The Sample You Approved Was Made on a Better Day

Something happens when a first sample comes out of a new die. The tool is fresh. The setter is often the person who built it. The engineer is standing at the machine, watching. The die itself is at shot four hundred, still carrying its original polish, still tight. Everyone in the building is paying attention to this one part.
Six months later your parts come off the same die at shot one hundred and twenty thousand, on a Tuesday, on the night shift.
Nothing was hidden from you. Nobody lied. The sample was the best case and production is the normal case, and those are two different processes wearing the same part number.
The gate only opens one way
Approval processes are designed to catch a bad part. They are not designed to catch a favourable process.
Think about what the sample run actually is. One cavity, or a few, held to a long cooling time because the setter is being careful and speed does not matter yet. Metal dosed by hand or by a ladle working well within its accuracy window. A machine chosen because it is the best match, not because it is the one that happens to be free. Inspector standing over every piece. The best six parts shipped, the rest absorbed into internal scrap.
Every one of those conditions is a reasonable thing for a supplier to do. At the sampling stage, you want the tool proven, and nobody wants to stress a process that has not been established yet. The trouble is that the resulting part is evidence about the tool, and you will read it as evidence about the process. Those are not the same claim, and the approval paperwork does not separate them.

Eight conditions worth auditing
Each row below is a condition that usually differs between the sample run and the production run. The last column is what you can reasonably ask to be written into the approval, so the gap is at least visible.
| Condition | Sample run | Production run | Write into the approval |
| Cycle time | Long cooling, deliberate pacing, speed irrelevant | Fastest cycle that holds the part, cost per shot decides | The cycle time the sample was made at, and the range allowed in production |
| Metal dosing | Hand ladle or auto ladle inside its comfort zone | Same ladle at rate, with temperature drift across the shift | Dosing method, metal temperature window, and how it is monitored |
| Cavities sampled | The best cavity, or the one the setter trusts | Every cavity, in rotation, on every shift | Which cavities produced the sample, and results per cavity |
| Machine | Best-matched press, freshly set | Whatever press is free when the schedule fills | Machine envelope and tonnage, not a machine serial number |
| Die condition | Shot 400, original polish, no heat checking | Shot 120,000, crazed gate, one weld repair | Shot count at approval and a re-approval trigger at a stated interval |
| Who is running it | Tool room setter, engineer present | Production operator, twelve-hour shift, no engineer | Whether the process sheet is complete enough for a new operator |
| Inspection | Every piece measured by hand | AQL sample from a lot | The sample plan and what happens to a lot that fails it |
| Downstream ops | CNC on a one-off setup, polished by the finisher | Fixture in a cell, batch finish, thickness at the low end | Machining allowance and finish thickness limits, measured on the sample |
Two more rows belong on this table if your part carries them. Ambient conditions matter in a shop without full climate control, because die thermal equilibrium shifts with the season and tool steel moves with it. And any operation that a pipe fittings manufacturer would recognise as pressure-critical, a sealing face or a threaded port, deserves its own line, because the leak test result on a hand-made sample tells you less than almost anything else on the list.
What the documents will not flag
A dimensional report can be perfectly clean and still describe a process that cannot hold the part. Here is the mechanism, and it is not a paperwork trick.
Suppose the drawing allows plus or minus 0.15 on a bore. On the sample run the die is cold and stable and every part lands at nominal. The report looks excellent. Now the die runs at rate, the die temperature climbs twenty degrees higher than it ever did during sampling, and the same feature starts drifting to the top of its band. Mid-shift it crosses. The dimension was never the problem. Thermal growth was, and nobody measured it because the sample run never produced the condition that causes it.
The classic way to hide this is sorting. Take a production lot of three hundred parts, measure all of them, keep the ones inside tolerance, and run the capability study on the survivors. The Cpk looks respectable. What you have actually measured is aptitude at inspecting, not capability at making. Ask how the sample for the capability study was selected before you read the number.
Put numbers on it and the shape becomes obvious. Say a housing has a critical bore at 40.00 plus or minus 0.15, and the sample run put it at 39.98 with a spread of 0.03 across six parts. That looks like a capable process, and on those six parts it was. Now the die reaches thermal equilibrium at production rate and the die block grows. Thermal expansion on an aluminium casting between a stable die and one running forty degrees hotter is worth roughly 0.3 to 0.5 millimetres of length on a part this size, and it does not act uniformly, because the thick section at the bore warms differently from the thin rib beside it. Your bore moves up three hundredths and the spread widens because the die is no longer at a steady state. The centre is now 40.01 and the natural spread reaches 0.08. Still inside. Push the die temperature another notch for a faster cycle, because that is the lever a production supervisor pulls when a schedule slips, and you cross the upper limit partway through a shift. Nothing about this is hidden. The sample was honest, and the report was accurate, and the process still moved out from under both of them.
There is a second version of the same problem. Metal casting services that run a first article on trial parts rather than on parts made under production conditions are not being lazy. They are doing the only thing possible at that stage, because production conditions do not exist yet. That is exactly why the approval needs to state the conditions rather than assume them.
When the sample is good and production is not
If the first production lot fails and the sample passed, work outward in this order rather than starting with the die.
Check the process sheet against the machine. Most failures at this point are a parameter that was set during sampling and never written down, so it did not get set the second time. Cycle time and die temperature are the two that go missing most often.

