Skip to content

Optional Killed-By Correlation — Inferred K/D (Stats Only) #116

Description

@ntatschner

Problem

Individual users see their own deaths and kills, but there's no way to infer killer ↔ victim pairings across the user base — and a naive approach would leak identifying info about who killed whom, or invite leaderboard abuse.

Approach

Opt-in per user, off by default. When two users have both opted in, server-side correlation across concurrent sessions infers killer ↔ victim pairs from kill/death events that happen within tight time + location windows. Each inferred match is confidence-scored.

Stats-surface only — no notifications, no leaderboards, no exposure of the inferred edges to either party beyond their own aggregated K/D number.

Acceptance criteria

  • User-level opt-in toggle (default OFF) with explicit privacy disclosure
  • Correlation runs only across pairs of users who BOTH opted in
  • Confidence scoring based on time + location window proximity
  • Inferred K/D appears in personal stats only; never on leaderboards or org views
  • Inferred edges are NOT exposed via any API to either party (only the count survives)
  • Audit-log entry on each correlation pass (count of matches, no identifying detail)

Risks / Open questions

  • Time/location window thresholds: too loose → false positives; too tight → misses. Needs tuning data.
  • Server-side correlation cost grows with concurrent-user count — what's the upper bound and ratelimit?
  • Privacy regression risk: even aggregate K/D might be back-inferrable in small populations. Threshold for showing the number?
  • Opt-in semantics: does opt-in apply only to future events, or retroactively? Probably future-only to avoid surprise.
Original draft notes

Opt-in per user, off by default. Server-side correlation across concurrent sessions to infer killer ↔ victim pairs from kill / death events that happen within tight time + location windows. Confidence-scored matches. Stats surface only — no notifications, no leaderboards, no leakage of the inferred pairing back to either party.


Tracking

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions