Every Production Outcome Has a Story in the Data.

The run begins as six records in five places. A weighbridge ticket states 24.0 t net. The ERP goods receipt, posted by a supplier-facing team, states 52,911 lb under receipt GR 88/A. The batch sheet refers to Lot 0088, and the laboratory export calls it LOT-88. The filler count sits in a spreadsheet. A pomace inventory adjustment appears later with no reference to the run at all.

Connection means establishing that these are the same event. Identifiers are resolved against site-specific rules, units are converted, and timestamps are aligned. The tank transfer log and line schedule then allow each movement to be attributed to the run. Every link is recorded along with the basis for making it, and links made on weaker evidence, such as a time window, are marked so they can be questioned. Originals are preserved unchanged.

24,000 kg went in and 15,480 kg of finished puree came out. Read naively, that suggests 8,520 kg of waste. It is the most common misreading in food manufacturing, because not every decrease in finished-product weight represents waste.

The balance classifies mass by what actually happened to it. Select each class below to see its definition and the evidence behind it.

With a reconciled history available, a yield figure can be compared with something more meaningful than an annual standard. For AP-2417, a model trained on this line’s past runs estimates 66.2%, with a range of 65.1% to 67.3%. It uses information known about the lot and the run, including solids, firmness, variety and product specification.

Actual yield was 64.5%. That is 144 kg below the bottom of the range and 408 kg below its centre. Most runs on the chart fall inside their ranges, even where their yields differ noticeably, and those differences do not call for investigation. This is the distinction between normal process variation and a deviation worth examining.

A flag prompts a question, and it arrives with material to answer that question faster: the factors most strongly associated with the difference, the runs most like this one, and the records that are missing. The value of the gap is stated as a range, which helps a reviewer decide how much effort it merits. The full brief for this run appears further down the page.

Investigation is where the plant’s own knowledge takes over. The people who know the line check the finisher, ask about the fruit, and look in the tank. Their conclusions become part of the record.

Suppose engineers act on the finisher finding and adjust maintenance. Did it work? Comparing average yield before and after can mislead, since better fruit arriving the following month would look like a process improvement. Comparing each run with its own expected yield removes much of that effect, because the model already accounts for the raw material.

The result is reported with its uncertainty, and quality outcomes are reviewed alongside it. An improvement that moved product toward a specification limit is a different result from one that did not.