Partner paper

Updated: 2026-09-17 · strapline reframe

Earlier body stamp retained for provenance: Updated: 2026-09-15 · R3 (accuracy-review merge). R3 applied the accuracy-review merge table: Bulgarian census denominators, FRAME Max order-size prices, and unsupported first/only/universal claims. Original analysis is retained and qualified; it is not silently deleted. This remains a working paper, not a validated product or forecast.

The operating system for aerial field work.

A proposed service OS for any size of land and any size of air company. Request/accept may use a familiar ride-hailing UX metaphor; that is secondary to the product definition.

Situation–complication–resolution spine. Facts are sourced. Unsourced numbers are labelled hypothesis or to be measured. Analyst totals disagree; they are ranges, not a model.

1. Executive summary

1. Aerial work exists. Hardware (DJI, XAG, Hylio) and Earth observation (Sentinel-2, Planet, Airbus) are on the market. Booking, live customer view and a regulator-shaped receipt are not a single proven product. Public offerings already overlap in the US, EU and UK. A landowner still often coordinates by phone; that is a workflow gap to test, not proof that no network exists.
2. Structure can lock people out, not physics. FAO SOFA 2025 / Lowder global farm-count and land-share figures are held out of the current investor headline until the primary table, year and denominator are reconfirmed. Bulgaria Census 2020 (publication No406, October 2022, hosted 2023 — not a 2026 census) [S48]: 132,742 total holdings, of which 5,464 without UAA. The reported 35.9 ha average applies to 127,278 holdings with UAA. The 37,452 count is the greater than zero and less than 1 ha UAA class (16,213 ha), not “≤1 ha.” Some named US DaaS operators publish acre minimums; a blanket 10–25 acre industry rule is not supported without named supplier evidence.
3. Two TAMs, one platform. Farm count and farm hectares are different markets. Serve only ≥50 ha estates and you address most land and almost no farms. Serve only smallholders and you address most farms and little land. The product hypothesis is a network any size landowner and any size air company can use. Instant, nationwide availability is a target subject to local delivery, permissions and operator supply — not a shipped capability.
4. Small jobs fail one-by-one. They work when batched. Hypothesis: pool nearby small jobs into one sortie; price the landowner per hectare; pay the operator on the pooled block. Live tracking is the proposed trust layer that makes a stranger acceptable on a 0.4 ha plot.
5. Software OS first. No drone factory. No FRAME Max in seed. EnduroSat published unit prices by order size: USD9.5m for one, USD8.1m for ten, USD6.7m for 100, USD4.2m for 1,000+ [S49]. The low end is a bulk-order bus price, not a single-satellite quote or full mission cost. EnduroSat is data / teleport / 16U for v1. Defense is civil and emergency only.
6. No verified incumbent combines the full proposed Canopy OS (instant book + any-size land + any-size operator + live customer view + satellite cue + EU chemical receipt) in Bulgaria, the EU or the UK. That is not a “first/only” claim. Closest named overlaps: AcreConnect / American Autonomy (US operations software and discovery; advertised from USD650/user/year [S02]); ProxyDrone.eu (Romanian B2B marketplace — PARTIAL overlap: job post, offers, execution report, payment; not a live landowner OS or BFSA receipt); FlyGuys, Droners, FlyNex, HireDronePilot and local BG contractors. Procore reported its DroneDeploy acquisition completed on 9 September 2026 [S46]; announced cash consideration was approximately USD845m, subject to adjustments [S47].
7. Bulgaria is the lab, not the company. Keep Census 2020 denominators distinct from the later MAF land-use series. Do not splice 2020 holdings with 2024 UAA. The Plant Protection Act already requires UAS treatment logs to BFSA. That is the receipt shape Canopy should write — a telemetry export is not automatically a legally accepted application record.

Recommendation. Build the service OS. Make 0.2 ha first-class via batching. Ship live tracking in the first slice. Pilot Bulgaria for 18 months. Do not buy airframes or FRAME Max. Go / no-go on 10–30 paying jobs with a live view and a BFSA-shaped receipt that a reviewer can accept. The current commercial experiment (R2/R3) is recurring property/construction evidence plus this own-field agricultural test; they are not simultaneous funded commitments.

