Technology & Stack Brief — Extensive Cattle Tracking (May 2026)

Confidential disclosure. This document describes the technology categories, standards, and stack layers under evaluation for the platform.

Companion documents. Read this brief as part of a three-document set:

Strategic Partner Brief — the vision, the architecture-at-a-glance (97/3 split, modular Cowbell, 1 km cow network, system diagram), the phased roadmap narrative with head-count targets, the partnership ask.

Market Landscape Brief — competitive landscape with URLs, categorical pricing & topology map, regulatory anchors.

Technology & Stack Brief (this document) — local-network and backhaul technology categories, edge intelligence tiering, software stack, standards compliance matrix, hardware-development phase table at the Member/Leader column level.

The architectural premise (the asymmetric 97 % Member / 3 % Leader hierarchy, the modular Cowbell carrier-and-comms-module pattern, the 1 km between-cow design target with an open Star-vs-Mesh topology choice) is established in the Strategic Partner Brief and is not restated here.

Local Cow Network — Member ↔ Leader Cluster

The cow-to-cow / cow-to-Leader link is the engineering centerpiece of the platform. Members operate against the Leader cluster described in the Backhaul section (multiple Cowbells per herd for high availability), not against a single Leader — the network must let any Member reach any Leader in the cluster rather than binding a Member to a specific Leader.

The design constraints (recap from the Strategic Partner Brief): minimum 100 meters, ideal 1 km between adjacent animals as the link-budget design target; maximum density of 500 Member nodes per 1 Leader Gateway; body-shadowing attenuation as the primary RF challenge; topology (Star vs. Mesh) explicitly open and selected by field measurement.

Technologies under evaluation — categories, not vendors

Category	Range (target)	Power profile	Notes
Sub-GHz IEEE 802.15.4g mesh	1–5 km / hop	Low	Zephyr-supported open stack; royalty-free; high throughput for behavioral data.
LoRa P2P / Mesh	5–10 km / hop	Ultra-low	Maximum range; very limited bandwidth; well-understood power budget.
Trickle-based open mesh (RFC 6206)	n/a	Low	Reference logic — same architectural pattern used in Thread / Matter and Smart-Grid mesh. Royalty-free re-convergence.
Industrial proprietary mesh (NeoCortec, Wirepas)	1–2 km / hop	Ultra-low / low	Decentralized "flat" mesh with high density. Proprietary licensing.
Mioty / TS-UNB	5–15 km	Low	Telegram-splitting; high robustness in dense, interfered environments.
DECT NR+	1–2 km	Low	Non-cellular 5G; massive density; future-proofing.
MeshCore / ZephCore	n/a	Low	Open-source MIT-licensed protocol; runs natively on Zephyr (ZephCore) for RAK3172; supports up to 64 hops (solving body-shadowing range limitations) without GPL copyleft restrictions.

All categories above are under active evaluation. Selection will be driven by the validation phase (measured battery budget, body-shadowing attenuation, mesh network convergence, vendor terms) rather than by an a-priori ranking — this brief intentionally does not signal an internal priority order. MeshCore (via ZephCore) is currently the selected baseline for the Phase 1 Foundation COTS pilot.

Design principle — royalty-free radio stack

Long-term unit economics at mass scale benefit from avoiding per-node licensing on the radio stack. Royalty-free, standards-based options remain in scope alongside proprietary alternatives where the latter offer a material technical advantage that closes the licensing gap. The 3+ year minimum (5-year desired) autonomy mandate and the asymmetric Leader/Member hierarchy are the dominant constraints either kind of stack must satisfy.

The Mesh power-budget challenge — explicitly acknowledged

Sustaining a high-fidelity mesh at the +3-year zero-maintenance autonomy mandate requires extreme optimization of:

Radio duty cycles (synchronous wake / sleep windows).
Asynchronous neighbor discovery (no constant beaconing).
Exception-based reporting (quiet most of the time, transmit only on biological/operational anomaly).

Whether the chosen topology is Star or Mesh, closing the power budget is the critical engineering exercise during the validation phase.

