A turnkey quote isn't a price. It's a list of assumptions someone made about your BOM.

A contract manufacturer, RAYPCB, published a field report this week about an industrial-controls client that outsourced a 3,000-unit build on a turnkey basis. The client's BOM specified capacitance, voltage, and package for a line of capacitors. It never specified dielectric.

The X7R part sat on a 22-week lead time. A Y5V part in the same package was in stock at half the price. It got bought.

The build passed AOI. It passed functional test at 25°C on the bench. Then it started failing in the field — intermittent brownout resets in cabinets running at 55°C ambient. The rework bill was $61,000, and the customer relationship didn't survive it.

Here is the part that should unsettle you: "Nobody had done anything wrong, technically." The Y5V met every parameter the BOM specified. Purchasing made a rational call on lead time and price. Every local decision was defensible. The failure was not a mistake — it was an unspecified parameter meeting a rational substitution. Nobody negotiated the substitution clause, and the system had no place where that negotiation could have happened.

Four incidents, one mechanism

This is not a one-off. Four independent substitution stories have surfaced in hardware circles over the past month, and they run the same invariant mechanism every time:

Incident	The substitution	The cost
EDN "Tales from the Cube" (Larry K. Baxter)	Purchasing swapped the specified 470-μH cordwood inductor for a smaller 470-μH part to save cents; DC current saturated the core, turning it into a 200-μH part	60% rejection rate for 3+ months, 2,000 dead units. "Purchasing received a reward for saving a few cents... but every failed terminal cost many dollars in stagnant inventory, test-technician time..."
Startup IoT gateway (Medium, Jul 2026)	Inherited BOM line read "H1102NL or equivalent" — no manufacturer, no alternate part number, no notes. A generic-trader part with off-spec winding passed visual inspection	200 boards assembled, ~30% failed link training. "Or equivalent. Those two words cost us three weeks."
RAYPCB turnkey report (Sep 2026)	X7R → Y5V, same capacitance, package, half the price	3,000 units, $61,000 rework, lost customer
Repair-lab case (flagged: single-source account)	Factory "drop-in replacement" buck converter — pins matched, internal feedback reference differed	50 bricked boards, recovered one feedback resistor at a time by hand

The mechanism never varies: underspecified BOM line → rational-local downstream substitution → parametric difference invisible at the substitution point → failure that passes every test except the field. Purchasing, sourcing, contract manufacturing — the substitution finds whichever discipline owns the cost lever and hides in whichever parameter the BOM didn't pin down.

The workaround is discipline. Discipline doesn't scale.

The standard advice — and it's good advice — is BOM-authoring discipline: specify every parameter, maintain an AVL with reviewed alternates, mark no-substitution lines. The H1102NL author rewrote his company's entire BOM practice afterward. RAYPCB's prescription is the same.

But notice what that prescription actually is: a human acting as the merge gate, standing between the BOM and the substitution, holding every parameter in their head. It's the same pattern as every other workaround in this industry — hand-built review replacing infrastructure that doesn't exist. And it fails exactly where it's needed most: turnkey builds, inherited BOMs, and purchasing organizations whose incentives reward the substitution.

An unspecified alternate is a change-impact visibility problem. "Or equivalent" is a string that should trigger a review before it ever becomes a reel on a pick-and-place.

Make the substitution visible

Every substitution in this cluster was a change made without an engineering review — and every one of them passed the checkpoints that existed. The BOM was the last document where the decision could have been cheap. By the time AOI, bench test, or end-of-line test sees it, the expensive part is already built.

ProductFlo's position: a substitution is a proposed change to the product, and changes go through the merge gate. Flag untested alternates. Flag CM substitutions without engineering review. Agents propose the diff; a deterministic layer scores the impact against the engineering constraints; a human approves. The $61,000 wasn't a manufacturing cost. It was the price of a change nobody reviewed.

---

Sources: RAYPCB turnkey report (Medium, Sep 2026; supplier-authored — incident detail as reported); EDN "Silo slip-up" (Larry K. Baxter, "Tales from the Cube"); H1102NL sourcing story (Medium, Jul 20, 2026; pseudonymous author, first-person account); buck-converter substitution case (single-source account — cited as pattern confirmation, not primary evidence).