A £6,000 phased MVP at £2,000 a month, each month gated on evidence, with hardware billed separately at cost. A free validation gate runs first, on the boards you have already bought.
POC is proven. 1 July 2026, Men's Shed: XIAO ESP32-S3 breadboard, three buttons toggling Watch / Shield / Emergency over BLE into the iOS app — the video you were sent.
Payment position (19 August). Against the £5,000 POC, £3,500 has been paid. That leaves £1,500 outstanding — the last of the three agreed monthly payments.
| POC | Amount |
|---|---|
| POC fee agreed | £5,000 |
| Paid to date — £500 + £1,500 + £500 + £1,000 | £3,500 |
| Remaining on POC | £1,500 |
Background reliability is already proven — and it was proven before the MVP was quoted. Range across the house through walls, with the screen locked and the app closed, on both iPhone and Android. That was the single biggest technical unknown in the whole product and it passed. It is stated as proven in the v2.0 proposal of 14 July.
What genuinely remains unproven is narrower and cheaper: battery life on the actual target battery, and accidental-trigger rejection. Then one item that cannot be tested yet at all — there is no housing, so nobody knows whether range holds once the board is enclosed rather than open on a bench. Antenna performance changes inside a case. That test happens the moment the first shell comes off the printer in M2, before the full batch is committed. Real work, but bench work on a known-good architecture — not an existential risk.
The sequencing rule: MVP work does not start until the POC balance is clear. That keeps a clean line between what you already have and what you are buying next.
Each month is a gate with evidence attached. Aaron pays monthly in advance; if a month isn't paid, work pauses and nothing is lost — the project simply sits at the last completed gate. That is deliberately easier on his cashflow than £4,000 lumps, and it matches how he is already paying the POC.
| Gate | Weeks | What it delivers | Evidence at the gate | Days | Fee |
|---|---|---|---|---|---|
| M0 No fee |
0 | C3 vs S3 bench-off, on the boards Aaron has already bought. The C3 carries an on-board (PCB) antenna instead of the external antenna soldered onto the POC rig, so range must be re-proven on the actual part before it is designed in. Everything here runs off USB — neither board has a battery, so nothing is bought: BLE range through walls at distance, C3 vs S3 side by side; full trigger with the phone locked and the app backgrounded; and deep-sleep and active current draw measured with a multimeter inline on the supply rail. No buttons or app work needed: a minimal BLE advertiser sketch on each board plus nRF Connect on the phone measures RSSI and the drop-out point directly — the same tool on both boards, so the comparison is like-for-like. | Two range runs per board, on the same day in the same places: through-walls (the product case) and open-street drop-out (the antenna comparison). Plus the locked-phone alert arriving, and current draw in µA asleep / mA active. A board recommendation. Photo of the rig and a short video, as with the POC gate. | 1 | £0 |
| M1 Design freeze |
1–3 | Unit design frozen on the board M0 recommends. What M0 could not test, because neither board has a battery: battery life on the target 500mAh LiPo and charging behaviour — a cell has to be bought first. Plus accidental-trigger rejection, single button + ring-pull confirmed, and the carrier-board schematic fixed — which now has a hard requirement attached: a battery voltage divider onto an ADC pin, so firmware can read the charge level. Seeed confirm the XIAO assigns no GPIO to the battery, so on the bare module the device cannot know its own battery state. Then — since none exists yet — the housing designed for the first time: slim, pocketable, ring-pull mechanism, printable. | Bench report with pass/fail per test. Frozen schematic + housing CAD, both ready to quote at JLCPCB / JLC3DP. | 4–5 | £2,000 |
| M2 Cash out |
4–7 | First housing printed and range re-tested with the board inside it — a hard checkpoint before the rest of the batch is committed. Then 12 units assembled and bench-tested. iOS app on TestFlight: alert flow, emergency contacts, location, phone-side audio capture, resumable evidence upload. Supabase backend live. | In-housing range result. Units in a box. TestFlight build installable on a tester's phone. An incident record end-to-end in the database. | 4–5 | £2,000 |
| M3 | 8–12 | 4-week field trial with Aaron-recruited testers — no waitlist exists, so these come from friends, family and contacts. Metrics collected: alert success rate, false-alarm rate, battery life, feedback. MVP exit report + costed Production specification. | Field trial report. Go/no-go recommendation. Production quote Aaron can take to a lender. | 3–4 | £2,000 |
| MVP total — ~3 months (M0 carries no fee) | 12–15 | £6,000 | |||
A safety device has to know its own battery level. The XIAO has no GPIO wired to the battery, so out of the box the firmware cannot read the charge state — meaning the unit cannot warn its user that it is nearly flat. Someone arms Shield believing they are protected, on a cell that died yesterday. That is a failure mode the product cannot ship with.
