Exhaust manifold dimensions drift about 0.1 mm after 300 parts because the cavity wears and wax lots change. A 5,000-cycle tool service and a per-batch cavity check hold the drift inside that band; without them it grows until the machining allowance is gone. The drift is a process number, not a defect count.
Two wear paths open at once, and they are measured differently. The cavity wears where the wax flows hardest, at the gate and along the runner bend, so a feature that sat at nominal after the first article moves a few hundredths of a millimetre. Wax lots change on a slower clock: a new lot can shift shrinkage by a fraction of a percent, which is small on the body and enough to matter on a flange face.
On this manifold the exhaust manifold dimensions drift showed as 0.1 mm over 300 parts, measured on the wax pattern at the flange centres. The cavity had moved in the direction the tool wears, and the wax lot change added to it rather than cancelling it. That combination is the normal case; the two effects rarely come in opposite directions.
Wear shows as a trend, not as one bad part. The first signal is a slow change in the flange centre distance on the wax pattern, moving in one direction across several batches. The second is flash that comes back at a corner that was ground to 0.05 mm, because the parting edge rounds as the tool cycles. The third is weight: a wax pattern that gains 0.2 g is telling you the cavity has opened somewhere.
A single pattern outside the band is usually a press setting or a cold tool. A trend across three batches is the cavity, and it is confirmed by measuring the cavity itself rather than by adjusting the press. Reading the trend costs one measurement per batch; missing it costs a machining allowance.
Three checks at three frequencies cover the drift without turning the line into a measuring room.
Every batch: five wax patterns measured on the flange centres and the body length. The trend across batches is the decision; a single value is only a flag.
Every 5,000 cycles: the tool is serviced, the parting faces are checked and the cavity is measured against the tool card. This interval is set by the wear rate, not by calendar time.
Every wax lot: a retained sample and a first-batch measurement. A lot change without a sample turns a known number into a guess.
The batch check reads the dimensional control in investment casting limit, not the drawing limit: it asks whether the exhaust manifold dimensions drift is still inside the band, not whether the last part passed.
Rework wins while the drift is under 0.1 mm and one-directional. A cavity that has moved 0.05 mm in one direction can be corrected by re-cutting the affected faces or by adjusting the compensation on the tool card: a day of tool room time. Re-tooling starts when the drift is not one-directional, when the parting face has been ground twice and no longer closes on a 0.05 mm gap, or when the correction needs to move more than 0.15 mm.
At that point the cavity is no longer the one the first article proved, and a new tool is cheaper than chasing the old one. A production case on stainless steel investment casting shows the same split: the first service is a rework, the third is a decision.
A 0.1 mm drift is not a scrap sentence. It says the process moved inside its band and the 1.5 mm machining allowance still absorbs it. It does not say the next 300 parts will behave the same way; that depends on the service interval and on how the next wax lot behaves.
It also does not say the casting is out of tolerance. CT6 under ISO 8062-3 allows 0.9 mm of form deviation, and a 0.1 mm drift sits well inside it. The number that matters is the allowance left for machining, and that is what the batch trend protects.
Send the drawing and the batch history, and we will return a drift budget with the three check frequencies and the tool-service interval that keeps the casting inside its allowance, priced with the vacuum investment casting route before the first batch runs.