Ground-Level Propagation & Antenna Requirements

The physical reality of cattle grazing dictates that both the Member (ear tag) and Leader (Cowbell) operate at roughly 1.2 meters above ground. This severely impacts theoretical radio ranges: - Terrain Obstruction: Because the devices are so low to the ground, the earth itself, along with pasture brush and uneven terrain, physically blocks the signal path. The theoretical 10+ km ranges marketed by vendors drop dramatically to 150–400 meters when operated "cow-to-cow". - Body Shadowing: A 500 kg animal is primarily water, which heavily absorbs radio frequencies. If several cows are standing between a transmitting Member and a receiving Leader, the signal is significantly weakened.

Architectural Mitigation: This physical reality is why we specify a link-budget design target of 100 meters to 1 km between adjacent animals, rather than relying on multi-kilometer ground-level links. While we cannot guarantee every Member is always within direct range of a Leader, we can statistically assume an animal is close to another animal. This constraint directly drives our ongoing field evaluation between a robust long-range Star topology (direct to Leader) and a multi-hop Mesh topology (relaying cow-to-cow).

Hardware Mandate: The hardware partner must prioritize rigorous Antenna Design. The Cowbell carrier requires dual optimization: an upward-facing antenna for satellite/cellular backhaul, and a secondary sub-GHz antenna heavily optimized for horizontal, ground-level propagation through a herd.

Backhaul — Leader Cluster → Cloud

A herd is served by a cluster of Leader Cowbells, never a single one — the ~3 % share of the herd carrying a Cowbell is sized for high-availability redundancy. Backhaul is therefore a cluster property, not a per-device property:

Multiple uplinks per herd. Each Cowbell in the cluster carries its own backhaul comms module, so the herd has N independent paths to the cloud (with N typically 3+ for any operational deployment, scaled with herd size).
Failover is local-network-driven. If any single Leader's link drops (battery depletion, satellite outage, physical loss of the animal), the cluster's remaining Leaders continue to ferry the herd's data; affected Members re-associate to a surviving Leader through the local cow network.
Heterogeneous backhaul mix per cluster. Different Leaders in the same cluster can carry different comms modules — e.g. one on terrestrial NB-IoT, two on NB-IoT NTN, one on LEO L-band — to harden the herd against any single network's regional outage. The hot-swap carrier model (described in the Strategic Partner Brief) makes this practical without N different SKUs.
Cloud-side de-duplication. Aggregated state arriving from multiple Leaders covering the same herd is reconciled at the cloud, so redundancy on the wire does not become duplication in the data product.

This section enumerates the backhaul technology categories that an individual comms module can implement; the cluster property above is what turns that category list into a deployable HA architecture.

Dual-Path Payload Constraints

The architecture strictly bifurcates payloads based on the transport path: - Primary Path (Cellular/NB-IoT): High bandwidth, low cost. Handles rich aggregated herd-state snapshots (~100 KB to 2.5 MB per day), standard payloads (JSON/CBOR) over TLS, and supports full OTA updates. - Fallback Path (Satellite/NTN): Low bandwidth, high cost. Handles only Edge AI-triggered exceptions (theft, mortality) and minimal daily heartbeats. Utilizes highly compressed custom binary payloads strictly capped at < 50 bytes per transmission to preserve unit economics. No OTA support.

Category	Networks under evaluation	Coverage	Standard	Notes
NB-IoT NTN (3GPP Rel-17)	Skylo, Sateliot	Global (LEO)	Standardized	Standards-based NB-IoT silicon with NTN firmware. The 3GPP-aligned future-proof path.
LEO direct-to-satellite (proprietary L-band)	Globalstar (RM200M / ST100 family)	Global / regional	Proprietary	Field-proven in Mercosur via regional VARs. Cost-effective for moderate-payload telemetry.
Terrestrial NB-IoT / LTE-M	National telcos (Antel UY, Tigo, Vivo, etc.)	Coverage-dependent	3GPP	Used opportunistically when in coverage; lowest per-message cost.
Direct-to-satellite BLE	Hubble Network	Global (LEO)	Proprietary, early-stage	Pilot-stage technology; under monitoring as a future option.
L-band proprietary (low-payload)	Astrocast	Global	Proprietary	Cost-effective alternative for low-payload, low-frequency telemetry.
Roaming / multi-network NTN aggregation	Monogoto, Syniverse	Global	Aggregator	Service-layer aggregation across multiple satellite + cellular networks; not a transport in itself.