The fix is trivial if it is designed in now and expensive to retrofit once the housing is cut: a two-resistor divider from the battery onto a spare ADC pin, plus a low-battery warning in the app. It is added to the M1 schematic scope for that reason, and it is the first question to put to anyone helping with the board layout.
The soft spot in M3 is people, not hardware. There is no waitlist: the ProofGuard page has not been shown to anyone and no sign-up effort has been made. Twelve testers therefore means friends, family and contacts — which is fine for catching bugs, battery problems and false alarms, and that is what the trial is for. It is not evidence of demand, and it should never be presented as such. Friends are kind; strangers are informative.
Two things follow. Recruiting twelve people takes longer than it sounds, so it starts in M1, not M3. And if demand evidence is wanted for the funding application, that is a separate, cheap exercise — put the page in front of strangers and count sign-ups — which can run in parallel and costs nothing but time.
Why M0 is free, and why it matters more than it looks: the boards are already bought and paid for by Aaron, and every test here runs off USB — so this costs a day and not a penny of hardware. Battery life and charging cannot come forward with it, because neither board has a cell in it; those stay in M1, where a LiPo gets bought. What it buys is the single most useful thing a funding application can carry: a demonstrated working prototype on the actual production-candidate part, not a concept. See the funding page — most Start Up Loan applicants have nothing built at all.
It also de-risks M1 before a penny of fee is charged. If the on-board antenna cannot hold range through a wall, or the C3 draws too much asleep, that is a board decision taken in week zero on parts already in hand — not after the schematic is frozen.
Why M1 still comes first, even though the hard part passed: the design cannot be frozen — and therefore no board or housing can be ordered — until battery life, trigger rejection and in-housing range are measured on the target parts. It is a short, cheap gate with zero hardware spend that stops £500 of components being bought against a design that turns out to brown-out on an SD write. The existential risk is gone; the expensive-mistake risk isn't.
Components, PCBs, enclosures and printing are billed as pass-through at cost + 10% handling, with receipts, and paid before anything is ordered. Paul does not float hardware. Keeping this visibly separate from the fee also shows Aaron exactly what is service and what is materials.
| Item | Choice for MVP | Per unit | × 12 units |
|---|---|---|---|
| Board Decide in M0 | XIAO ESP32-S3 plain (~£6–7.50; $7.49 direct from Seeed) or ESP32-C3 SuperMini (diymore 4-pack, ~£3/unit — ordered by Aaron 28 July, in hand). Neither is the Sense variant — dropping the mic/SD/camera board roughly halves the S3 cost and makes it smaller. Measured 26 Aug: the plain S3 is physically SMALLER than the C3 SuperMini, which reverses the earlier assumption that the C3 served the slimness goal. Both in hand; benched against each other in M0. | £3–7.50 | £36–90 |
| Charging C3 only | Needed only if C3 wins — the XIAO charges a cell itself at 100mA over USB-C (Seeed spec), the cheap C3 minis do not. ⚠ Generic TP4056 modules ship set to 1000mA — that is 2C into a 500mAh cell and a genuine fire risk. The PROG resistor must be changed (I = 1200/R, so 10kΩ ≈ 120mA). Cleaner alternative: an MCP73831-based charger, 100mA by default, no modification. | £0.50–1 | £6–12 |
| Carrier PCB | 2-layer, buttons + reed switch + LED + battery, XIAO socketed (JLCPCB) | £3–8 | £36–96 |
| Battery | 500mAh LiPo, integrated protection circuit, pre-fitted leads and genuine JST-PH connector — UK distributor, both to dodge UN38.3 air-freight grief and because the protection circuit is the part that must not be a guess. Board side takes a mating JST-PH pigtail so cells are keyed and swappable. | £6 | £72 |
| Ring-pull trigger | Magnetic reed switch + magnet on lanyard | £1–2 | £12–24 |
| Button | Single tactile switch (press patterns select state) | £0.50 | £6 |
| Enclosure | 3D-printed nylon, designed from scratch in M1 — slim is the priority, benchmarked against the tiny China sample. No injection mould at this quantity. | £5–10 | £60–120 |
| Lanyard, screws, sundries | — | £1 | £12 |
| Spares & failures | +25% parts allowance | — | £75–115 |
| Parts sub-total — 12 units + spares | £405–575 | ||
| Fabrication & assembly — not yet quoted (see below) | TBC | ||
At ~£3 against £8–10 the C3 saves roughly £60–84 across twelve units and is the smaller board. Worth taking seriously. Three things have to be measured before the batch is committed to it:
Recommendation: the 4-pack is already bought and in hand — Aaron ordered it on 28 July — so this costs nothing to settle. Run C3 against S3 as a straight bench comparison in M0 on range and current draw; charging and battery life follow in M1, once a cell is bought. Pick on data, not price. Production is unaffected either way — that's a custom PCB with a pre-certified module, C3 or S3.
Working default: the S3 for the whole MVP. On the evidence so far it leads on everything except unit price — range through walls and locked-phone triggering are already proven on it, it charges a cell onboard at the correct 100mA with no extra parts, it has a U.FL socket for an external antenna, and Seeed document a 14µA deep sleep. The C3’s advantages have now largely evaporated: on measured dimensions the plain S3 is the smaller board, and with the Sense variant dropped the real price gap is about £3.50–4.50 a unit, not the £5–7 first assumed. What remains of the C3 case is also eroded by the TP4056 and its wiring.
Across twelve units that price gap is about £45 — noise, not a decision. At production volume it is thousands. So M0 is not really choosing the MVP board; it is buying C3 data cheaply now to inform the production board later, on parts already in hand. The MVP runs on the S3.
And the picture genuinely reverses at module level, which is why the C3 data still matters: the bare ESP32-C3-MINI-1 is 13.2 × 16.6 × 2.4mm against the ESP32-S3-MINI-1 at 15.4 × 20.5 × 2.4mm. The C3 module is meaningfully smaller and cheaper — it is only the dev board built around it that is not.
M0 needs no battery on either board: runtime is predicted from current draw. Powering the C3 from a cell is not a wiring job but a small power-supply design, and one question decides it — which regulator diymore fitted. A LiPo runs 4.2V down to ~3.0V while the ESP32-C3 needs a stable 3.3V. If it is the common AMS1117, its ~1.1V dropout cannot hold 3.3V from a 3.7V cell and it will brown out as the battery sags; a low-dropout part would cope. That marking is readable off the board with a magnifier and is worth checking before any C3 battery work is contemplated. It is also the clearest illustration of why the C3's £3 is not the whole price.
Battery safety — the controls that actually matter. Nearly all of the risk is designed out by what gets bought, not by care at the bench. Buy protected cells with leads already fitted, and never apply heat to a cell tab — tabs are spot-welded, not soldered. Fit a keyed JST-PH connector so polarity cannot be reversed, and meter the cell first: connector polarity is not standardised between suppliers. Solder with no cell connected, bring the rig up current-limited, and if a connector is ever cut off, cut one wire at a time — both at once shorts the cell across the blades.
Charge on a non-combustible surface, never unattended. A LiPo bag is fire-resistant containment, not fireproof — useful, and never sealed shut. A 500mAh cell holds about 1.9Wh, roughly a tenth of a phone: real hazard, manageable scale.
Sheds carry retired engineers, and schematic review, a simple two-layer carrier layout and a DFM check are well within that. Taken up, it could remove the design line from the funding ask entirely. Three caveats: RF is not general electronics knowledge, so follow the module vendor's reference layout exactly for antenna keep-out rather than improvising; agree IP ownership in writing before work starts, however informally, because an unassigned contribution becomes a problem the moment the business is transferred or seeks lending; and keep volunteer help to the MVP — the production board for a safety product belongs with someone paid and accountable.
Parts estimates pending real quotes from JLCPCB / JLC3DP. Lead times run 2–3 weeks on PCBA and printing, which is why ordering starts the moment M1 passes rather than after the software is finished.
The POC was a breadboard on the bench powered over USB. That cannot become a field-trial unit — it will not fit a pocketable housing and it has no battery. Turning a proven circuit into twelve things a tester can carry needs three jobs that are not in the parts table and are not in the £6,000 fee:
| Job | Who does it | Cost |
|---|---|---|
| Circuit board design | An electronics engineer — schematic capture and PCB layout. Not a software job. BLE layout in particular has antenna keep-out and ground-plane rules that are easy to get wrong and expensive to discover late. | Not quoted |
| PCB assembly (PCBA) | China, almost certainly JLCPCB — their Economic tier is built for exactly this size of run. Setup, stencil and per-joint charges apply on top of the parts above. | Not quoted |
| Final assembly & test (box build) | Board, battery, switch and lanyard into the printed housing, then tested. By hand. | Not quoted |
Twelve units is below the threshold where a Chinese contract manufacturer will build finished products for you. Turnkey box-build is quoted against volumes far larger than this; at twelve, the setup cost per unit is absurd and most will simply decline. That is an assessment from how the industry prices, not a quote we have been given — but it is the right assumption to plan against. China does the boards; the units get finished by hand here.
The unknown can be removed this week, for nothing. Quotes are free: an EE design quote for the carrier board, a JLCPCB PCBA quote once a schematic exists, and a JLC3DP print quote against the housing CAD. Until those three numbers are in, any fabrication figure on this page would be a guess. Nothing gets committed and no batch gets ordered until they are.
Where it lands matters for funding, not just for cost: PCB design and tooling are exactly the kind of line a Start Up Loan is meant to cover — see the funding page. It belongs in that ask, not buried in the MVP fee.
Nothing below is committed and nothing below should be spent until the MVP field data justifies it. It is included so Aaron can see the real shape of the road and take a costed plan to a lender.
| Production item | Indicative | Note |
|---|---|---|
| Custom PCB (ESP32-S3-MINI, pre-certified module) | Fee TBD | Unit cost drops to ~£5–8 at 1,000 |
| Injection mould tooling | £2,000–3,000 | Only economic at 500+ units |
| UKCA/CE + EMC + Radio Equipment Regs | £5,000–15,000 | Depends how much the pre-certified module covers. 8–12 weeks. |
| Battery transport compliance (UN38.3) | TBD | Needed to ship product |
| Managed hosting migration | TBD | Non-negotiable: a paid safety service cannot run on a consultant's VPS |
| Paul's Production fee | £15,000–25,000 | Scoped properly at MVP exit, not now |
| First run, 500–1,000 units | TBD | ~4–6 months from MVP exit |
This is the honest reason the MVP exit report matters commercially: it is the document that turns "my mate's building me an app" into something a lender or an investor can price. The government-backed loan scheme Aaron asked about on 26 July will want exactly this.
Don't plan around funding arriving. Grants and lending are slow — months, not weeks — so no part of the MVP should be sequenced on money that hasn't landed. Treat funding as upside: if a loan or grant comes through, it goes towards the production version, the fabrication costs and the launch push — not towards accelerating the MVP. The MVP is deliberately scoped to be affordable month by month without it.
Still open: whether Paul's professional indemnity insurance covers safety-adjacent product work. Worth a call to the broker before M2, not after.
| # | Decision | Owner | By when |
|---|---|---|---|
| 1 | ESP32-C3 boards — ordered and delivered. M0 bench-offs them against the S3 (range, locked phone, current draw) at no fee | Paul | Now |
| 1b | Order a 500mAh LiPo — neither board has a battery, so battery life and charging cannot be tested until one is bought | Paul | Before M1 |
| 1c | Route A or B for the twelve units — hand-wired protoboard, or a designed carrier PCB. Decides whether an electronics engineer is needed at all | Paul + Aaron | M1 |
| 1d | Get the quotes. Only the two design jobs are quotable now (carrier PCB, enclosure CAD) — PCBA and printing cannot be quoted until files exist, then both price instantly online. Free either way, and they close the only open cost on the project | Paul | M1 |
| 1e | Rebuild the firmware for the C3 target — S3 is Xtensa, C3 is RISC-V, so the POC build does not copy across. A minimal BLE advertiser is enough for M0; the full port waits for the board decision | Paul | M0 |
| 2c | Get the China sample measured — its dimensions are the slimness target for the housing CAD | Paul | M1 |
| 3 | One button + ring-pull confirmed (POC used three) | Paul + Aaron | M1 |
| 4 | Batch size 12 vs 15 vs 20 | Paul + Aaron | M1 exit |
| 5 | Runtime target — suggest ≥30 days standby / ≥4h active | Paul | M1 |
| 6 | PI insurance cover for safety-adjacent work | Paul → broker | Before M2 |
| 7 | Find 12 testers. There is no waitlist and no signups — realistically friends, family and contacts. Confirm Aaron owns this | Aaron | Before M3 |
| 8 | JLCPCB / JLC3DP quotes (needs schematic + CAD uploaded — ask Paul first) | Paul | M1 week 3 |