# Ingestion Synthesis — Fresh Founder Input vs Ratified 18-Claim Spine > **Matter**: AGRO-P-003 · **Date**: 2026-05-09 (pre-filing, 7-day window) · **Authority**: 18-claim spine ratified 9/9 unanimous Patent Council 2026-05-09 — **binding, additions only**. Strictly Private & Confidential. --- ## (A) Executive summary The fresh input does not unsettle the 18-claim spine; it extends it on three orthogonal axes the corporate council did not address. The *Apple/on-device* artefact reframes the technical effect from latency-and-bandwidth to **fixed-cost-per-inference and structural data-sovereignty for regulated services** — sharpening the s.1(2)(c) defence. The *Dean/Dally* artefact identifies **agent-native primitives** (persistent containers, shared KV-cache, MCP/SSE) absent from claim 17. Prior HealthAgroAI and Pulse Agro AI sessions add concrete claim-suitable elements: single-call multimodal diagnostic, batched herd inference, WhatsApp/messenger transport, Pi 5 + Coral TPU value-tier as co-equal limb, and named local-LLM specifics. Recommendation: add 6 new dependent claims (Y1–Y6) and three spec-level technical-effect framings; do not touch claims 1–18. --- ## (B) Per-input extraction (elements NOT already in the spine) **Input 1 — Apple / on-device.** Cost-structure framing (cloud=variable, on-device=fixed) distinct from latency/bandwidth in §3.1. Regulated-services-on-prem framing parallels UK VMD / EU MDR. "No rackable form factor / no clustering software / no on-prem identity" implies a fleet-clustering limb absent from spine. **Input 2 — Dean/Dally.** "Tool latency, not model latency" supports an agent-native API technical effect. Three primitives — persistent containers, BranchFS, shared KV-cache — none in spine. **Input 3 — HealthAgroAI top-5.** Agent-native API; single-call multimodal diagnostic; herd-batch endpoints; MCP server; SSE streams. None in claim 17. **Input 4 — HealthAgroAI hardware.** Named local clinical LLM (meditron:7b) distinct from claim 9's generic LLM. "80% diagnostics offline" — inference-layer envelope (claim 6 covers transport only). **Input 5 — Pulse Native AI.** **WhatsApp/messenger transport** absent from spine (claim 17 is EHR/PMS-only). Specific stack: Llama 3.1 70B cloud; Mistral 7B / Phi-3 / LLaVA 13B / Nomic Embed edge. **Pi 5 + Coral TPU (£100) vs Jetson Orin Nano (£200)** present in §1.2 but not yet a claim-1 limb. **Webhook ingest** of weather / soil / EID / walk-over weigher data extends claim 1(a) to heterogeneous sensor input. **Input 6 — HealthAgroAI patentability assessment.** System + method + **UK Registered Design** triad — design track is a parallel filing, flag for separate workstream. **Input 7 — Pulse patentability assessment.** "System + method + device + **follow-up-chain**". Follow-up-chain (local verdict → deferred re-sample, not cloud escalation) is an orthogonal fourth behaviour absent from claim 2. --- ## (C) Consolidated NEW claim-element candidates | # | Element (short form) | Cross-ref to spine | |---|---|---| | C1 | Agent-native API surface — MCP server, SSE event stream, batch-inference endpoint | extends claim 17 | | C2 | Single-call multimodal diagnostic endpoint (biomarker + image + free-text → one hypothesis) | extends claims 2 + 17 | | C3 | Persistent-container execution + shared KV-cache across inferences per animal-ID | new | | C4 | Messenger-channel user surface (WhatsApp Business API / SMS / RCS) | new (claim 17 is API-only) | | C5 | Locally-resident clinical LLM Markush (7B-medical / 7B-general / 3B-compact) | extends claim 9 | | C6 | Edge-gateway hardware Markush — Jetson Orin Nano OR Pi-class SBC + external TPU | extends claim 1(b) | | C7 | Heterogeneous on-farm webhook ingest (weather / soil / EID / walk-over weigher) | extends claim 1(a)/(b) | | C8 | Deferred-follow-up sampling protocol (4th orthogonal behaviour to three-band) | extends claim 2 | | C9 | On-farm fleet-clustering — shared identity + model-weight cache across gateways | extends §1.4 | | C10 | Cost-structure technical-effect framing (fixed-cost-per-inference at edge) | spec-level (Section E) | --- ## (D) Recommended new dependent claims (UK form) > *Note: all are dependent on claims 1 or 2 of the ratified spine; none amends an existing claim.* **Y1.** The system of claim 1, wherein the edge gateway exposes an agent-native application programming interface comprising a Model Context Protocol server endpoint, a Server-Sent Events stream of diagnostic events, and a batch-inference endpoint configured to receive a plurality of animal identifiers in a single request and to return a respective plurality of candidate-hypothesis outputs. *Effect:* reduces tool-latency-bound round-trips for software-agent consumers; Aerotel-survives as a contribution to *operation of the computing system* (Emotional Perception aligned). **Y2.** The system of claim 1, wherein the edge gateway provides a single-call multimodal diagnostic endpoint configured to receive, in one request, a sequence of biomarker readings, image data, and free-text contextual data, and to return a single candidate-hypothesis output and confidence score. *Effect:* eliminates serialised round-trips across separate inference services; measurable latency reduction; consolidated inference path, not clinical interpretation. **Y3.