The hot-swap carrier model means the choice across these backhaul categories does not need to be made once-and-for-all at the carrier-design stage; multiple categories can be supported by different comms modules on the same carrier across different SKUs or deployment regions.

Antenna design is the deepest open challenge in realizing the modular vision — each comms category has different frequency, gain, polarization, and ground-plane requirements that a fixed carrier antenna cannot serve uniformly. Two patterns under evaluation: antennas on the comms module itself (Blues Notecard pattern) vs. multi-feed antenna real estate on the carrier. See the Strategic Partner Brief for the framing; the choice is part of the carrier-design work with the hardware design partner.

Failure modes — what the cluster guarantees, and what it does not

Failure	Cluster behavior	Detection horizon	Notes
Single Leader battery depletion	Members re-associate to surviving Leaders; herd data path continues. Cloud emits a "Leader degraded" alert on next reporting cycle.	Within one Leader heartbeat interval (minutes).	Scheduled service window normally addresses this before depletion.
Single Leader satellite link outage	If the Cowbell carries other backhaul (terrestrial / different satellite per the heterogeneous-mix property), it continues. Otherwise its share of herd state queues in the local store-and-forward buffer until the link returns.	Member-level coverage continues; cloud notices missing reports within one fixed-cadence interval.	The heterogeneous-backhaul-mix-per-cluster pattern is the primary mitigation.
Single Member silent for >24 h	Cloud emits a "lost or mortality candidate" alert against that animal ID. Leader cluster has no further mitigation — physical recovery / inspection is the human-side action.	24 h (configurable).	Distinguishing battery failure from animal mortality from theft requires the post-event acceleration profile prior to silence.
Entire Leader cluster down (worst case)	Members buffer locally to the limit of their own MCU storage; no telemetry reaches cloud. Local cow network may continue to operate Member-to-Member but cannot exfiltrate.	Cloud alerts on N consecutive missing herd-level reports.	Mitigated by the cluster-floor of 3 Leaders; addressed operationally by herd inspection if it persists.
Satellite-tier outage region-wide	Same as single-Leader satellite outage but applied to all Cowbells of a given backhaul class in the region.	One fixed-cadence interval.	The heterogeneous-mix-per-cluster property is the primary mitigation; multi-satellite-network selection is the secondary.

Positioning Strategy — Open Evaluation

Two design paths are under evaluation for Member-node positioning:

Path A — On-Device GNSS. Independent geolocation per Member. Highest spatial resolution. Subject to power-budget validation; GNSS cold-fix energy is the dominant draw.
Path B — Mesh-Relative Inference. ToF / RSSI triangulation against the high-precision GNSS anchor on the Leader Gateway. Significant power saving on the Member; loses absolute spatial resolution between Leader anchors.

Both paths are compatible with the same MCU + radio combination on the Member, so the choice is firmware-level, not silicon-level.

The Leader carries high-precision multi-band GNSS in either path — the spatial accuracy of the herd map ultimately depends on the Leader anchor.

Edge Intelligence

Computation is tiered to match the power budget available at each level. AI / model inference does not live on the Member node — the Member's job is classification and reporting, nothing more. Analytics live at the Leader.

Member nodes (ear tags) — classify and report, no AI on-device

Continuous accelerometer sampling with deterministic activity classification — Grazing / Ruminating / Walking / Lying / Panic — implemented as low-cost rule-based logic on the MCU. No model inference, no on-device learning, no historical analytics.
Two reporting cadences to the Leader:
- Frequent regular heartbeat with the latest activity vector at a configurable interval.
- On-exception transmission, immediate, when a local rule trips (sustained anomalous pattern, panic, theft-class acceleration).
The on-device complexity stops at classification. This is what keeps the Member power budget compatible with the +3-year zero-maintenance autonomy mandate.

