DEV Community

Gavren
Gavren

Posted on

The Second Sample Passed Every Check and Would Not Fit

The second supplier's sample passed inspection on every dimension. Fitted to a frame built by the first supplier, the axle would not enter the bore — a fraction of a millimetre out. Neither drawing was wrong. Neither inspection report was wrong. What was missing was any requirement that the two tolerance bands had to mesh.
Why does dual sourcing fail at the interface?
Because verification in a dual-source setup is naturally one-sided. Each supplier works to its own drawing, inspects with its own gauge, and submits a report that comes back green. Nobody fits the two parts together, because fitting them together is not inside either supplier's scope of delivery.
Interface deviations also add up. Supplier A machines the axle on the upper half of its band; supplier B reams the bore on the lower half of its band; both are conforming, and the assembly is an interference fit. Each report is defensible on its own. Together they describe a joint that cannot be closed.
Interfaces fall into three kinds: shaft-and-bore fits, clips and inserts, and threads with fasteners. The first causes the most trouble because its deviation decides whether the part enters at all. Clips fail through elasticity and dimensional drift. Threads fail through hole pattern offset, where every thread is good and the pattern is not.
There is another layer underneath: an interface moves during transport. Packed and shipped separately, the bore and the shaft can both lose roundness or straightness before they meet at the assembly plant, so two conforming parts still produce a poor first-pass yield.
The problem usually surfaces after volume ramps, because at low volume the two suppliers share a batch of hand fitting and the deviation is quietly removed. When the volume rises and the hand fitting stops, the deviation reappears with no process change to explain it.
What is dual sourcing actually buying?
The table below is the only one here: the element on the left, the single-source risk in the middle, and what dual sourcing has to pay for on the right.
| Element | Single-source risk | Extra cost of dual sourcing |
|---|---|---|
| Drawing and tolerance | No comparison, so deviation goes unchecked | Interface dimensions on both drawings must come from one source |
| Tooling and moulds | One mould, one point of failure | A second set, plus clear ownership of it |
| Gauges | Each side measures its own, results not comparable | One shared gauge set, or periodic cross-checking |
| Sealed sample | A single reference | One sample per supplier, and the pair must assemble |
| First-article approval | One signature is enough | Assembly verification written into the approval |
| Change notification | A change affects one supplier only | Any change at either one is notified to the other |
The sealed sample line is often treated as a formality. What works is one sample per supplier, then the two samples assembled once before sealing; only then does a seal mean anything. A seal without an assembly leaves each side holding a piece of evidence and nobody holding proof that the two parts were ever compatible in the first place.
The mould line deserves its own sentence. If the supplier pays for the second mould, who owns it when the relationship ends is frequently left unwritten. The executable version has the buyer funding and holding the tool, or agreeing a buy-back at residual value when the relationship ends.
Every line on the right says the same thing. What dual sourcing buys is not a second supplier; it is a second set of tooling and gauges that agrees with the first set. Sign the second supplier without doing that and you have a backup list, not backup capacity.
Why must the tolerance bands overlap?
Because an interface belongs to two parties at once. Supplier A's axle and supplier B's bore make sense only as a pair. Looked at separately, each drawing is internally consistent and each inspection is correct.
The workable instrument is a separate interface specification: only the dimensions that cross the supplier boundary, with tolerances set by what has to assemble rather than by what each plant finds easy to machine. The buyer holds that document, and both suppliers work to it instead of to their own habits. It is also the only page a new supplier can be handed on day one without exposing either incumbent.
An interface specification that lives at one supplier is not a specification; it is that supplier's drawing with a different title block.
Who performs the assembly verification should be settled in advance. It can sit inside the assembly plant's first-article approval, or be delegated to a third party as a one-off fit check. It does not belong at either supplier, because that asks one party to approve the other's parts.
A form worth copying: nominal shaft diameter with its band, instrument, datum and sampling point; the same four for the bore; then one sentence saying the pair must assemble by hand without tools. That last sentence looks crude, and it is the only one an inspector can apply on the spot.