2. Situation

Demand for aerial work is already priced. Service-market reports commonly cite spray at $8–25 per acre and mapping at $5–20 per acre or $500–1,500 per mission. US corn and soy drone spray is often $12–18 per acre. Observed offer prices, not Bulgaria quotes.

Airframes are not the bottleneck. DJI Agras is the default global spray fleet. XAG sits in Western DaaS fleets. Hylio, Guardian, and Pyka serve large-acre and NDAA-sensitive buyers. We will not out-manufacture DJI in year one.

Imagery is not the bottleneck. Sentinel-2 access is free on a five-day cycle; free access is not zero all-in processing, storage or delivery cost. Planet sells daily. Airbus sells tasking. EnduroSat sits in Sofia with a Balkan constellation story, a teleport, and a 16U line. FRAME Max is a different bus class with order-size unit prices (USD9.5m / 8.1m / 6.7m / 4.2m). It is not required to fly a field job.

What exists on the ground is a contractor market. A landowner calls. An operator texts. A price is agreed on WhatsApp. The job is flown. A PDF is typed later, if at all. The landowner does not see the pass. The small plot is refused or quoted a punitive minimum. The small operator cannot see a queue of nearby jobs. The ministry cannot surge the same network in a fire.

The missing object, as a product hypothesis, is a working request → match → pool → fly → watch → receipt → cue loop that a buyer will pay for. Instant availability is a target, not a present capability.

3. Complication

Three locks. MECE: demand-side, supply-side, trust-and-state.

3.1 Demand-side lock — minimum acres

Some named US DaaS operators publish acre minimums. A universal 10–25 acre industry floor is not established by this review. Where a named operator does impose a 10-acre (~4 ha) or 25-acre (~10 ha) minimum, a 0.8 ha vineyard or most Bulgarian holdings under 10 ha are out of that operator’s standalone quote.

This is rational for a standalone sortie. Ferry, mix, setup, and cleanup do not scale down with plot size. It is not required as a platform rule. A platform that can see density can batch. Whether buyers and crews will pay for that batching is a measured hypothesis.

Global structure makes the lock material. ~85% of farms are under 2 ha — ~500 million smallholders on ~9% of land. They produce about 16% of global crop dietary energy. They are not “80% of the world’s food.” That figure is a myth. We do not use it. They are still the majority of customers if the product is a booking.

Bulgaria is the same shape among holdings with UAA. Census 2020: holdings under 10 ha are the majority of count and a small share of hectares. The 13,366 holdings ≥50 ha hold 84% of UAA. A DaaS that copies a US acre floor would be a product for those 13,366, not the 127,278 holdings with UAA (132,742 including holdings without UAA). Holdings already fell 64% versus 2010. Consolidation is not a reason to ignore the remaining small plots.

3.2 Supply-side lock — WhatsApp dispatch

A two-drone family crew and a 20-drone contractor are both operators. They find work by phone, dealer referral, and season gossip. No live queue. No pooling across landowners. No standard receipt. Small air companies cannot see enough density to make a 0.5 ha job worth the drive. Large air companies ignore those jobs because dispatch is built for blocks.

The lock is mutual. Small landowners cannot find a crew that will come. Small crews cannot find a cluster that pays.

3.3 Trust-and-state lock — no live view, no artefact

A landowner who has never met the pilot will not send a stranger over vines with a tank of product on a text thread. Live tracking is the proposed trust layer (a secondary UX metaphor familiar from ride-hailing). That mechanic is missing here.

The state has the opposite problem. Bulgaria’s Plant Protection Act requires UAS treatment logs (coordinates, time, volume) to BFSA. EASA treats drop of material as Specific category; PDRA-S01 is the usual path. Directive 2009/128/EC bans aerial spray of plant-protection products, with national derogations. The legal artefact is required. A Canopy export is a candidate pack; it is not automatically a legally accepted pesticide application record until BFSA / the relevant directorate accepts the fields. The typed PDF is the failure mode we propose to replace.