Leader gateways (Cowbell) — local analytics layer

Aggregates the activity streams from the Members in its local mesh.
Runs the platform's Edge AI analytics workload locally on the Leader (e.g., via Zephyr/TinyML) — gait / lameness pattern detection, herd-level behavioral anomalies, estrus signatures, theft escalation. Edge AI is a strict financial mandate to compress data before it hits expensive satellite links.
Non-Volatile Storage Mandate: To survive satellite blind spots, the Leader must buffer offline telemetry. The calculation requires 8-16MB of non-volatile memory (~7.2MB for 3 days of offline herd caching for 500 nodes + ~4MB for dual-bank OTA staging = ~11.2MB minimum requirement).
Two reporting cadences to the cloud platform:
- Fixed regular interval with aggregated herd state.
- Alerts, immediate, when an analytic threshold trips.
This tier is also where exception traffic from Members is de-duplicated and contextualized before any satellite or cellular bytes are spent on it.

Cloud / platform

Cohort analytics across herds, behavioral drift over biological cycles, regulatory reporting (SNIG, EUDR), and integration APIs.

Algorithmic basis (peer-reviewed) — Leader-tier analytics

Intellectual Property Note: JAAB retains full IP ownership of all application-level algorithms, Edge AI models, and behavioral classification logic. The hardware partner provides the compute substrate; JAAB provides the intelligence.

Lameness detection: lying-bout frequency / duration changes, validated days before visual limping appears. References: Wageningen University, Lincoln University NZ.
Estrus detection: mounting-behavior signatures aggregated across Members.
Theft detection: non-biological acceleration patterns (e.g., sustained transport > 20 km/h).

A direction we are exploring — not yet a locked decision — is to package these analytics as plug-ins: sandboxed behavior modules that can be field-replaced and remotely updated over-the-air on a deployed Leader without re-flashing the device firmware or pulling devices from the field. The intent is to let the Leader-tier analytics evolve over the +5-year mechanical lifecycle of the Cowbell carrier, and to keep the underlying hardware intentionally hardware-agnostic so the same plug-in can run on equivalent industrial processors. Candidate runtimes (e.g., WebAssembly / WASI) are under evaluation; the actual runtime selection is part of the architecture work to be done with the hardware design partner.

Edge Operating System & Software Stack

Layer	Choice	Rationale
Edge OS	Zephyr RTOS	Real-time efficiency, royalty-free, broad driver support, large vendor-agnostic community. Firmware sovereignty for long-term scaling.
Edge logic	TinyGo / Wasm	Safe, portable, hot-swappable behavior modules. Hardware-agnostic.
Cloud + high-bandwidth messaging	NATS / JetStream	"Dial-tone" reliable on the cloud side and over terrestrial cellular / LTE-M / Wi-Fi backhaul where bandwidth is plentiful. Store-and-forward across intermittent links; proven at financial-transaction scale (Mastercard, Walmart edge). NATS is not used over the satellite backhaul — see next row.
Satellite / low-bandwidth backhaul encoding	Per-network compact framing (CBOR / protobuf-style / custom binary, byte-budget tuned per satellite network)	Satellite networks (NB-IoT NTN, LEO L-band, direct-to-sat BLE) have hard MTU caps, per-message billing, and tight power budgets — a general-purpose pub/sub like NATS is the wrong shape. The Leader serializes aggregated state into the smallest possible network-specific frames (often bit-packed) for the active comms module, and the cloud-side gateway decodes them back into NATS events on arrival. Encoding choice is per satellite network and is part of the comms-module integration.
Edge runtime + cloud control plane	FluxRig (JAAB-developed, in active development)	JAAB Tech's in-house orchestration layer. Integrates the Zephyr / plug-in / messaging layers above into a single runtime. Deploys plug-ins from the cloud control plane to the Leader edge runtime, bridges the satellite-encoded backhaul into NATS on the cloud side, manages the store-and-forward fabric across intermittent backhaul, coordinates herd- and cohort-level state.
Cloud analytics	DuckDB + standard cloud OLAP	Fast SQL on aggregated platform data; not on-tag.
API layer	GenAI-ready structured JSON	Natural-language queries over herd data ("which paddock has the most lame cows?").

Layered messaging — the wire protocol differs by tier. Member ↔ Leader uses a lightweight sub-GHz binary protocol (not NATS). Leader → cloud uses NATS where the backhaul is terrestrial/cellular, and per-network compact framing where the backhaul is satellite. NATS is the cloud-side and high-bandwidth fabric, not the universal wire protocol — choosing NATS for satellite would burn power and money for no benefit.

