Would Your Parts Still Pass the Bin Test?

A Demonstration in a Workshop Near Paris

On 30 August 1785, Thomas Jefferson, then the American minister to France, wrote a letter to John Jay about something he had just seen with his own eyes. A French gunsmith named Honoré Blanc had presented him with the parts of fifty musket locks, taken to pieces and arranged in compartments. Jefferson did what any good engineer would do with a claim like that. He tested it: “I put several together myself taking pieeices [sic] at hazard as they came to hand, and they fitted in the most perfect manner.”

That is the oldest interchangeability record I know of, and I keep coming back to how it was established. Blanc did not declare his parts interchangeable. He demonstrated it, in front of a skeptical witness, who reached into the compartments at random. Call it the bin test. The claim, any piece replaces any other piece, was earned in an afternoon, and the witness put his name on it in writing.

Some 240 years later, the industry is once again arguing intensely about interchangeability. Last month, in Interchangeable Does Not Mean Identical, I put the two leading answers to plant-specific manufacturing side by side: Prof. Dr. Jörg W. Fischer’s Object Type Cascade, a.k.a. the “Part Proxy” since CADTALK’s podcast episode with him in May 2025, and re-identification with a CM2 equivalent item record. I weighed their pros and cons and argued that both can work; which one fits depends on your governance maturity and systems landscape. But I closed with a question I deliberately left open: whichever structure records the interchangeability claim, who validates it? The longer I sat with that question, the more I kept thinking about Jefferson’s letter. Last month was about where to record the claim. Blanc would ask the question this article is about: who demonstrated it?

What Did Blanc’s Test Actually Prove?

The bin test worked because the test matched the life of the part. A musket lock in 1785 had a short list of things it would ever be asked to do: be assembled, be fired, be swapped in the field by an armorer. Every dimension of interchangeability that mattered could be witnessed in a single sitting. Pick at hazard, assemble, done. The test covered the lifecycle because the lifecycle was small.

The modern descendant of that demonstration is three checkboxes on a change form: Form, same. Fit, same. Function, same. Somewhere in those 240 years, we swapped the demonstration for a declaration. We kept shrinking the number of witnesses until one design engineer, working from one desk, could settle a claim that now has to hold for assembly lines on three continents, service vans, spares warehouses, diagnostic software, and a regulator scoping a recall.

A part number is the memory of a demonstration. Most organizations only kept the number.

That is not nostalgia. It is an operational gap with a body count, and we do not need a hypothetical to see it.

What Happens When the Number Outlives the Proof?

In April 2006, a General Motors engineer approved a redesign of the ignition switch used in the Chevrolet Cobalt and other small cars. The new switch had a longer detent plunger and a stiffer spring, correcting a switch so far below torque specification that its own designer later called it “the switch from hell.” The redesign was the right engineering decision. What followed was not: the redesigned switch kept the part number of the old one, as GM’s own Valukas investigation later documented.

Think about what that single identification decision did to everyone downstream who reasons with part numbers, which is everyone. Investigators could not explain why the failures stopped appearing from the 2008 model year on, because the data said the switch had never changed; when one engineer wondered whether the switch had been modified, the engineer responsible for it said no. The defect’s true population stayed invisible for years. The change only became widely known when a plaintiff’s expert X-rayed and disassembled switches from early and late model years and found different internals behind the same number. By the time the recall came in 2014, it covered 2.6 million vehicles, and the compensation fund eventually tied 124 deaths to the defect. GM signed a deferred prosecution agreement with a $900 million forfeiture.

Note what kind of failure this is. The old and new switches were not interchangeable; one met the torque requirement, and one did not. Under any honest assessment, that difference in Function demanded a new part number. That is last month’s argument in its starkest form: interchangeable does not mean identical, and the number must say so. Re-identification exists precisely to make that kind of difference visible, and skipping it did not just weaken traceability. It actively falsified the fleet: every analysis, every warranty query, every complaint trend ran on the false premise that one number meant one construction. In other words, the GM case is the bin test run in reverse. Instead of a demonstration proving that two constructions are one part, a declaration concealed that they never were.

Both Architectures Record the Claim. Neither Demonstrates It.

The pressure behind the Part Proxy debate is structural, and I described it last month: one engineering definition increasingly gets realized in multiple plants, from multiple suppliers, under local content rules, and the identification scheme has to carry that without going quietly false. The Part Proxy carries it by giving each plant its own material number, born from one engineering part, under Fischer’s unhedged rule that “material numbers must be revision-free!”, because the same material must always be 100% interchangeable. The equivalent item record carries it by giving each construction its own part number and cross-referencing the numbers that fully replace each other.

But notice what both structures quietly assume: that somewhere, somebody validated the claim they record. The Part Proxy asserts that all materials born from one part are 100% interchangeable. An equivalent item record asserts the same thing about two numbers. Both are declarations. Neither is a demonstration. I believe that the industry is standardizing the machinery for recording the claim while leaving unsettled what the claim must prove, and who witnesses it.

What Would a Modern Bin Test Look Like?

Start with what the claim must prove. The conventional Form, Fit, and Function view is incomplete, and I have argued before that FFF alone is not enough. In my book, and in last month’s article, I define interchangeability across four dimensions, adding Interface: the interaction between parts, and between parts and their environment and users, from communication protocols to APIs. Form, Fit, Function, and Interface (F3I). The definition bears repeating:

Parts that are capable of fully replacing each other in any application. Fully: achieving the same results without any special measures, except for normal adjustment.

