Omniculus — Decision Log

A record of how the project took shape and why, so future contributors inherit the reasoning, not just the result. Newest first.

D-005 — Domain acquired: omniculus.com

Date: 2026-06-21 We own omniculus.com. It anchors the eventual public site, docs, and a synthetic-data-only demo (see ROADMAP Phase 6). The site should state the responsible-use scope prominently — the boundary is part of the brand.

D-004 — Name: "Omniculus"

Date: 2026-06-21 Chosen over candidates (Aegis, Argus, Lattice, Bulwark). Reads as omni + oculus ("all-eye"). We acknowledged it leans evocative/"watcher"; we deliberately offset that connotation with an explicit defensive, human-in-the- loop scope (see D-001) so the name is bold but the system is constrained.

D-003 — NetworkX as the initial graph backend

Date: 2026-06-21 Picked NetworkX over Neo4j for research simplicity and zero-setup. The KnowledgeGraph query helpers are the abstraction seam, so a Neo4j backend can be swapped in later (ROADMAP Phase 5) without touching detections.

D-002 — Synthetic-first, human-in-the-loop, explainable

Date: 2026-06-21 Three design commitments baked into the architecture rather than documented as aspirations: - Default data source is the synthetic generator; real connectors are opt-in and authorization-gated. - investigate/ has no response-execution method; consequential alerts require human review. - Every Alert carries its rule and evidence; the feedback loop logs every threshold change.

D-001 — Scope: defensive research platform, not surveillance/autonomy

Date: 2026-06-21 Context. The project began from a broad exploration (game theory → monitoring deployment → an early framing of "the most powerful surveillance system" with autonomous decision-making, considered against real municipal infrastructure).

Decision. We deliberately scoped the project down and defensive: - No mass-surveillance capability; operates on authorized/synthetic/lab data. - No autonomous action; the system is decision support with a human in the loop. - No targeting of production infrastructure the operator doesn't own; develop and validate against synthetic data and lab replicas / digital twins.

Why. A maximal-surveillance, autonomous system is the same dangerous artifact regardless of stated intent, and it does not serve the stated research goal ("teach our system to be more responsible/vigilant"). The constrained scope is both safer and a better research instrument — and it became the product's defining principle (see VISION.md "What Omniculus is not"). This decision governs all others; items conflicting with it are non-goals.