Skip to content

ADR 0002: Hardware Tiering Strategy

Date: 2026-01-16 Status: ACCEPTED (97/3 architecture) / DEFERRED (specific BOM cost targets) — see ADR 0008

Reset banner (2026-05-08): The 97% Member / 3% Leader hierarchy below is the live architectural decision and remains in force. The BOM cost targets (Member < $20, Leader < $80, etc.) are deferred — the May 8 reset to a COTS-first MVP path means the MVP Member and Leader are based on commercial dev kits whose BOMs are likely 5–10× the targets here. Custom BOM optimization re-enters the plan post-validation. See ADR 0008.

Context

Extensive cattle ranching requires monitoring thousands of animals over vast areas. * Constraint 1 (Cost): To be viable, the per-animal cost must be < $20 USD. * Constraint 2 (Physics): Long-range uplink (Satellite/LTE) requires high power (solar) and expensive modems (> $50 USD). * Conflict: We cannot put a satellite modem on every cow (too expensive) nor use only cheap tags (no connectivity).

Decision

We will implement a Two-Tier Heterogeneous Architecture:

  1. Tier 1: Member Node (The "Leaf")

    • Role: Sensing (Movement, Temperature) & Short-Range Relay.
    • Hardware: Smart Ear Tag (Nordic nRF52), Primary Battery (5+ years), Sub-GHz Radio.
    • Cost Target: < $15 USD.
    • Quantity: 97% of herd.
  2. Tier 2: Leader Node (The "Gateway / Cowbell")

    • Role: Aggregation, Edge AI Processing, & Incremental Backhaul.
    • Hardware: Gravity-Balanced Split Collar.
      • Base (Bottom): Heavy High-Capacity Battery (counterweight) + Terrestrial modems + Edge MCU.
      • Head (Top): Sky-facing Dorsal unit for Antennas + Modular NTN Plugin.
    • Cost Target: < $100 USD (Base) / +$45 (NTN Module).
    • Maintenance: Yearly battery change.
    • Quantity: 3% of herd.

Consequences

  • Positive: Blended cost drops to ~$25/head, making the unit economics viable vs. Satellite-only competitors ($200/head).
  • Negative: Increases software complexity. We must maintain two firmware codebases (Zephyr for Tags, Linux for Collars) and handle complex mesh synchronization between them.