The interface specification does not need to be long. One page is enough: the dimensions that cross the boundary, each with its tolerance, instrument, datum and sampling point. Those five elements are the same five as in any tolerance clause; the difference is that here the buyer issues them once instead of each plant filling in its own.
Sharing one gauge has practical obstacles: it must travel, be calibrated, and fit into two production calendars. The common compromise is one gauge per plant with a quarterly cross-check against a common master, and a record kept of every cross-check.
A stricter version requires interface dimensions to be inspected with one shared gauge, or requires the two gauge sets to be cross-checked against each other on a schedule. Without a common gauge, two reports cannot be compared — and a report that cannot be compared is more dangerous than no report, because it looks like evidence.
Who should carry the switching cost?
Switching cost is usually recorded as a little extra administration. The bulk of it sits in four places: a second mould, a second gauge set, duplicated first-article approval and sealing, and the line trials and ramp-up at each switch.
Those four are chronically underestimated at the decision stage because none of them appears on a quotation. The common error is to compare the second supplier's unit price with the first and switch on a few points of difference; spread across the volume, those points are usually consumed by the four items above.
How it is amortised matters as much. Spread per unit, a one-off cost looks enormous on a small volume and dual sourcing appears uneconomic; spread per year, it thins out until it is invisible. The honest version lists the two separately and lets the decision-maker see which line is one-off.
The clean way to count it is per event rather than per unit. Mould and gauge are one-off, approval and sealing are one-off, trials and ramp-up happen at every switch. Counted per event, it becomes obvious how small a volume is too small for a second source, and the answer usually surprises the person who argued for the switch on unit price alone.
Which categories need two sources, and which do not?
The criterion is not spend and not risk appetite. It is the consequence of interruption weighed against the difficulty of substitution at short notice. Where a stopped line is expensive and no alternative exists at short notice, dual sourcing is warranted. Where the part is standard and available from the market, a single source plus safety stock is cheaper.
One category gets overlooked: parts from a dedicated mould. Their substitution difficulty is not machining capability, it is the tool. Dual sourcing such a part does not mean finding a second machinist; it means either duplicating the mould or writing a right to retrieve it into the contract.
For products in this sector, structural parts and interface parts repay a second source. Finishing and packaging rarely do — the substitution market is wide enough, and switching them does not disturb an interface.
One last test of a dual-source plan: ask each supplier separately what they would do if the other stopped tomorrow. The answers usually reveal whether an interface specification actually exists or whether both are working to their own drawings.
The plan is only as strong as its least documented link, and in most dual-source arrangements that link is not the mould and not the drawing. It is the gauge sitting on a bench in two buildings.
FAQ
The second supplier is cheaper. Should we switch?
Count the four one-off costs first, then the ongoing cost of an interface specification and cross-checked gauges. Look at the unit price difference afterwards; the conclusion usually changes.
Both work to the same drawing. Can they still be incompatible?
Yes, and this is the trap most often missed. The same drawing does not mean the same point inside the band. Either end of a tolerance produces a conforming part, and the two ends will not assemble. Nothing in either inspection report will look wrong, which is what makes it expensive to find.
Should a share of volume be held for the second source?
Yes, or it will drift away. How much depends on how fast it can ramp and how much interruption you can survive; some buyers set the share by the supplier's capacity rather than by preference. Price is not the variable in this decision.
Who should write the interface specification?
The buyer, because only the buyer sees both suppliers. Let either supplier write it and it will drift toward that supplier's machining habits.
All values in this article are illustrations rather than figures from a standard; sourcing and acceptance terms are best agreed in the contract and confirmed in writing.

Top comments (1)

Collapse
 
hiredov profile image
Dov at HireDov •

Great example of why a green first-article report can still hide a failed assembly. One practical addition is to freeze a “golden pair”—one axle and one bore assembled and approved together—then make the first-article plan test fit across samples, not only each part against its own nominal drawing. Record the mating-part lot and gauge ID with the result; otherwise a later mismatch becomes detective work across two suppliers and several spreadsheets. For production changes, tie revalidation to any change in tooling, material lot, or measurement method. A backup source is real only when the handoff is repeatable.