Most explanations of EVT, DVT and PVT are a glossary entry with a table of unit counts attached, and the unit counts in different explanations disagree by a factor of twenty. Nobody is lying. The terms were coined inside very large consumer electronics programmes and then borrowed by everyone else, and each source describes the scale it knows.
We have written about why the three gates exist and what it costs to overlap them in Someone Has to Own the Calendar, and about the requirements document as the cheapest prototype you will ever build. This piece is the operational one. What does each build have to prove, what gets frozen on the way out, where do certification, tooling and firmware actually sit, and how does a team of two compress the sequence without lying to itself?
Why nobody agrees on the numbers
Here is what four sources say about build sizes, in units.
| Source | EVT | DVT | PVT |
|---|---|---|---|
| Bolt (hardware VC, 2015) | 20 to 50 | 50 to 200 | 500+, or 5 to 10% of first run |
| Instrumental (2016, updated 2024) | 100 to 1,000 | 300 to 2,000 | 1,000 to 20,000 |
| Jaycon (design firm, 2022) | 20 to 1,000 | 50 to 2,000 | 1,000+ |
| Venture Mfg (manufacturer, 2026) | 5 to 20 | 50 to 200 | 100 to 1,000 |
Instrumental sells inspection software to large electronics brands, and its figures read like a phone programme. Bolt invested in startups. Neither is wrong about its own world. The useful conclusion is that unit count is an output, not an input.
Work it out from the test plan: the units drop testing will destroy, the environmental chamber set, the samples the certification lab keeps, the reliability set, field testers, and two golden samples. Add a margin for fallout. If the plan needs 60 units and you build 30, you have quietly decided which tests will not happen. Write that decision down.
Instrumental’s rule for the commonest self-deception: if you are running side configurations of 20 units in DVT, you should call it EVT2.
EVT: does the design work when built the way it will be built?
EVT is the first build assembled by the contract manufacturer rather than by your engineers, and it is the first time the board house brings your PCBs up in a production setting. Bolt describes the usual state of affairs: parts come from first-shot tooling, so expect flash, wrong colours, missing texture and poor fit, with an engineer stationed at the factory for the build.
Sources disagree on what counts as EVT material. Instrumental says 3D-printed parts do not qualify. Fictiv and Formlabs allow printed, CNC-machined and vacuum-cast parts. Our working rule: a printed part is acceptable where its material is irrelevant to the test, and never near an antenna, a seal, a snap fit or a thermal path.
Instrumental’s build guide gives the rhythm: about two weeks of build, two to three of test, one to two of iteration. Formlabs puts the whole EVT stage at three to six months. Bolt treats a second EVT as common.
What goes wrong. Depending on complexity, up to 40 per cent of EVT units may fail and need analysis, according to Instrumental, with antenna tolerances, glue processes, hand soldering and environmental seals as the usual suspects. A mismatch between a supplier’s outgoing quality report and the factory’s incoming inspection costs four days to a week on its own.
What EVT must prove. One production-intent configuration that meets the functional, performance and reliability requirements, with every functional test station present and collecting data. Predictable Designs adds a pre-compliance EMC scan at this stage, “a half-day at a local lab”, which is the cheapest certification insurance you can buy.
What freezes on exit. The electrical architecture, the main silicon and radio module, the PCB outline and stackup, and the mechanical envelope including parting lines. These freeze because the production tool is about to be ordered against them.
What it costs. Formlabs puts a refined rapid prototype with vacuum-cast overmoulding at $500 to $1,000 a unit, so 20 to 50 units run $10,000 to $50,000 (about €9,000 to €43,000) before engineering time. Renting a pre-compliance chamber costs $500 to $2,000 (about €430 to €1,700) a day, per MarkReady. For how the gates add up into a funding plan, see our piece on funding the whole phase.
DVT: can it be built repeatedly, from the tools you will ship from?
DVT is where the hard tools arrive. Every made part comes off a production tool or a mass-production-capable process, built on a production line, and the point of the build is to qualify those tools and measure true yield. Bolt notes it is usually the first time you see production cosmetics: in-mould decoration, pad printing, final colours. Instrumental puts the mass-production hard tool at about 12 weeks to make, which is why it gets ordered at the end of EVT.
Formlabs lists the battery: environmental chamber, thermal cycling, vibration, ESD, ageing, cosmetic, wear and drop testing, plus user testing in context. Jaycon notes that consumer electronics companies destroy millions of test units a year this way, which is why DVT quantities are larger.