IoT Device Management & Fleet Operations

Managing a fleet of thousands of autonomous devices requires a robust, remote-first device management strategy. The hardware and firmware must support the following lifecycle capabilities, orchestrated by a central cloud platform:

Platform Flexibility Note: JAAB Tech actively develops FluxRig, an in-house orchestration layer built heavily around NATS / JetStream for high-bandwidth telemetry and state synchronization. While we expect FluxRig to serve as the default platform for this solution, we remain fully open to evaluating alternative IoT platform and data solutions proposed by our hardware partners.

Standardized Management Protocol (e.g., LwM2M): To ensure interoperability and efficient payload handling over cellular links, the use of Lightweight M2M (LwM2M) or an equivalent standardized protocol for device management, configuration, and FOTA is explicitly encouraged.
Over-The-Air (FOTA) Updates: The Leader (Cowbell) MCU must support dual-bank flash memory to enable seamless firmware updates with safe rollback capabilities in case of a corrupted download. Crucially, OTA updates are only permitted over high-bandwidth primary links (Cellular/NB-IoT) and are strictly disabled over Satellite/NTN.
Digital Twin & Configuration Sync: The platform maintains a cloud-side "Digital Twin" of every Leader and Member. Configuration parameters (e.g., reporting cadences, Edge AI alert thresholds) are synced asynchronously when the device next checks in.
Remote Diagnostics & Telemetry: Beyond cattle behavioral data, the system must report hardware health telemetry. This includes battery discharge curves, internal temperature, and backhaul signal strength (RSSI/SNR). This data allows the platform to predict device failures or battery depletion before the node drops offline, scheduling maintenance during the next rancher visit.

Security & Trust

The platform operates in environments where physical access to devices is uncontrolled and the Internet link is intermittent.

Edge cryptography: target PSA Certified Level 2 for hardware-isolated root-of-trust. Behavioral reports digitally signed at the edge — converting raw sensor data into audit-ready evidence for institutional buyers and regulators.
Zero-touch provisioning: SGP.32 v1.2 iSIM — eliminates field-labor sync at deployment scale.
Immutable provenance logs: a layered store-and-forward fabric — sub-GHz framing on the local cow network, per-network compact framing on the satellite tier, NATS JetStream on the cloud and on terrestrial backhaul — together act as a tamper-evident flight recorder. Each tier uses the wire protocol appropriate to its bandwidth and energy budget.
Data sovereignty: rancher / institutional buyer owns the data. The platform is the utility, not the walled garden.

Standards Compliance Matrix — Hardware-Relevant

The market and regulatory rationale for each standard below is in the Market Landscape Brief. This matrix maps each standard to the device class that must satisfy it.

Standard	Scope	Member tag	Leader Cowbell
ISO 11784:2024 / 11785:2026	Animal RFID identification (HDX)	Required	n/a
ICAR certification	EID conformity	Required	n/a
FCC Part 15 / regional sub-GHz	902–928 MHz unlicensed	Required	Required
CE RED	EU placement	Optional Phase 1	Optional Phase 1
IP67 (Member) / IP69K (Leader)	Ruggedization	Required	Required
3GPP Rel-17 NTN	Satellite NB-IoT/LTE-M	n/a	Required (Phase 2+)
SGP.32 v1.2	iSIM zero-touch provisioning	Optional	Required
AGESIC IoT Security 2019	Uruguay government IoT data-handling baseline	n/a (cloud / platform-side)	n/a (cloud / platform-side)
PSA Certified Level 2	Edge cryptography root of trust	Strategic target	Strategic target

Per-country RF licensing (Argentina ENACOM, Brazil Anatel, Paraguay CONATEL, Colombia ANE) and personal-data-protection regimes (Uruguay Ley 18.331, Brazil LGPD, etc.) are tracked in an internal compliance roadmap; deployment in any country other than the Uruguay testbed will require a country-specific certification pass.

Device end-of-life

The Member tag ships and stays on the animal until slaughter or natural death. The deployment plan includes a recovery-and-recycle posture for spent tags at slaughter-yard processing points, with the device materials selected to be compatible with standard agricultural-electronics recycling streams. The Leader Cowbell, by contrast, is rancher-managed and has a scheduled-service lifecycle with battery refresh.