Complication, one line: the people with the most farms cannot book; the people with the most aircraft cannot see a queue of small jobs; nobody watches the pass; the register is filled by hand.

4. Resolution — Canopy as a proposed job OS

Canopy is a proposed service operating system. Not an airframe. Not a satellite. The layer that would make any size of land and any size of air company first-class if matching, batching, live view and accepted receipts work.

The intended motion is operating-system job booking and matching, not enterprise SaaS:

1. Landowner opens a map. Drops a parcel. 0.2 ha and 10,000 ha use the same request.
2. Job type: survey, spray, spread, forest health, civil/emergency.
3. Network matches licensed operators by proximity, airframe, licence, tank, and weather window.
4. Nearby small jobs are batched into one sortie so the operator’s block clears an economic floor (hypothesis: §8).
5. Landowner watches the pass live — aircraft, progress, remaining hectares, hold reasons — as a rider watches a car.
6. Receipt is proposed from labelled source evidence (crew-device beacon, OEM export, or aircraft telemetry — provenance required). BFSA field names, EASA log, insurer pack. A crew phone in a truck is not the aircraft path.
7. Satellite and weather sit on the same map: cue, hold, verify.
8. Optional: a supplier delivers the labelled mix to the pad. Chain of custody into the receipt.

Same OS for forestry and civil protection. Payloads stay civil. Not weapons.

Out of seed scope: a drone factory; a FRAME Max purchase; becoming a pesticide company; offensive ISR. EnduroSat is data, teleport, and 16U tasking. FRAME Max remains year-3+ optionality at published order-size unit prices, not a seed line item. The Bulgaria test does not require it.

Small plots are not a discount SKU. They are how density is created for operators. Large estates capture hectares and compliance complexity. Both must work on day one.

Product pages: product/. System narrative: product/00-system.md.

5. Market

Treat analyst totals as order-of-magnitude. Do not average them into a fake point estimate. Do not use the global farm-count / land-share rows below as the current investor headline until the primary FAO/Lowder table, year and denominator are reconfirmed.

Exhibit 1 — Global farm structure

Source: FAO SOFA 2025 / Lowder update. Sourced fact.

Exhibit 2 — Dual TAM

We play both. That is the mandate. Not a later phase.

Exhibit 3 — Bulgaria holdings (Census 2020 results publication No406, October 2022, hosted 2023)

Sourced fact [S48]. Not a 2026 census. 35.9 ha average applies to 127,278 holdings with UAA, not to all 132,742 holdings. Holdings fell 64% versus 2010.

Holdings under 2 ha with UAA: ~57,250 holdings, ~44,300 ha. Under a named 10-acre (~4 ha) minimum, most of the first three UAA classes are unbookable as standalone jobs. They are bookable as a pooled cluster if density and crew economics work.

A later MAF land-use series exists (methodology differs from Census 2020). Do not splice the 2020 holdings denominator with a later area series. Hold the 2024 UAA/arable totals out of the current investor headline until the primary table, year and denominator are reconfirmed.

Exhibit 4 — Global drone market ranges

Named analysts. Ranges disagree. Not a forecast we own.

Hardware-plus-precision-ag is a mid-single-digit billion market. Pure spraying services are still small (~$0.4B in 2026) and the faster-growing slice in this set. Canopy sells the service OS and the take-rate on jobs, not airframes.

Observed offer prices (not a Canopy price list): spray $8–25 / acre; mapping $5–20 / acre or $500–1,500 / mission; US corn/soy often $12–18 / acre. Bulgaria €/ha: to be measured. Forestry and civil protection are mission-priced until a buyer contract exists. Do not add them into the agri-drone totals.

