ARBITER
HarvestGuard
A solar cold-storage pod — and the verified-data credit rail it builds for the smallholder farmers finance cannot see.
DomainAgriculture & Food Security
StageDesigned · engineer-reviewed
Ask10-pod pilot, Meru, Kenya
The Problem
The cold chain stops before the smallholder farmer.
<1%
of the world's refrigerated storage is in Africa
50%
of the fruit & vegetable harvest lost after harvest (up to)
33M
smallholder farms carry that loss each season
<6%
of commercial loans ever reach these farmers
GCCA/IIR Global Cold Storage Capacity Report (2023) · FAO/IIR post-harvest loss, SSA (2022) · FAO Family Farming; One Acre Fund (2023) · IFC, Scaling Farmer Financing in SSA (2024).
The Thesis
The pod is the wedge.
The verified data is the moat.
01
Cold at the farm gate
A solar pod, sub-village scale, billed per crate-day over M-Pesa — one 4–13 °C zone, crop-cohorted by the app.
02
A record by construction
Every use is device-signed and payment-reconciled — a history, not a claim.
03
A credit identity
The first credential a lender can act on — cold storage as an on-ramp to finance.
System Design
Off-the-shelf refrigeration, sized for 35 °C and no grid.
1.5 m³ · 80 mm PUR290 · Secop BD35F330 W · Victron 75/3516× EVE LF100 · 4.8 kWhESP32-S3 · X.509 · offline-first
Engineering Model
A full CAD model — down to the crate, the compressor, the seal.
01 · CAD · cutaway
Real part footprints: Secop BD35F compressor · Victron MPPT 75/15 · 16× EVE LF100LA (4S4P) · ESP32-S3 · 330 W panel. 80 mm layered wall, 2 crate shelves, sealed bulkhead.
02 · FEA mesh
Full assembly discretised for thermal simulation — Fusion 360 FEA is a Phase-0 deliverable.
03 · Thermal
Surface heat-ingress ≈ 99 W · peak on the sun-facing roof & door seal · from the first-principles budget.
Materials & Construction
Chosen for heat, humidity, and dust.
FRAMEPowder-coated galvanised steel — welded, corrosion-proof in humid heat, locally fabricable.
DOORSteel + EPDM magnetic gasket (UV-stable, 18–24 mo) · 3× 304 stainless hinges · 12 V solenoid bolt.
FLOORAnti-slip aluminium tread plate — drains condensate, takes crate load.
REFRIGERANTR290 (propane), GWP 3 — Secop BD35F hermetic DC compressor.
HEAT EXCH.Roll-bond aluminium evaporator + finned condenser, externally vented, 12 V fan.
EXTERIORWhite PU topcoat — reflects solar load; black would absorb 20–30% more heat.
Engineering Feasibility
The energy budget closes — with margin.
1.34
kWh/day generated — 330 W array, Meru sun
1.10
kWh/day used — Secop BD35F at 35 °C, ~83% duty
+235
Wh/day surplus at an 8 °C interior setpoint
~89h
battery autonomy with no sun (16× LF100, 4.8 kWh)
A first-principles budget, not a datasheet quote — Phase-0 bench testing validates the BD35F and real autonomy under dust, door-openings, and the 40 °C worst case.
Independent first-principles thermal/energy analysis. Verify vs. Secop BD35F R290 performance table and NASA POWER irradiance (Meru, 0.05 °N).
The Trust Layer · core IP
A record a lender can trust — by construction.
01
Device-authenticated
A per-pod X.509 certificate; a record from any unauthenticated source cannot enter the system.
02
Independently witnessed
Each transaction reconciles against the M-Pesa settlement API — a payment network no party controls.
03
Tamper-evident
Append-only, offline-first telemetry; gaps from lost connectivity are marked, never hidden.
The Credit Pipeline · in development
Behavioral data → a score a regulated lender can act on.
We use transactional, behavioral data — a stronger credit signal than yield prediction (Accion, 2026). Trained only after the pilot; scores stay advisory until a lender validates.
Positioning
Proven parts, an unfilled seam.
ColdHubs · InspiraFarms · Koolbokssolar cold-chain
Proved solar cooling and pay-per-use, but stay cold-space or hardware providers, not a farmer's credit rail.
Tala · FarmDrivealt-data credit
Proved thin-file lending from behavioral data — but on phone/agronomic data, not a verified physical-storage record.
HarvestGuardthe seam
A trust rail, not a landlord: every stored crate becomes a farmer credit record no incumbent builds.
ColdHubs 2024/2025 (coldhubs.com) · InspiraFarms Asset Finance · Koolboks (TechCabal 2025) · Tala 2023 · FarmDrive (FAO). Competitive scan to our knowledge, from public disclosures.
Validation & Milestones
A testable path, with kill criteria at every gate.
Phase 0 · Bench
M1–3
Prove it under failure
±0.5 °C hold · 100% state recovery · ≥18 h autonomy.
Phase 1 · Pilot
M3–15
10 pods, Meru
≥95% uptime · ≥70% utilisation · 100% M-Pesa reconciliation.
Data gate
M12–14
Train model v1
≥200 farmers × ≥3 cycles × ≥12 months.
Phase 2 · Scale
M15–24+
100 pods · live credit
AUC > 0.65 · a lender validates on holdout.
The pilot's central deliverable is not revenue — it is the first verified dataset proving pod records predict repayment.
Status & The Ask
Designed, engineer-reviewed,
ready to pilot.
Done
Hardware design, independent engineering review, firmware & data architecture, energy budget.
The ask
A 10-pod field pilot in Meru County, Kenya, with a local operations lead and an NGO partner.
Honest risks
Credit model needs a lender partner for labels; autonomy is validated 8–13 °C; 40 °C is a bench item.
HarvestGuard · A product of Arbiter · dayeon603@gmail.com
Key References
Every figure is sourced.
Cold chain. GCCA/IIR, Global Cold Storage Capacity Report (2023).
Post-harvest loss. FAO/IIR, Post-harvest losses in SSA (2022).
Smallholders. FAO Family Farming; One Acre Fund (2023).
Credit gap. IFC, Scaling Farmer Financing through AgTechs in SSA (2024).
Climate. FAO, Food Wastage Footprint (2013).
Alt-data credit. CGAP, M-Shwari (2015); Tala Impact Report (2023).
Ag credit. FarmDrive (FAO e-Agriculture; RFILC 2020); Apollo (GSMA 2024).
Caveat cited. Digital Frontiers / Accion (2026).
Cold-chain precedents. ColdHubs (IIR 2024); Koolboks (GSMA 2024).
Data → finance. Pula (FarmFit, 2026).
Figures tagged SOLID / APPROX per verification; the "3–5% in cold chain" figure is footnoted as an industry estimate, not a primary FAO statistic.