Phased Hardware-Development Path

The four phases, the strategic intent of each, and the head-count targets are documented in the Strategic Partner Brief. The same four phases are listed below at the per-device level — what hardware ships at each phase, on the Member and Leader columns. The two views describe the same roadmap, not different roadmaps.

The Member and Leader follow different hardware-evolution tracks because their constraints are different:

The Leader (Cowbell) has a forgiving form factor and energy budget — COTS satellite-gateway carriers with hot-swap comms modules are viable from day 1, and the custom Leader is a later optimization.
The Member (ear tag) has hard constraints (≤ 20 g, +3-year zero-maintenance autonomy, ICAR-coexistent ID layout, sub-GHz radio in a tag-class housing) that no commercial dev kit satisfies as a deployment target. Member custom-design is required from the Foundation phase. Bench prototyping may use eval boards for power-budget characterization, but the deployed Member is custom from day 1.

Phase	Member Node	Leader Cowbell	Focus
Foundation (current)	Custom-design begins. Sub-GHz radio + IMU + ICAR-coexistent ID coexistence in ≤ 20 g housing. Bench eval boards used only for power-budget characterization — the deployed Member is custom.	Off-the-shelf hot-swap satellite-gateway carrier (a socketed cellular / satellite modem, e.g. of the Blues Notecard class) on a partner-built mechanical housing. Validates the modular comms pattern before committing to a custom Cowbell carrier.	Member: close the link budget, body-shadowing, and +3-yr autonomy on real custom silicon. Leader: validate the modular-comms pattern with off-the-shelf modems before committing to a custom Cowbell carrier.
Custom Hardware Production	Refined custom Member (production-ready)	Custom Cowbell carrier with hot-swap comms (production housing)	Ruggedized field durability and energy-harvesting integration on both classes; production tooling.
Precision Biosensing & E-Paper (aspirational)	Optional Bolus pairing (short-range RF) to the Member. Evaluation of e-paper displays for digital visual identification.	Same as previous phase	Health / biosensing telemetry and dynamic visual IDs for ranch operations. Scope and feasibility under evaluation.
Mass-Market Scale	High-volume standard silicon (or conditional custom ASIC if required for cost)	Custom Cowbell carrier	Extreme BOM reduction at mass-market volume. ASIC path is optional.

The current phase is Foundation. Off-the-shelf modems on the Leader let us validate the modular-comms pattern before building a custom carrier. The Member is custom from day 1 — its constraints rule out off-the-shelf hardware as a deployment target.

What the Hardware Design Partner Engages With

The high-level engineering domains and the strategic-partnership rationale are in the Strategic Partner Brief. This section enumerates the specific technical scope for a hardware design partner.

We are seeking specialized engineering expertise on the mechanical and electronic carrier design for both classes:

Cowbell Leader carrier — ruggedized housing; hot-swap comms-module connector; MCU + GNSS + sensor integration; energy-harvesting power path; antenna feed design; certification path.
Member ear-tag carrier — ≤ 20 g housing; ICAR-coexistent ID layout; energy-harvesting integration with ultra-low-power MCU + sub-GHz radio; +3-year autonomy validation; certification path.

Logistics & Onboarding (Zero-Port Provisioning)

To support mass-market deployment by non-technical ranch hands, the physical device lifecycle must adhere to "Zero-Port Provisioning": - Deep Sleep Shipping: Devices must ship in a deep sleep state drawing < 5µA. - Zero Physical Ports: Activation occurs via a magnetic reed switch or NFC tap. There are no buttons, USB ports, or physical switches exposed to the elements. - Field Configuration: Initial provisioning and diagnostic queries are performed locally in the field via a BLE or NFC mobile app.

About

This initiative is being incubated under JAAB.tech, a consultancy active in mission-critical distributed systems, with the intent of forming a dedicated operating entity for the cattle-tracking platform. Inquiries: andres@jaab.tech.

Confidential — General Technical Vision subject to change. Proprietary and Confidential at JAAB Tech. Prepared by Juan Andrés Antoniuk