6. Competitive landscape

Closest named US analogue: AcreConnect / American Autonomy (ex-Rantizo software). Pattern to copy: match licensed operators to jobs and keep operations records. Pattern not to copy: assuming US acre floors, US chem labels, or the absence of a live landowner view and EU receipt. Sale of operations in 2025 is evidence the software layer can be separated from the crew; it is not evidence they serve 0.4 ha plots in the EU. Regional PARTIAL overlap: ProxyDrone.eu (RO marketplace). No full Canopy OS was verified in BG / EU / UK.

Exhibit 5 — Competitive matrix

Local EU contractors are the real alternative in Bulgaria today. Switching cost is a phone number. The barrier is density and trust.

7. Where to play / how to win

Where to play. Bulgaria first (UAS already named in the Plant Protection Act; EnduroSat in Sofia). CEE is expansion, not the pilot. Survey first; spray second, on a confirmed derogation. Forest and civil/emergency as the same OS. No weapons. All land sizes and all operator sizes on day one. No platform hectare minimum. Pooling is a launch feature. Software and network only. Partner airframes. Partner EO (Sentinel-2 now; EnduroSat tasking as stretch). No factory. No FRAME Max.

How to win. Be a network a 0.6 ha owner and a 6,000 ha estate both open, and a queue a two-drone crew and a regional contractor both trust. “First” is not a supported market claim.

1. Instant request. Parcel on a map. Price or range before a phone call.
2. Batching as the default for small jobs. Landowner sees their plot. Operator sees a cluster.
3. Live track. If the landowner cannot watch the pass, we are a contractor directory.
4. Receipt as a candidate legal artefact. Labelled source evidence writes BFSA / EASA field names. Acceptance by BFSA remains unproven.
5. Cue from space and weather. The job exists because a tile or a window said so.
6. Supplier attach. Tank mix to the pad, label-legal, chain of custody.
7. Civil surge. Ministry books the same network.

Winning is not cheapest $ per hectare against a tractor. Not a Chinese airframe with a Velocity sticker. Not a constellation press release.

8. Unit economics of small jobs

Hypothesis. Sourced offer prices are inputs only. Bulgaria realised prices, ferry times, and cluster densities are to be measured.

A standalone small job fails because time does not shrink with the plot. Take the sourced US range of $12–18 / acre for corn/soy. Convert only for illustration: 1 ha ≈ 2.47 acres, so $30–45 / ha if that tariff travelled. It may not. A 0.6 ha plot at an illustrative $35 / ha is ~$21 of GMV. Drive, mix, rinse, weather abort, and the chem log sit against that $21. A standalone acre floor follows for that operator. At 10 acres (~4 ha) the same $35 / ha is ~$140. At 25 acres (~10 ha) it is ~$350. Those floors are why the 37,452 Bulgarian holdings in the >0 and <1 ha UAA class often never see a standalone quote.

Exhibit 6 — Pooling (hypothesis, not measured)

Conditions: plots share one weather window and one ferry radius (radius to be measured); labels are compatible or the sortie splits by product; the operator is paid on the cluster; the landowner is not forced to wait indefinitely — max wait and a “fly anyway” price are to be designed. Survey jobs pool more easily than spray. Learn on survey.

Failure mode: clusters never form. Then the any-size claim fails. That is a go / no-go input. Large jobs do not need this math. They must still run on the same rails (live, receipt, cue).

9. Product system

One job object. Six products around it. Pages in product/.

Thin slice: Request survey → operator accepts → live map → receipt. Spray is slice 2, after one legal path is confirmed. Cue uses Sentinel-2 and open weather first. Supply and Duty are designed in the first 18 months; not necessarily shipped.

AI is in the loop: prescription, match, in-flight hold, post-job change detection, receipt mapping. The human remains PIC and licensed applicator.

10. Eighteen-month roadmap (Bulgaria pilot)

A proof that the proposed job loop works on real holdings, including sub-2 ha plots.

Success bar: 10–30 paying jobs with live view; a non-zero share on holdings <2 ha, flown as pooled clusters (exact share to be set at Gate 0); one forest or civil-protection demo, not a weapons demo; EnduroSat data / ground-station first; written go / no-go on TAM + regulation + operator supply.

11. Scale path

Long form: Canopy_Scale_Roadmap.md. Not everywhere at once. The 18-month pilot is Phase 0 only.

Exhibit 7 — Geographic phases

OS scales without hardware. Five levers: marketplace density (operators + batched small jobs); local chem partners; per-country receipt; satellite as bought data (Sentinel now; commercial / EnduroSat later); Stripe split payout already in the stack. Not a factory. Not a FRAME Max purchase (published unit prices USD9.5m / 8.1m / 6.7m / 4.2m by order size [S49]). Shared software does not confer public-safety or military qualification.

Civil / emergency — hard constraints. Duty is a requester type on the same Job OS. In scope: wildfire, flood, SAR, comms relay, damage assessment, chemical-spill mapping. No weapons, munitions, targeting, or offensive ISR. Complementary to civil protection and critical-infrastructure resilience — not a defense prime. EO cues and verifies; it is not a weapons sensor.

Exhibit 8 — Why quick deployment is the product

Sourced note: Canopy_Rapid_Deploy_Benefits.md. Additional dollar or hectare figures are not extrapolated beyond the sourced note.

12. Risks

13. Recommendations

1. Adopt the mandate as a product rule. Any size landowner. Any size air company. Instant availability is a target, subject to local delivery and approvals. No platform hectare minimum. Pooling is how small jobs may pay. If a screen asks for a minimum hectare, the design has failed.
2. Build the OS, not the factory. Marketplace, live view, receipt, cue, supplier hook. Partner airframes. Partner EO. Human remains PIC.
3. Do not buy FRAME Max in seed. EnduroSat scope is data, teleport, and 16U. Revisit an owned high-power bus only if the data layer is the moat and revenue exists.
4. Pilot in Bulgaria on the Census shape, not the 35.9 ha average of holdings with UAA. Book the >0 and <1 ha and 1–<2 ha UAA classes as first-class. Do not headline unverified later land-area series. Founder-owned test land is a separate experiment (~1,000 acres / ~405 ha stated base) and does not prove marketplace liquidity.
5. Ship in this order: Request → Fly (including batch) → Live → Receipt. Cue on Sentinel-2. Spray only on a written legal path. Duty as a demo, not a weapons programme.
6. Treat pooling economics as a measured hypothesis. Instrument ferry, hop, cluster size, wait, and operator net per hour. Kill or redesign if clusters do not form.
7. Price as a take-rate plus optional seats. Illustrative take-rate 15–25% of GMV is a hypothesis, not a board number.
8. Keep defense civil. Wildfire, flood, SAR, comms relay, damage assessment, chemical-spill mapping, forest pest, post-event inspection. No munitions. No targeting. No offensive ISR. Duty is a requester type, not a second stack.
9. Scale in phases, not “global DaaS.” Bulgaria → Balkans/SEE → wider EU → UK, then selected non-EU. US is a later any-size wedge against named acre-minimum incumbents, not a year-1 launch. Scale the OS by density, local partners, country receipts, bought EO, and Stripe — not by buying airframes or FRAME Max.
10. Go / no-go at month 18 on 10–30 paying jobs with live view, a non-zero small-plot cluster share, one civil/forest demo, and a receipt a directorate would accept. If the jobs are only large estates, we have not built Canopy.
11. Write nothing into a raise that assumes a drone plant or a constellation. Software-led Canopy is a marketplace + compliance + data company. A FRAME Max owner is a space-infra company. Do not confuse the two.

Appendix — Sources

Facts in the body are limited to the items below plus named competitor and regulatory instruments. Market totals are analyst estimates and disagree. Bulgaria €/ha, cluster density, take-rate, and wait tolerance are not sourced.

1. FAO, The State of Food and Agriculture 2025, and Lowder et al. update — global farm count ~570–579 million; <2 ha ~85% of farms and ~9% of land (~500 million smallholders); >50 ha <1% of farms and ~75% of land; >1,000 ha 0.1% of farms and ~half of farmland; smallholders ~16% of global crop dietary energy. The “80% of the world’s food” claim is not used.
2. Bulgaria, Ministry of Agriculture and Food, Census of Agricultural Holdings 2020 results publication No406 (October 2022, hosted 2023) — 132,742 total holdings; 5,464 without UAA; 127,278 with UAA; UAA 4,564,152 ha; 35.9 ha average on holdings with UAA; 37,452 holdings in the >0 and <1 ha UAA class; holdings −64% versus 2010. [S48]
3. A later MAF land-use series exists and uses a different methodology. Do not splice it with Census 2020 holdings. Hold later UAA/arable totals out of the current investor headline until the primary table is reconfirmed.
4. Grand View Research, agriculture drones — $3.4B (2025) → $3.9B (2026) → $13.2B (2033), 18.8% CAGR.
5. Fact.MR, agriculture drones — $6.5B (2026).
6. Fact.MR, drone spraying services — $0.4B (2026) → $3.8B (2036), 25.3% CAGR.
7. Service-market reports (trade summaries) — spray often $8–25 / acre; mapping $5–20 / acre or $500–1,500 / mission; US corn/soy drone spray often $12–18 / acre. Not Bulgaria field quotes.
8. Company public materials and trade press — DJI Agras; XAG; Hylio; Guardian Agriculture; Pyka; AcreConnect / American Autonomy; ProxyDrone.eu (PARTIAL marketplace overlap); named US operator minimums where published (not a universal 10–25 acre rule); Rantizo sale of operations in 2025; DroneDeploy / Procore (completion reported 9 Sep 2026 [S46–S47]); Taranis; Sentera; Planet; Sentinel-2; Airbus; EnduroSat Balkan.
9. EnduroSat public materials — FRAME Max published unit prices by order size: USD9.5m (1), USD8.1m (10), USD6.7m (100), USD4.2m (1,000+) [S49]; 0.5–3 t payload; 4 Gbps. Not a v1 purchase. The USD4.2m figure is a bulk-order unit price, not a single-satellite or full-mission cost.
10. Regulation (EU) 2019/947 and 2019/945 — Open / Specific / Certified. Drop of material → Specific.
11. EASA PDRA-S01 — predefined risk assessment path for certain Specific operations.
12. Directive 2009/128/EC — aerial spraying of plant-protection products banned; national derogations possible.
13. Bulgaria, Plant Protection Act (as amended) — UAS named; treatment logs to Regional Food Safety Directorate / BFSA.
14. European Commission, COM(2025) 1021, Food & Feed Safety Omnibus — proposed drone-related adjustments; watch 2026–27. Not treated as enacted law here.
15. Canopy research notes: research/00-thesis.md through research/08-defense-emergency.md; legal/regulatory-sources.md.
16. Scale and rapid-deploy notes in this folder: docs/Canopy_Scale_Roadmap.md; docs/Canopy_Rapid_Deploy_Benefits.md.
17. DTN Progressive Farmer, 5 August 2026 (Taylor Moreland / Agri Spray Drones) — timeliness of the spray/scout window.
18. AgriNews, 13 August 2026; Queensland DPI; Ohio State CFAES fact sheet — small / irregular / sloped / wet / remote parcels; forestry spray often >30 ft AGL; complement not replacement.
19. ICAERUS, Horizon Europe 101060643 (ended ~2026) — LCA on CO2 and agrochemical use; open-call environmental trials. SPADE, Horizon Europe 101060778 — agri + forestry + rural observation as one services market.
20. FAO, State of the World’s Forests 2024 — drones + AI / lidar / optical as forest-management innovation.
21. DroneXL, 26 June 2026; GreekReporter, 23 May 2026; eKathimerini — Greece 2026 civil-protection drone plan.
22. Dryad Silvaguard / Silvanet — Eberswalde demo; ERDF co-fund €3.8M; Dryad claim that unwanted wildfires are up to 20% of global CO2 (Dryad, not FAO). Suppression R&D is Dryad’s roadmap, not Canopy v1.

Velocity EU — Canopy working paper — 16 August 2026 — CONFIDENTIAL