Check the metal. A different heat, a different ingot supplier, or a higher share of internal returns in the charge will move fluidity and filling behaviour without changing the certificate on file. Chemistry also explains most cases where a casting looks right and then cracks after a week, which is a metallographic question rather than a dimensional one.

Which cavity, and which machine. A failure that appears on one cavity out of four is a tooling conversation. The same failure across every cavity on one machine is a parameter conversation. Those two get confused constantly, and the confusion costs weeks.
Die condition sits last on this list but early in the calendar. At a hundred thousand shots the gate area has heat checked, possibly been welded, and the filling pattern is no longer the one your sample was cast with. Ask for the shot count and the repair history before you ask for a tooling change.
Making the sample mean something
You cannot eliminate the gap. You can shrink it and you can document it.
Run the sample at production cycle time. This is the single highest-value change and it costs almost nothing except patience. If the part cannot be made at production rate, you want to discover that at shot four hundred rather than at shot forty thousand.
Sample every cavity. A four-cavity tool that has one good cavity and three marginal ones will pass a single-cavity sample and fail in the field. Ask for a dimensional result broken out per cavity.
Keep the process sheet that made the sample. Not a summary. Shot profile, intensification pressure, die temperature setpoints, vacuum level if there is vacuum, biscuit thickness, cycle time, and the metal temperature at the moment of the shot. A sample without its process sheet is an object, not a standard.
Retain the physical sample, signed and dated, in a place where both sides can reach it. Then when a dispute starts in month eight about what colour the finish was supposed to be, the argument lasts four minutes instead of four weeks. Boundary samples for cosmetic limits are worth more than any written description of a surface.