Now run Blanc’s demonstration against that definition, in a modern setting. The compartments are no longer fifty boxes on one bench. They are spares warehouses on three continents, supplier stock, service vans, and units already twenty years in the field. “Taking pieces at hazard” now means: any unit, from any plant, from any approved supplier lot, into any application ever issued that is not yet end of life. And “they fitted in the most perfect manner” now has to hold for more than assembly. The service manual must match the internals of the unit you are holding. The diagnostic tool must pair with its firmware. The spares kit, the repair procedure, the upgrade path must all achieve the same results without special measures.

Jefferson could witness Blanc’s claim alone in an afternoon, because in 1785 assembly was the whole story. No single person can witness a modern F3I claim, because its dimensions are seen from different desks. Engineering sees geometry and performance. Manufacturing sees process capability. Service sees diagnostics, manuals, and field tooling. Oleg Shilovitsky, analyzing why Form-Fit-Function decisions fail without product memory, puts his finger on exactly this when he observes that in many organizations FFF is “treated as a quick judgment call” that “can be resolved by a single person who knows the design well.” The ignition switch detent plunger went through a single pair of hands. The bin had witnesses; the checkbox has none.

Who Witnesses the Demonstration Now?

This is where I think Fischer’s most durable contribution lies, more than in the Part Proxy itself: the Form-Fit-Function Corridor I pointed to at the close of last month’s article, the corridor an interchangeability claim must hold across the whole lifecycle, from first design intent to final service action. The corridor is, in essence, the modern bin test stretched over decades. And a demonstration stretched over decades cannot be an event. It has to be a process.

That process exists, and it is the reason I keep pointing to the CM2 framework in this debate. In the CM2 framework, Naming, Numbering and Reuse states the identification principle in one line: identification numbers are used to control and signify interchangeability. In other words, the number is not a label. It is the standing record of a proof, and it may only be reused where the proof still holds. Where constructions remain fully interchangeable under different numbers, CM2-500 cross-references them in the equivalent item record I covered in depth last month; where a substitution is scoped or temporary, an approved alternate item record carries the change authority for its use. The claim is never left as an unqualified sentence.

And crucially, CM2 defines who the witnesses are. The CM2 change process is a closed-loop process in which the change leader identifies all affected items and datasets, along with impacts on cost, schedule, risk, customers, supply chain, and internal sites, with high-risk changes going before a Change Review Board. That is Blanc’s workshop rebuilt as governance: the people who represent all aspects of F3I are brought in front of the claim before it is made. An equivalence record created outside that loop is not a proof. It is a rumor with a template. Had the 2006 switch redesign passed through such a loop, the question “does the old switch fully replace the new one in every application?” had an obvious answer, and with it, an obvious re-identification.

Before anyone ticks “no re-identification required,” ask what the demonstration would be: who would have to take this unit at hazard, and in which application, for the claim to fail? If nobody in the room can answer that, the assessment has not happened yet.

What Should You Do First?

Not pick an architecture. Three things come first.

First, run the bin test as a thought experiment on your highest-risk part numbers. Take a part built in more than one plant or bought from more than one supplier. If you pulled units at hazard from each source, would the manual, the spares kit, the diagnostic tool, and the repair procedure hold for all of them, in every application? If nobody can say, then the interchangeability your systems assert was declared, not demonstrated.

Second, give every F3I dimension a named witness. For each critical part family, write down which desk validates Form, Fit, Function, and Interface: engineering, manufacturing, service, supply chain. If a dimension has no name next to it, that is the column where your next unwitnessed change hides. And make the re-identification decision itself carry co-signatures from those desks; re-identification is not about following a decision tree, it is about seeing the full scope of interchangeability before making the claim.

Third, make your change process the witness. Route supplier and plant realization changes through cross-functional impact analysis, and record the outcome deliberately: a new number where the proof fails, an equivalent item record where it holds, an approved alternate where it is scoped. Whether you then enforce the engineering-manufacturing boundary with a Part Proxy or with disciplined identification is a real choice, and an honest one, but only after the claim itself has been earned.

Closing Thoughts

Blanc’s real innovation was not the tooling that made parts alike. It was making the claim testable, and submitting it to a witness who reached into the compartments at random. The part number was born as the memory of that demonstration.

Let me be precise about what that demands today, because it is not fifty units assembled in front of a visiting diplomat. Most real interchangeability claims will never rest on a full physical demonstration, and no honest process requires them to. Some rest on a qualification test of a sample lot. Some rest on simulation, valid within the limits of a model that was itself validated against hardware. Some rest on engineering analysis and similarity to a construction with a decade of field history. That is not a corruption of Blanc’s test; it is engineering. Evidence comes in grades, and the grade should match the risk. But each of those is still a demonstration of something, with a defined scope and known limits, and the discipline is to record which one the claim rests on, and where its coverage ends. What the ignition switch case shows is not evidence of the wrong grade. It is a claim backed by no evidence at all: a checkbox where the record of proof should have been. The Part Proxy debate will run its course, and I suspect different companies will land on different structures, which is fine. What is not fine is any structure, old or new, asserting interchangeability backed by nothing but the assertion itself.

Last month’s question was where to record the claim. This month’s question is what the claim rests on, and who stands behind it. An organization needs both answers, because a record without evidence is exactly what Blanc refused to offer Jefferson: a paper claim.

So here is my closing question: the last time your organization declared two constructions interchangeable, what did the claim rest on (a test, sample, simulation, similarity), and who put their name to it? Let me know what you think.