What goes wrong. Instrumental’s description of a typical DVT is unsentimental: high functional fallout, cosmetic yield at zero because suppliers ship scratched parts, a key process such as gluing still out of control, nightly calls with vendors and a queue of designed experiments. Sofeast, a Shenzhen quality firm, argues that the costliest DVT failures are omissions: reliability tests never run, misuse cases never considered, materials new to the product never qualified. It points to the Philips respiratory device recall, where degrading sound-proofing foam led to millions of units recalled at a cost just short of $1 billion before litigation, as the kind of problem DVT exists to catch.
What DVT must prove. Parts from the production tools meet every requirement, including cosmetic and environmental ones; any yield problem has a corrective action you have high confidence in; and regulatory testing has been done on units that represent production.
What freezes on exit. The design. The BOM, including the approved alternates you will be glad of when a part goes on allocation. The factory test firmware. From here, the tools may be tuned but not redesigned.
What it costs. Tooling dominates. European figures from Formlabs put an aluminium tool at €2,000 to €5,000 (about $2,300 to $5,800) and a steel tool at €5,000 to €100,000 (about $5,800 to $115,000), with complex multi-cavity steel above €100,000. A US quote example from TEAM Rapid shows the trade-off: $12,000 for a single-cavity tool at $2.40 a part against $32,000 for four cavities at $0.95. Chinese tooling can cost upwards of 40 per cent less than a typical European or US quote, per Sofeast. Then add certification, which we price in detail in the certification map.
PVT: can the factory do it at rate, with the real line?
PVT tests the process, not the product. In Bolt’s account the tooling does not change and the product changes only if something goes badly wrong. You measure yield, volume, cycle time and rework time, and you build the apparatus of production: assembly jigs, programming fixtures, traceability codes linking component serials to unit and box, and final QA in which every unit boots, connects and reflashes production firmware automatically. Operators train on the line they will run.
Formlabs adds the paperwork that founders forget: first article inspection, the packaging and the user manual. Fictiv adds the test that founders forget: drop and vibration on the gift box and the master carton, to ISTA-3A, because a product that survives a drop and a carton that does not is still a return.
What goes wrong. Instrumental says an outstanding issue is almost always present at the start of PVT, a vendor’s yield is often far below what was promised and gates the whole line, and cosmetic yield can start at zero and take weeks to recover. Teams often colour-code PVT builds red, yellow and green as the line matures. Sofeast’s list of PVT failures is mostly about setup: work instructions, line balancing, mistake-proofing and in-line inspections nobody defined.
What PVT must prove. Mass-production yield at mass-production speed on at least one line.
What freezes on exit. Everything, including the packaging artwork and the labels, which is where the regulatory differences below start to matter.
What it costs. PVT units are meant to be sold, so the build itself is mostly your first-run BOM and assembly cost for 5 to 10 per cent of the order. The real cost line is risk: Formlabs puts a tooling change during production at up to $50,000 (about €43,000), plus weeks or months of delay.
The gate checklist
| Stage | Units (startup scale) | Frozen on exit | Exit criteria | Typical cost range |
|---|---|---|---|---|
| EVT | 20 to 50 | Architecture, silicon and radio, PCB outline and stackup, mechanical envelope | One production-intent configuration meets functional, performance and reliability requirements; all test stations collecting data; pre-compliance scan clean | $10,000 to $50,000 (€9,000 to €43,000) in units; $500 to $2,000 (€430 to €1,700) per pre-compliance day |
| DVT | 50 to 200 | Design, BOM with alternates, factory test firmware | Hard-tool parts meet all requirements including cosmetic and environmental; corrective actions for yield issues; formal EMC, radio and safety reports on production-equivalent units | €5,000 to €100,000 ($5,800 to $115,000) per steel tool, or €2,000 to €5,000 ($2,300 to $5,800) aluminium; certification on top |
| PVT | 5 to 10% of first run | Everything, including packaging and labels | Production yield at production speed on one line; fixtures and test stations validated; transport testing passed; sellable units | First-run BOM and assembly for the PVT quantity; a production tool change can reach $50,000 (€43,000) |
Unit counts follow Bolt, MacroFab and Formlabs; the ranges by source are in the first table. Dollar and euro figures are each quoted in the source’s currency and converted at the euro reference rate of 1.1537 dollars for 16 September 2026, rounded.
Where certification actually lands
Bolt, Formlabs and Jaycon all place regulatory testing in DVT. Predictable Designs, in a January 2026 piece, places formal certification at pilot production, with pre-compliance at EVT and DVT, on the grounds that certifying units which then change wastes the result.
The two positions reconcile once you notice that the label on the build is not the point. What matters is that the tested unit is production-equivalent: final PCB revision, final antenna and matching, final RF power tables and security features in the firmware, enclosure parts off the hard tool. In a well-run programme those units exist in late DVT, and the certificate is a DVT exit criterion. If your DVT still changes the board, your certification is not done, whatever the schedule says.
The arithmetic says to start early. MarkReady reports that roughly half of consumer electronics products fail EMC at an accredited lab on the first attempt. A retest without a board re-spin costs $2,000 to $5,000 (about €1,700 to €4,300) and two to four weeks; after a re-spin it costs $5,000 to $15,000 (about €4,300 to €13,000) and six to twelve weeks. That second path has to fit before PVT, or PVT builds boards that are about to change. Predictable Designs gives the clocks to plan against: one to two weeks of FCC testing for an unintentional radiator if it passes first time, two to three months or more for a custom intentional radiator, two to eight weeks for CE, and six to twelve weeks for safety certification once reports and reviews are included. Lab backlogs come on top. Using a pre-certified radio module, they note, shortens all of it.
The US and EU differences that move a gate
Most of this is physics, and physics does not care which market you ship to. Three differences change when things have to happen.
FCC pre-sale rules shape the end of PVT. Under 47 CFR 2.803 you may advertise a device before authorisation and take conditional orders, provided you disclose prominently that delivery depends on authorisation, keep ownership until the sale completes and keep records for 60 months. For a device that needs certification, you may hand stock to others for pre-sale purposes such as packaging or moving it to distributors only once accredited lab testing is complete and the application has been filed with a certification body, and each unit or its packaging must carry a removable label saying it cannot be delivered, displayed or operated until certified. No unit reaches an end user before the grant. For devices under Supplier’s Declaration of Conformity, nothing may be physically transferred until the process is complete. In practice, PVT output can move into a US warehouse before the grant if the test reports exist.
An NRTL listing is tied to a factory. No federal law requires a UL or ETL mark on a consumer gadget sold direct, but US retail and insurers expect one. UL’s follow-up services documentation says the initial production inspection takes place during the first production run, products should not ship with the mark until it is complete, and adding a factory triggers a new one. For Type R labels, unannounced follow-up visits usually happen at least four times a year. The PVT factory is therefore the listed factory, and switching manufacturer after PVT restarts that part of the clock. If you are still choosing, our piece on choosing a contract manufacturer covers what to ask before you sign.
EU labelling lands on PVT artwork. The CE mark is self-declared, so the EU has no grant to wait for, but it has addresses to print. Under the General Product Safety Regulation, applicable since 13 December 2024, the product must carry a batch or serial number, and the manufacturer’s name, registered trade name or trademark, and postal and electronic address must appear on the product or, where size does not allow, on the packaging or an accompanying document. The responsible economic operator’s name and contact details, including a postal address, must appear on the product, its packaging, the parcel or an accompanying document. A US or Asian company selling into the EU without an entity there needs that operator appointed before the packaging artwork is released for print, which in a normal programme means before PVT.
Firmware does not follow the hardware gates
Firmware milestones line up with the gates only loosely. Memfault’s guidance is the clearest we have read: keep a purpose-built manufacturing firmware that rarely changes and is separate from the application, and load the application at the last test station. That lets the application keep moving after DVT, once the factory processes are locked. Plan a day-0 update that has to be ready when devices reach customers, not when they leave the line, which decouples the ramp from the software release and adds more than four weeks to the software schedule. Treat over-the-air updates as insurance, with cohorts, staged rollouts and mandatory intermediate versions.
Mapped onto the gates: EVT needs bring-up firmware and a first factory test image good enough to run every test station. DVT needs the factory firmware frozen, along with anything a certification lab will test: RF parameters, and the security behaviour that the RED cybersecurity standards now examine. Fictiv’s advice to keep firmware and electrical engineers on standby during DVT builds is practical, because programming and built-in functional tests are what most often halt a line. PVT needs the production image programmed by fixture and verified on every unit. The application can trail, provided the day-0 update and the update path have been proven on PVT units.
Compressing the sequence when there are two of you
A two-person company cannot run Instrumental’s programme. It can compress honestly.
Cut quantities, not gates. Fewer deliberate tests beat a large build with no plan.
Buy certainty off the shelf. A pre-certified radio module or a certified battery pack removes the two biggest certification timeline risks Predictable Designs identifies: custom radios and custom batteries.
Use bridge tooling as the production tool. Protolabs quotes aluminium tools from around $1,500, delivered in seven days or less, good for roughly 10,000 shots and often more. Its German site guarantees prototype tools for at least 2,000 shots, with finished parts in one to two weeks. If your first year’s volume fits inside that, your DVT tool is your production tool. That is not skipping DVT. It is making DVT cheap enough to do properly.
Make PVT your first sellable run. Bolt already defines PVT units as sellable. Keep the exit criteria and add a stop rule written in advance: below a stated yield, the line stops and the units do not ship.
What does not count as compression:
- Building with EVT parts while claiming DVT exit criteria. Instrumental calls this XVT and says that if it stood for anything, it would be “No Validation Test”.
- Ordering steel before EVT reliability results. Formlabs notes that hard tooling at the end of EVT can skip DVT and calls it risky and generally inadvisable; the calendar piece covers what it costs when the gamble fails.
- Certifying units that will change. The certificate describes the unit you tested.
- Passing a gate on waivers. Write the exit criteria before the build, in the requirements document, and sign them. If you change the criteria after you see the results, the build has a different name: EVT2.
Salitronic suggests a 20 to 30 per cent contingency on top of the estimate for the whole project. For a team that is compressing, that is the minimum.
Order of operations
- Write exit criteria for all three gates into the requirements document before EVT is ordered.
- Size each build from its test plan, including destroyed units, lab samples and two golden samples.
- Choose a pre-certified radio module, and check its grantee against the FCC Covered List.
- Run a pre-compliance scan on EVT units and again on early DVT units.
- Freeze architecture, silicon, PCB outline and mechanical envelope at EVT exit, then order production or bridge tooling.
- Freeze factory firmware, RF parameters and security behaviour before formal certification testing.
- Book certification labs early enough that a six to twelve week re-spin retest still finishes before PVT.
- Appoint the EU responsible economic operator and settle US NRTL scope before packaging artwork is released.
- Run PVT in the factory that will hold the NRTL listing, with fixtures, traceability and transport testing in place.
- Prove the day-0 update and the update path on PVT units before the first one ships.
None of this is exotic. It is the same three questions in the same order, asked by people who decided in advance what a good answer would look like.
Sources
- The Illustrated Guide to Product Development (Part 4: Validation) — Bolt
- Hardware engineers speak in code: EVT, DVT, PVT decoded — Instrumental
- Why hardware products get delayed: EVT, DVT, PVT Decoded, Part 2 — Instrumental
- EVT DVT PVT: A guide to engineering builds (whitepaper) — Instrumental
- How to Avoid Design Mistakes: EVT, DVT, and PVT — Jaycon
- EVT vs. DVT vs. PVT: Understanding PCB Development Cycles — Venture Mfg
- Validation Testing in Product Development: From POC to EVT, DVT, PVT, and Mass Production — Formlabs
- Wie schätzt man die Spritzgusskosten ein? — Formlabs
- Pre-Production Hardware Testing Methods — Fictiv
- DVT in PCBA Product Development — MacroFab
- How Long Does It Take to Certify a New Product — Predictable Designs
- EMC Pre-Compliance Testing: Save Money Before the Lab — MarkReady
- Aluminum Mold Tooling for Injection Molding — Protolabs
- Die Vorteile vom Kleinserien-Spritzguss — Protolabs
- U.S. Injection Mold Tooling Quotes: Compare Real Costs — TEAM Rapid
- What Determines the Cost of a Plastic Injection Mold in China? — Sofeast
- DVT and PVT mistakes that lead to product recalls — Sofeast
- De-Risk Product Launches with Device Reliability — Memfault
- 47 CFR 2.803, Marketing of radio frequency devices prior to equipment authorization — Cornell Legal Information Institute
- FUStart: Follow-Up Services presentation, June 2023 — UL Solutions
- EU Regulation 2023/988 on general product safety: detailed guidance — GOV.UK
- How Much Does Electronic Product Development Cost? — Salitronic
- Euro foreign exchange reference rates as of 16 September 2026 — Banca d’Italia