Set a re-approval trigger. Either a shot count, or a die repair that touches cavity geometry, or a material grade change, or a gap in production longer than a year. A tool that has been welded in the gate area is not the tool that made your approved sample, and the approval should say so.
Where the honest conversations happen
A supplier who resists running the sample at rate is usually a supplier who knows the tool will not hold at rate. That is worth finding out early, and it is worth paying a little more to avoid.
The same logic runs through the rest of the route. Whether you buy Die Casting Parts in aluminium or run Zinc Casting for a small housing, whether it is custom die casting for an OEM programme or OEM casting for a replacement part, the sample approval is the last cheap moment to change your mind. Later, you are arguing about a die that already exists.
Our own scope covers die casting services from DFM and tooling through aluminum die casting and hot chamber zinc, then 3/4/5-axis CNC machining, blasting, anodising, powder coating, plating, insert installation, sub-assembly and leak testing. It is one aluminum alloy processing manufacturing route with no undeclared handovers, so the shop that makes your sample is the shop that runs your production. Typical aluminium casting products and aluminium casting parts include die cast housing enclosures for industrial electronics, telecom and EV housings, motor housings, gearbox covers, LED heat sinks, valve bodies and structural brackets. Every custom casting job is built to drawing or sample, and the aluminum casting route is quoted with its machining allowance stated rather than assumed.
Alongside casting we run metal fabrication and custom metal fabrication lines, covering fabrication services from laser cutting and press brake forming through TIG and MIG welding, plus structural fabrication and aluminum fabrication from extrusion and plate. A custom fabrication or custom metal fab bracket is sometimes the honest answer for a low-volume part that would not justify a die. Aluminum metal casting and aluminium metal casting are both in scope, and we will say when a part should be machined from cast metal stock instead of cast at all. It is a shorter conversation than the one you have in month eight.
As a metal manufacturing partner we would rather run casting services for a customer who checks all eight rows above than one who signs off on a photograph of six good parts. Aluminum die casting suppliers are easy to find. The ones worth keeping are the ones whose sample and production runs look like the same process.
Metal casting service comparisons usually stop at a price per kilo, and metal fabrication and die cast metal work gets measured the same way. It is the wrong measure. The number that actually matters is the one produced at rate, on the twelfth shift, on a die at a hundred thousand shots. Any of the metal casting services you are comparing can hand you a sample price. Ask for the production number before you approve anything.
Frequently Asked Questions
Why would a supplier make the sample under better conditions than production?
Usually not out of deception. At the sampling stage the objective is to prove the tool, so the setter holds a longer cooling time, runs a single cavity and watches every part. Those are sensible choices in isolation. They become a problem when the buyer reads the sample as proof of what production will look like, which is the natural reading and the one most approval forms encourage.
Is it reasonable to refuse a sample not made at production cycle time?
Yes, and it is a normal request rather than an unusual one. State it before the sample is scheduled, not after it arrives, because a sample made at rate takes longer to set up and the supplier needs to plan the machine time. If a supplier refuses outright, that answer is more useful than the sample would have been.
How many cavities should the sample come from?
All of them, with results reported per cavity. A multi-cavity die does not fill every cavity identically, and the differences are predictable from the runner and gating layout. Sampling only the cavity the setter trusts tells you about the shop's judgment and nothing about the tool.
What is the difference between an FAI and a capability study?
An FAI is a dimensional report on parts from the tool, confirming the tool can produce a conforming piece. A capability study is a statistical claim about the spread of a running process, usually expressed as Cpk and built from a sample of consecutive production parts. An FAI can pass on a tool that will not hold tolerance at rate. A capability study run on sorted parts can pass on a process that never will.
Can I ask to see the process sheet that produced the sample?
Yes, and it should be a deliverable rather than a favour. Ask for shot profile, intensification pressure, die temperature setpoints, vacuum level, biscuit thickness and cycle time. If the supplier cannot produce it, the sample has no reference point, and reproducing it later becomes guesswork.
Should the approval include a shot count?
Yes. Record the shot count at approval and agree a re-approval threshold, commonly after a die repair affecting cavity geometry or on a stated interval. Heat checking is normal ageing, but a repaired gate changes how the cavity fills, so the approved sample stops being a valid comparison from that point.
What if the drawing changes after the tool is cut?
The approval lapses on any change that touches form, fit or function, and the tool needs to be re-sampled against the new revision. Freezing the drawing before steel is cut is the cheapest way to avoid this, and it is the single most common cause of a launch running late. If a change is unavoidable, agree who pays for the tool modification before the change is made, not after.
Does any of this apply to a simple part?
Less to the dimensional question, more to the finish and the downstream operations. A simple bracket rarely drifts out of tolerance at rate, but its coating thickness will sit at the low end of the range in a batch process, and its machined faces will come off a fixture rather than a programmer's one-off setup. Audit the rows that apply. Finish and secondary operations cause more disputes than casting dimensions do.

