foreground are treated as suspicious; apps that consume disk, CPU, or network while rarely used are flagged for review; and system or user-protected apps are handled conservatively rather than flagged aggressively. Every recommendation is required to carry its evidence — the design principle is that the system explains why a suggestion exists rather than issuing an unexplained verdict, and it never uninstalls anything autonomously. The architecture is layered in three parts: OS-specific collectors that read app inventory, startup entries, usage events, and optional resource signals; a shell-independent core holding the local data model and recommendation engine that converts those facts into ranked, explained suggestions; and a desktop shell for review, keep/ignore decisions, and guided uninstall handoff, with the core deliberately kept independent of the shell so the eventual UI layer (Tauri, Electron, or native) can vary. The name plays on the economic concept of revealed preference, applied here to software rather than purchases, and the project's tagline frames it as the same technique applied to what a person actually uses versus what they think they use. Status: in-flight, with the recommendation model, evidence-based explanation principle, and three-layer architecture specified, though the provided material does not confirm which shell has been built or whether a working collector exists for any specific OS yet.