The fastest way to blow your hardware launch budget is to treat manufacturing as a single step: design done, send Gerbers, get units back. Real production runs through a sequence of validation gates — EVT, DVT, PVT — and each one exists because skipping it costs an order of magnitude more later. Founders who learn this the hard way usually find out during PVT that a connector they picked is now end-of-life, after they have already paid for tooling.
Here is what each gate actually checks, and where first-timers lose money.
EVT — Engineering Validation Test
EVT answers one question: does the design work at all? You build a small batch — typically 5 to 20 units, often hand-assembled — to prove the architecture.
What you test:
- Core functionality: does the firmware boot, do the radios link, does power sequencing behave.
- Major design assumptions: thermal headroom, current draw, antenna performance on the real board.
- Component choices under real conditions, not just on the bench breadboard.
EVT is where you expect to find problems. Finding three bugs here is normal and cheap. Finding them in PVT is a disaster.
DVT — Design Validation Test
DVT answers: does the design meet spec, repeatably, and can it be built? This batch (50–200 units) is made closer to the real production process. The board, enclosure, and assembly should be near-final.
What you test:
- Full spec compliance across units, not a golden sample.
- Regulatory pre-scans: FCC/CE, drop, vibration, ESD. Catch a radiated-emissions failure here, not after you have shipped.
- DFM and DFA: is anything painful to assemble or test at volume? This is the gate where a good factory or sourcing partner earns its fee by pushing changes that cut per-unit assembly time.
Skipping or rushing DFM is the single most common expensive mistake. A footprint that a human can hand-solder but a pick-and-place line struggles with turns into yield loss on every batch forever.
PVT — Production Validation Test
PVT answers: can the factory build it at volume, at the target yield, on the real line? This is a pilot production run — hundreds to low thousands — using mass-production tooling, jigs, and operators.
What you test:
- Yield and cycle time on the actual line.
- The end-of-line test jig and test logs.
- Packaging, labeling, and the full assembly SOP.
PVT units can usually be sold. By now the design must be frozen — changes here mean re-cutting tooling.
A realistic timeline
For a moderately complex IoT device with custom tooling, a sane schedule looks like:
- EVT: 2–4 weeks (plus board fab/lead time)
- DVT: 3–5 weeks (includes pre-compliance and tooling)
- PVT: 3–4 weeks
That lands a first production order in roughly 8 to 12 weeks if nothing surprises you — which is the window I'd plan around for a first run, with a buffer because something always slips.
Where founders skip steps and pay
The pattern is always the same: collapse EVT and DVT into one batch to save a month, freeze the design too early, then discover at PVT that the enclosure traps heat or a part is unsourceable. Re-tooling a mold is a four-figure cost and weeks of delay — far more than the gate you skipped would have taken.
The actionable move: write down a pass/fail exit criteria list for each gate before you build that batch. "EVT passes when the device runs the full self-test for 24 hours across 5 units" is a real gate. "Looks good" is not.
If you are running this solo without a team in Shenzhen, the DFM review and gate management is exactly where an engineering-led partner helps. Someone like China Sourcing Agents — where the person reading your BOM is a hardware engineer — can run DFM before EVT and keep the factory honest through each gate, which is what actually compresses that 8–12 week window instead of letting it sprawl.
Treat EVT/DVT/PVT as forcing functions, not bureaucracy. Each one is a chance to fail cheaply.
Top comments (0)