A Proposal for Scrappy

The Shield Works R&D team has reviewed the full document set. What follows is not the Phase 1 output. It is the position our engineering work starts from, set out here so the scope and the value of Phase 1 are clear before any commitment is made.

One point on how to read it. Identifying these questions is not the same as answering them. What Phase 1 does is convert the positions below into documented engineering recommendations, supported by supplier input, and a costed plan that can be funded in stages. That is the difference between knowing where the risk sits and having a product that can be deployed at scale.

The document set is unusually good. The PRD is a controlled requirements document with 41 logged engineering decisions, each traceable to a named owner and a piece of evidence, a field failure log drawn from live deployments rather than speculation, and an honest register of what remains open. It is well past the standard of brief we normally receive, and it makes this section possible. The findings below are offered in the same spirit.

The engagement scope note states that on technical content the PRD wins and on engagement scope the note wins. We have worked to that hierarchy. It matters because the original enquiry of 22 July predates the PRD, and 3 of its headline requirements have since been superseded, one has been clarified rather than changed, and the design status has moved. One further point of scoping departs from the PRD as well as the enquiry and is recorded at the foot of the table. We have scoped to the PRD throughout, and record the differences here so there is no ambiguity later.

One point of clarification on the LED, because the two things are easily conflated. The call record notes that the LED matrix needs to become a screen. That is the on-device display requirement, and we have scoped it. It is a separate matter from illumination, which the PRD deliberately deletes on the grounds that a light source over a deep bin creates hard shadows and hot spots and makes recognition worse. We agree with that reasoning and have not reintroduced a lamp.

Most of this product is well understood. Two questions dominate cost, performance and risk, they are connected, and neither is resolved in the current documentation. Phase 1 carries each as a dedicated engineering workstream.

The sealed envelope, and what has to happen inside it. The unit must dissipate in the region of 15 to 18 watts typically, and must stay within limits at the 25.5 watt ceiling that COMP-03 permits, inside a sealed stainless enclosure, in a hot kitchen, with passive cooling and no fan, while keeping a downward camera lens clear of grease aerosol and condensation film, and while carrying cellular antennas that have to work from inside a metal box in a basement. Four requirements pull against each other here. Sealing fights heat rejection. Heat rejection fights silence and the seal. A pressure equalisation vent is mandatory rather than optional, because a sealed metal box that heat cycles in a humid room will condense internally and the first surface it will condense on is the camera window, but that vent is also a deliberate hole in the ingress boundary. And a stainless head blocks radio, so the antennas need a non-metallic radome panel, which is another aperture in the same sealed, hygienic, premium surface. Two further apertures belong on the same list and are easily overlooked. The weighing deck has to move freely for the scale to work at all, so the base carries a mechanical discontinuity that cannot simply be welded shut. And the external antenna bulkhead required by COMP-06 is a blanked penetration on every unit, whether or not a given site uses it.

The PRD is candid that this is the largest open area, and the evidence currently offered for the thermal position is field experience across roughly 120 units. That evidence is real but it does not transfer, because the current unit is not sealed. Sealing the enclosure changes the thermal problem entirely. This needs a stated ambient envelope and a validated conductive path to the enclosure wall, not an assumption, and that path is sized against the sustained worst case rather than the typical figure, because a design that only holds at the typical load throttles at the ceiling.

There is a related mechanical problem inside the same workstream. The camera has to sit roughly 1.1 to 1.3 metres above the platform to clear the bin rim, and the unit has to ship flat in a single parcel. That means a sectioned pole. Sectioning the pole is precisely what produced the wobble the enterprise evaluation rejected. Making a demountable joint that is genuinely rigid, indexed against rotation, sealed, and able to carry power up and a digital signal down is not a catalogue part. In our assessment it is the highest technical risk item on the product and it needs prototyping and cycle testing rather than specification.

Stability is the part of that problem the document set does not address at all. MECH-03 requires no perceptible wobble and a head that holds its camera framing under normal knocks, and MECH-12 specifies passive levelling feet on a graded kitchen floor. Nowhere in the requirements is there an overturning criterion, a sliding criterion, a stated footprint, or a method for validating any of them. That gap matters, because a free-standing tower carrying a head at 1.1 to 1.3 m is a materially harder case than it appears, and it needs an overturning calculation before a footprint is fixed. On the mass estimate in section 2.3, a 28 kg unit on a 600 mm base resists overturning about its front edge with roughly 80 newton metres, so a horizontal push at 1.3 m of around 60 newtons, under 7 kg of force, is enough to begin lifting it. A loaded bin on the deck improves that considerably, which means the governing case is the empty unit with the bin wheeled away for emptying, and that is precisely when staff are moving around it.

There is a second and separate case at deck level. Pushing a bin that can weigh 100 kg up the ramp applies a horizontal force into the unit before much of the bin weight has transferred onto it, and on a wet kitchen floor the passive levelling feet of MECH-12 may not hold it. Sliding is as much of a failure as tipping and it is the more likely of the two. Both cases pull against MECH-04, because the cheapest source of stability is mass and the weight target asks for the opposite. That conflict is better named than discovered. The available resolutions are footprint and mass distribution, high friction feet that remain cleanable under MECH-11, and if neither is sufficient an optional floor or wall anchor as a documented site variant. We would flag that last option carefully, because anchoring into a client's building reintroduces the permission conversation that cellular-primary connectivity was chosen to avoid.

The weighing chain, which is the product's entire data claim. The commercial promise is audit defensible waste data, so accuracy is a hardware outcome. The current specification is lighter than the application warrants in five respects. Automatic taring corrects the zero point but does nothing for span or gain drift, so a cell can degrade silently over a lease. A single point load cell with a bin parked off centre, or mid roll up a ramp, is a moment loading problem rather than a simple weight. Load cells drift with temperature, and a kitchen is a temperature cycling environment. A bin liner change is not a bin change: liner weight varies by tens of grams against a 50 gram display division, which is the same order as the quantity being measured. And the specification does not say where the tower stands. If the pole foot lands on the weighing deck, the mass of the pole and head becomes permanent dead load, and a lean or a knock at head height is read as weight rather than rejected, because a sustained lean is not the transient that WEIGH-06 filters. The pole has to mount to the base chassis with the deck carried on the load cell alone, and any harness crossing that boundary has to be routed as a deliberate slack loop of negligible stiffness rather than left to the assembler. A harness pulled taut across a weighing boundary is a force shunt, and it is one of the most common ways a correctly specified scale is ruined in assembly.

None of that is a reason to change the architecture. It is a reason to engineer the chain deliberately, to specify a factory calibration process with traceability rather than a field step, and to define the settle window and transient rejection behaviour explicitly. Phase 1 establishes that specification and the end of line calibration station that enforces it. One consequence is worth stating early, because it reaches back into the base concept. An isolated deck and a washdown seal pull against each other, since a bonded perimeter skirt is watertight but stiff, and stiffness at that boundary is measurement error. The route washdown platform scales normally take is to abandon the perimeter seal, make the deck a fully welded sealed pan with the cell inside it, and allow the void beneath to wet and drain. That is a different base architecture from a deck, pan and skirt, and it changes both the fabrication and what the leak test is actually testing.

Requirement MECH-04 sets a maximum unit weight of 23 kg. It is worth being precise about what that number is, because two different limits are folded into it. Single person handling is a manual handling limit and it applies to the packed parcel rather than to the bare unit, so packaging comes out of the same allowance. Carrier single parcel limits vary by region and by carrier and should be confirmed against the carriers actually used. On our arithmetic the 23 kg figure is not achievable for a unit with an integrated ramp, and the gap is wider than a bare fabrication estimate suggests. It is better raised now than discovered in engineering.

The footprint has to be derived before the weight can be, and the document set does not state one. The enterprise feedback recorded in the PRD is that a standard bin, either an EN 840 wheelie or a round Brute, does not fit the current scale footprint, so the new platform is sized from the bin rather than measured off the current geometry. Working from MECH-07, the platform is sized by the EN 840 120 L wheelie, which is in the region of 505 mm wide by 560 mm deep, subject to confirmation of the actual make and model under open item 3 of section E2. Adding positional tolerance for an untrained user pushing a bin on, and edge clearance so the bin body does not overhang and load the deck edge, gives a deck of approximately 600 by 600 mm. That is the smallest defensible platform rather than a target, and it holds only if the wheel and lug contact points of the bin fall inside it while the wider body overhangs. The round Brute sits comfortably within the same deck. We have used 600 by 600 throughout below.

On published sheet weights for 304 stainless at that footprint, a 2 mm deck is 5.7 kg, a 1.5 mm base pan 4.3 kg, and a 1.5 mm skirt at 70 mm high 2.0 kg. The deck gauge is not easily reduced, because a loaded bin puts 60 to 80 kg through each wheel on a small contact patch and 1.5 mm material will dent locally under that. With a ramp at 5 to 7 kg, discussed below, that is a base weldment in the region of 18.7 kg. Adding the load cell and its mounting at around 2 kg, 4 levelling feet at 0.4 kg, a pole assembly at roughly 4 kg, a head with its enclosure and internals at around 3 kg, and harness, glands and fasteners at 0.5 kg gives a net unit of approximately 28.5 kg, or approximately 31 kg packed. The shortfall against a 23 kg packed allowance is therefore in the region of 8 kg.

The ramp is the swing item, its mass is set by the gradient, and the gradient is set by the deck height. MECH-08 requires a gentle one person gradient for a bin that can weigh 100 kg loaded. At a 70 mm deck height and a 1 in 10 gradient the ramp is 700 mm long, which in 2 mm stiffened stainless is close to 6.7 kg. At 1 in 8 it falls to around 4 kg, but the gradient is then no longer gentle, and 1.5 mm material unsupported over that span will deflect under a loaded bin. Any ramp weight that does not state the gradient it assumes is not a figure at all.

That exposes a conflict between three Must requirements which the document set does not reconcile. MECH-07 sizes the platform to the 120 L wheelie, MECH-02 and PKG-01 require the whole unit to pack into a parcel of approximately 600 by 600 by 300 mm, and MECH-08 requires a ramp long enough to be gentle. A 600 mm deck cannot enter a 600 mm parcel, and a 700 mm ramp cannot lie flat in it. The parcel envelope is therefore not a reference dimension with slack in it. It is the binding constraint on the platform, and one of the three requirements has to move. We would rather put that in front of you now than resolve it silently in engineering.

Deck height is the dimension that governs all of this and it is not specified anywhere in the document set. It sets the ramp length and therefore the ramp mass, it sets the package space available for the load cell and its mounting, and it is what MECH-09 is really asking about when it calls for a low platform with no high lip. Fixing it is a Phase 1 output, and every figure downstream of it is provisional until it is fixed.

There are good engineering answers and Phase 1 will land one. A ribbed or swaged deck in thinner material with local doubler plates only under the wheel tracks recovers 1.5 to 2 kg for a small forming cost. An open drained frame in place of a full base pan recovers 3 to 4 kg and improves drainage at the same time. Lowering the deck shortens the ramp and takes mass out of both at once. An aluminium ramp would remove around 4 kg, but it introduces a dissimilar metal joint in a washdown environment and a visible finish break against BRAND-02, so we would want to test that against the premium finish requirement before recommending it. Steel grade is not a lever here, since 316 is only around 1% heavier than 304. The decision that actually resolves this, though, is whether the ramp travels inside the parcel. Making it a separate detachable item takes 6 to 7 kg out of the shipped tower, simplifies the base weldment, and is consistent with the unbox order being the build order in ASSY-05. With an integrated ramp we do not believe 23 kg packed is reachable. With a detachable ramp it is close. That is the question we would put to you, rather than a request to relax the target.

The PRD names washdown survival as the critical quality bar and specifies IP66 across the unit with IP69K at the base. We agree completely on the priority. It is worth confirming the IP69K element specifically, because of what that test actually is.

IP69K is a close range, high pressure, high temperature test drawn from automotive and food processing machinery practice: water at approximately 80 degrees Celsius, delivered at 80 to 100 bar, at a distance of 100 to 150 millimetres. For enclosures over 250 millimetres the test is performed manually across the whole surface rather than on a rotating table, and the Scrappy assemblies are well over that threshold. Meeting it drives real cost into seal design, continuous rather than stitch welding, gland selection and, critically, 100% pressure decay testing on every unit in production, because an ingress claim of that grade cannot be established by sampling.

The question Phase 1 answers is whether the actual cleaning regime at the sites concerned requires it. If kitchen staff are directing a pressure washer at close range then IP69K at the base is the correct specification and we will engineer to it. If the real duty is hose down and spray at normal distances, IP66 across the unit with hygienic design to EN 1672-2 principles delivers the same practical outcome at materially lower cost and lower production test burden. We need the water temperature, the pressure, the chemicals in use and the working distance from the lead caterer to close this. It is recorded as an open item in the PRD itself.

One point regardless of the answer. IP69K is a type test, not a build standard. Passing once in a laboratory does not mean a thousand units hold it in the field. Whichever target is set, the production process needs a leak test that proves it unit by unit, and that is designed into the line rather than added to it. Two qualifications on how that test is defined. A pressure decay test demonstrates leak tightness against an acceptance limit, and the limit has to be correlated to the units that passed the type test rather than asserted on its own. And the pressure equalisation vent has to be plugged for the test to run at all, so the vent's own seal is verified as a separate operation rather than covered by the decay result.

The certification position is more favourable than it first appears, and one element of it is a genuine gap in the current plan. The summary below is the position in four lines, with the reasoning behind each set out underneath.

CE marking is currently recognised as a route for placing this product on the Great Britain market. Under the Product Safety and Metrology etc. (Amendment) Regulations 2024, in force from 1 October 2024, the United Kingdom recognises CE marking indefinitely across 21 product areas, including every regulation that bears on this device: electrical equipment safety, electromagnetic compatibility, radio equipment and RoHS. The same core technical evidence is therefore expected to support both the European Union and Great Britain routes, subject to confirmation of any Great Britain specific documentation, labelling and economic operator requirements, so a separate UKCA conformity assessment campaign is not presently expected. Two qualifications. Because Positive Carbon is established in Ireland, the Great Britain supply route will need to identify the Great Britain importer and confirm the associated labelling and compliance responsibilities. A UK authorised representative may also be appointed for specified functions, but whether one is required or commercially useful depends on the final route to market. And the Product Regulation and Metrology Act 2025, which received Royal Assent on 21 July 2025, is an enabling Act that gives ministers powers to alter the Great Britain product regulatory framework through secondary legislation, so this is a position to be confirmed at the point of placing on the market rather than treated as permanently settled. We will confirm it in Phase 1 rather than rely on it, and we would note that a UKCA mark may still be asked for by an enterprise buyer as a procurement condition rather than a legal one, which is a commercial question worth settling early.

The device is not a legal for trade instrument, and that must be protected. The Non-automatic Weighing Instruments Directive exempts instruments used for internal process control, goods inspection and customer information. Scrappy sits squarely in that exemption, so no conformity assessment, no M mark and no Measuring Instruments Directive obligation applies, and the PRD is correct on this point. The important qualification is that the trigger is the use, not the design. If a caterer ever bills a client from the reading, or a waste contractor charges by weight, or a rebate or penalty is calculated from it, the instrument moves into the regulated category and requires EU type examination plus verification of every unit. Given that the product exists to produce costed waste data, we recommend the intended use is stated explicitly in the customer contract and the user documentation.

Radio certification does not transfer automatically from the module. There is no modular approval concept under the Radio Equipment Directive. Integrating a pre-certified cellular module creates a new radio product, and the module certification does not carry across when antenna design or placement changes. Because the antennas sit inside a stainless head behind a bespoke radome, a reduced radio verification campaign is unavoidable. This is far cheaper than a full campaign but it is not zero, and it needs to be in the budget. The same campaign should carry the radio frequency exposure assessment under EN 62311, which is easily overlooked and applies here because the antennas sit at head height, where an operator stands to read the display.

Cybersecurity is a requirement that is not currently in the plan, and it sits across the scope boundary. Since 1 August 2025, radio equipment placed on the European Union market must meet the cybersecurity requirements under Article 3.3 of the Radio Equipment Directive, assessed against EN 18031, with no transition period. Where the standard is applied in full and no restricted provision is engaged, self assessment is available; where a restricted provision is engaged, a notified body becomes mandatory. Separately, reporting obligations under the Cyber Resilience Act begin on 11 September 2026, with the main obligations applying from 11 December 2027. The Act is a distinct track rather than an extension of EN 18031. Evidence produced for EN 18031 maps usefully onto it, but presumption of conformity under the Act will come only from its own harmonised standards, common specifications or a certification scheme once those are published, so the two should be planned as two pieces of work with shared evidence rather than one.

The division of work follows the scope boundary in section 8. We define the security requirements the hardware and its embedded layer have to meet, establish the assessment route, and carry the assessment and test cost in the certification line in section 5.2. Implementation sits in the on-device software layer, which we have assumed remains with Positive Carbon: secure boot and secure storage, access control, signed update delivery, event logging, and the vulnerability handling process behind them. The Cyber Resilience Act obligations that continue over the product life, including the reporting duty from 11 September 2026, the software bill of materials and the stated support period, fall on the party placing the product on the market, which is Positive Carbon rather than Shield Works. Neither obligation appears in the current documentation, which is why we have raised it here rather than in the exclusions alone.

Two further items sit with Positive Carbon rather than with us, as the party placing the product on the market: WEEE producer registration in the United Kingdom and each European Union market of sale, and packaging producer responsibility obligations. At around 800 to 1,000 units a year these are not trivial, and on the mass estimate in section 2.3 the annual tonnage places the company well above the small producer threshold in the United Kingdom. We flag them so they can be planned rather than met late.

The PRD holds the bill of materials at or under EUR 1,000 per unit, with an indicative figure of EUR 972. Our first observation is a definitional one and it matters more than any single line. That EUR 972 is a parts total, taken at European distributor list prices in single unit quantities and excluding inbound freight. It carries no assembly labour, no factory calibration or functional test, no packaging, no quality control, no factory overhead, no tooling or certification amortisation and no margin. It is a useful engineering discipline. It is not a comparison for a delivered unit price, and setting the two side by side without saying so would mislead in both directions.

Our second observation is more interesting. Reviewed line by line against the Chinese supply base, the bill of materials is materially wrong in both directions, and roughly right in total by coincidence.

The mechanical side is where the cost genuinely sits, and it is worth being specific about why, because it changes where design effort should go. On our cost model for an assembly of this type and quality, welding and the weld dressing that follows it account for roughly a quarter of the cost of the fabricated mechanical assembly. Cutting, forming and the machined joint components account for around a further fifth, and raw material for only about an eighth, with assembly, finishing, inspection and overhead making up the balance. These are proportions of the fabrication, not of the delivered unit. The reason is that grinding a weld flush destroys the surface finish locally, so the whole panel then has to be refinished to blend, and large areas get refinished because of small welds. That is a design problem rather than a purchasing problem, and the levers are to move welds out of sight, to select a finish that blends rather than one that shows a direction, and to reduce weld length. Every metre of weld designed out is worth several times more than a reduction in sheet gauge.

The practical conclusion for Positive Carbon is that the parts budget was never the constraint on this product. There is real headroom against the EUR 1,000 ceiling once the electronics are sourced properly, and our recommendation is that the headroom is reinvested rather than banked: into a designed thermal path, a genuinely certified load cell rather than a lower rated part with an optimistic datasheet, hygienic levelling feet, IP-rated antenna bulkheads, and the enclosure engineering that makes the unit survive. Indicative unit pricing is set out in section 5.

One market condition should be stated plainly because it affects every figure in this document. Memory and storage pricing has moved violently through 2026, with contract prices for both DRAM and NAND rising sharply across successive quarters and no meaningful capacity relief expected before late 2027. Single board computer pricing has followed. Any figure we quote on the compute and storage lines carries a validity window, and the supply agreement will carry a pass through mechanism on those lines rather than a fixed price we would have to pad to survive.

Phase 1 is the first paid engineering engagement in the program. It is the start of development rather than preparation for it. It resolves the questions in section 2, converts the open items in the PRD into documented engineering positions and recommendations, and delivers a costed, stage by stage development plan before any money is committed to tooling or testing.

Phase 1 is delivered by the Shield Works R&D team, led by our Head of R&D, working alongside C2W sourcing, compliance and commercial engineering resources.

Duration: approximately 8 working weeks from the Engineering Kick-off Workshop, which we would schedule within 2 weeks of project confirmation and receipt of the outstanding technical files. The workshop can be held remotely if travel cannot be arranged inside that window, so the program is not gated on diaries, and the teardown then proceeds with our own team in Zhuhai. The pacing items are the supplier engagement in Deliverable 5, which needs real engineering responses from stainless fabrication and load cell suppliers rather than headline quotes, and the thermal and ingress workstream in Deliverable 3.

What Phase 1 establishes, and what it does not. Phase 1 delivers engineering positions, recommendations and the data requirements that the decisions in section 2 depend on. Full specification, detailed design and physical validation sit in Stage 01 and the stages that follow it, and are costed in section 5. That boundary is drawn deliberately: it keeps the fee proportionate to a decision-making phase, and it avoids paying for detailed specification of a geometry that the decisions in this document may change.

Phase 1 is structured around six deliverables, each of which is real engineering or commercial work product.

The fee includes one consolidated client review and one revision of the Development Roadmap. Additional work arising from new requirements, or from source data that turns out to be materially incomplete, is agreed separately before it is started.

One thing worth being explicit about. The purpose of Phase 1 is not to push this product into production. It is to establish the engineering and commercial basis for doing so, and to do it before larger commitments are made. If the work shows that the thermal envelope cannot be met inside the ingress target, or that the accuracy claim cannot be held over a lease at the target cost, we will say so plainly and set out what would need to change. Establishing that for USD 16,500 is a better outcome than establishing it after tooling has been cut.

The scoping call raised the downstream side of the engagement: buffer stock, direct to site shipping, spares and a returns and repair loop. We would rather be precise about what we do and do not operate than describe a capability we do not hold.

What we operate directly. Manufacture, assembly, integration, calibration and test at the Shield Works facility in Zhuhai. Component supply and supplier management. Layered quality control from incoming inspection through in process checks to final and outgoing inspection, with independent C2W pre-shipment inspection and container loading checks. Bonded warehousing and buffer stock holding in China, so finished units can be built to forecast and released against order rather than built to each order. Consolidation and export staging. Export documentation, VAT rebate handling and freight coordination, either through our own freight agents or a forwarder Positive Carbon nominates. Direct shipment from China to a nominated destination, including direct to site, by sea for volume and by air for urgent or early requirements. Spares kitting and supply of field replaceable modules against the same quality process as production units.

What we do not operate. We do not hold warehousing in Ireland, Great Britain or continental Europe, and we will not describe in-region consignment stock as our own capability. For a leased fleet rolled out across three regions, in-region stock and a regional swap and repair depot are the right answer commercially, and returning a unit to China for repair is not economic. What we can do is design for it: field replaceable modules, near-toolless service access that does not break the ingress seal, a defined spares list, and diagnostics that identify the failed module before an engineer travels. We would also be straight about why, because it is a capability question rather than a commercial one. In 21 years we have not operated an end client service interface, and we would not start on your program. Our clients hold that themselves, because they are on the ground, they hold the customer relationship and they have the field and customer service resources that a swap and repair loop actually runs on. Positive Carbon is the right party to own that, and we would rather say so than take it on and do it less well than you would. We are happy to share recommendations from our network on European logistics and repair providers, contracting directly with you.

The forward commercial model. Once a supplier set is confirmed and production begins, you receive a single unit cost per unit that covers goods, quality control, supplier management, packaging and logistics coordination. These are not billed as separate service fees. Tooling and fixtures are quoted separately and are never carried inside the unit price. The Phase 1 Indicative Quotation exists to establish those unit costs with confidence before any production commitment is made.

One point specific to a leased fleet. Because Positive Carbon owns and services every unit rather than selling it, build quality lands directly on your service cost and your unit economics rather than on a consumer warranty. The field failure log in the PRD makes that concrete: defective units reaching a customer, scale calibration drifting in the field, units dropping offline without recovery. Our quality process, the end of line calibration station and the 100% leak test are aimed at that specific exposure rather than at a general standard.

The Phase 1 fee covers all six deliverables in section 4. The activities below sit outside this engagement. Where they are needed they are scoped and costed within the Development Roadmap, or coordinated through approved external partners. Sequencing them this way keeps unnecessary cost out of the early stages and avoids committing to tooling and testing before the fundamental questions are answered.