** The method of claim 2, further comprising maintaining a persistent execution environment at the edge gateway in which model weights and a shared key-value cache are resident in memory between successive inference calls for a given animal identifier, such that inference latency for a subsequent reading from the same animal is reduced relative to a cold-start inference. *Effect:* per-inference latency reduction tied to specific cache-residency behaviour; not a mental act, not a business method. **Y4.** The system of claim 1, further comprising a messenger-channel transport layer configured to deliver the diagnostic decision-support output to a registered user device via at least one of: a WhatsApp Business API channel, a Short Message Service (SMS) channel, and a Rich Communication Services (RCS) channel. *Effect:* integration of a low-bandwidth high-availability transport into the cyber-physical pipeline; COMVIK-survives because technical features are the transport-tier integration, not the message content. **Y5.** The method of claim 2, wherein, when the confidence score falls within a defined sub-band of the second band, the routing decision comprises scheduling a subsequent biomarker reading at the wearable biosensor at a defined interval rather than invoking the cloud-tier classifier, and revisiting the routing decision on receipt of the subsequent reading. *Effect:* a fourth, orthogonal behaviour added to the three-band policy; reduces cloud bandwidth and sensor-tier power consumption. **Y6.** The system of claim 1, wherein two or more edge gateways within a single farm site are arranged as a fleet sharing a common on-premises identity-provider service and a shared model-weight cache, such that a model-weight update applied to one gateway is propagated to remaining gateways without egress to the cloud tier. *Effect:* on-prem clustering with bandwidth and data-sovereignty effect; Alice/Mayo Step 2 cleared by specific hardware coordination. *(Optional Y7/Y8 reserved for the value-tier hardware Markush (C6) and heterogeneous-sensor webhook ingest (C7); both filing-ready.)* --- ## (E) Strengthening framings for s.1(2)(c) ("program as such") Pre-empt the GB2514773.7 examiner's program-as-such characterisation by broadening the spec's technical-effect register on three axes: 1. **Cost-structure effect.** *"The integrated edge-tier inference architecture converts a per-inference variable cost (cloud round-trip) into a fixed cost amortised over the wear-life of the edge hardware — a deterministic compute budget at the farm tier independent of network conditions and cloud-tier pricing."* Anchored in operation-of-the-computing-system (Emotional Perception AI [2024] EWCA Civ 825), not economics. 2. **Structural data-sovereignty effect.** *"The architecture preserves data sovereignty as a structural property of the system, not a policy overlay — raw biomarker readings are never transmitted from the edge tier under any operational mode."* Effect is on the information-flow topology of the computing system, not on the legal regime (UK GDPR / DPA 2018; EU AI Act Annex III; UK VMD records-of-use as ambient context). 3. **Agent-throughput effect.** *"Where the consumer of the diagnostic output is a software agent, the agent-native API surface (MCP / SSE / batch) eliminates the human-paginated, human-rate-limited tool-latency that otherwise dominates end-to-end response time."* A measurable machine-to-machine throughput effect — outside any "program as such" reading and inside Aerotel step 4. These three anchor the contribution in system-operation properties (cost determinism, information-flow topology, M2M throughput), inoculating against the GB2514773.7 program/algorithm-as-such framing. --- ## (F) Defensive design-around closures 1. **Structural on-prem closure.** Y4 + Y6 + Section E recast the inventive concept as **structural on-prem-by-design** (not deployable-on-prem), closing a competitor route via "private-cloud deployment". 2. **Agent-consumer closure.** Y1 + Y2 + Y3 close the route of a competitor exposing only an agent-shaped API to a third-party agent harness; a competitor agent-native veterinary CDS still infringes Y1/Y2 if it touches the edge tier as described. 3. **Value-tier hardware closure.** Disclosing both Jetson Orin Nano and Pi 5 + Coral TPU as alternative limbs of claim 1(b) closes the "we used cheaper hardware" workaround. 4. **Heterogeneous-sensor closure.** Webhook ingest of weather/soil/EID/walk-over-weigher data forecloses reuse of the inventive method on non-ISF inputs in the same architecture. 5. **Messenger-channel closure.** Y4 forecloses a WhatsApp-first competitor that would otherwise sit outside an EHR/PMS-only claim 17 — material in emerging-market livestock deployments. 6. **Registered Design parallel track.** Per Input 6, queue a UK Registered Design on (i) wearable enclosure form factor and (ii) edge-gateway housing — separate workstream, same week. --- *Memo ends. Recommend Patent Council micro-vote on Y1–Y6 before Day-2 specification polish per §5.3.*