My First 100 Candles and Last 100 Candles From the Same Batch Perform Differently — How Should I Sample Retained Products to Identify Where the Process Drifted?

Look for the shape before the cause

You already know the first 100 candles and the last 100 perform differently. That is a comparison of two blocks, and it cannot tell you what happened between them. The difference could have built up steadily over the run, or it could have appeared all at once at candle 87 when something changed. Those two shapes have different causes, so the first job of sampling is to find which one you have.

A slope — performance changing a little with every tray — points to something that varies with time: wax cooling and thickening in the melter or pitcher, oil distribution changing as the wax stands, lighter notes leaving the hot wax as the run goes on, or a pouring routine that slows as the operator tires.

A step — two plateaus with a sharp edge — points to a moment: a melter refill, a second oil container, a reheat after lunch, a new pourer, a new carton of jars. Steps are often easier to fix, because once you find the moment, you find the cause.

Reconstruct the run before you pull samples

Sampling by position only works if you know which candle was poured when. Gather everything that records the order: numbered trays, time-stamped photos, the batch sheet, the pour log, even the order jars were labelled. Then list every event during the run — a refill, a break, a reheat, a new oil bottle, a change of pitcher, a change of person.

This is often where the answer first appears. Many makers call something "one batch" when it was actually two or three melter fills scented separately. A 20 kg melter cannot hold 40 kg of wax at once, so a 200-candle run from it is at least two melts, each with its own oil addition, temperature history and hold time. Each melt is a sub-batch with its own potential drift.

A sampling plan that separates slope from step

Space samples evenly through the run so a slope shows up, and add extra samples either side of every known event so a step shows up. Keep the plan written down before you pull anything, so you are not tempted to choose candles that look interesting.

Fourteen samples, including the reference, is a suggested starting point to validate. If the results show a change between two positions, go back and pull candles between them to narrow the boundary.

Test the samples so the shape is real

Read the fill-weight column with some care. Empty jars from one carton vary a little, so a few grams either way on a single candle means little; a run of consecutive candles all drifting lighter means a lot. What you are looking for is a trend or a jump across several positions, not one odd value.

Plot net fill and throw rank against candle number. The picture is usually clearer than any single comparison.

Worked example: 40 kg of Citrus Lemon in CSI 464

A maker poured 200 candles of 200 g from 40 kg of scented wax at 8% of total weight: 36.8 kg of CSI 464 and 3.2 kg of Citrus Lemon. Wax at ₹507.40 per kg (from the ₹5,074.00 10 kg pack) is 36.8 × ₹507.40 = ₹18,672.32. Oil at ₹2,832.00 per kg is 3.2 × ₹2,832.00 = ₹9,062.40. The run holds ₹27,734.72 of wax and oil.

Suppose the samples show two different pictures. In the first, throw ranks fall steadily through each melt and recover after the refill, and net fill weight falls slightly towards the end of each melt. That is two slopes: something within each fill changes with time. Citrus Lemon's page warns that citrus top notes are volatile, so a long hold with the wax hot is a strong suspect, with cooling wax dragging in the pitcher behind the fill-weight drop.

In the second picture, candles 1–99 are consistent, candles 102–200 are consistent and weaker, and the step sits exactly at the refill. That points to the second melt receiving less oil, oil at a different temperature, or a longer hold before pouring began. The batch sheet should show which.

What happens to the 200 candles

The sampling result also decides what you can sell. In the step picture, candles 1–99 match the reference and can be released once burn tests pass; candles 102–200 are held until you decide whether they are simply a lighter version you can sell under a clear description, or below your standard and written off. In the slope picture, the boundary is blurrier: release the zones that match the reference, hold the tail of each melt, and test a few more candles near the edge before drawing the line.

Either way, do not blend held candles into the next batch by remelting them with fresh wax without validation. That spreads the drift into a batch that was otherwise fine.

Confirm on a small control, then fix the process

Sampling tells you where and roughly what; a control tells you why. For the slope, make two 1 kg portions from the same materials, pour one straight away and hold the other hot for as long as the long melt was held, then compare. For the step, recreate the second melt's recorded oil weight and temperature at 1 kg beside the first melt's. If the control reproduces the difference, you have the cause.

Hold times, portion sizes and sample positions in this page are suggested starting points. Validate them in your own workshop: the right hold limit for a 20 kg melt of citrus in soy may be very different from the right limit for a woody oil in coconut soy.

Then fix the process, not the formula. For a time-dependent slope, shorten the hold per melt or scent smaller portions in sequence. For a step at the refill, weigh and add each melt's oil separately and record it. Pour retained candles from the start, middle and end of every future run so the next drift is caught while it is small.

Where this page fits

CSI already has guides that cover parts of this question. This page covers the failure investigation for the question above; these go deeper on the rest:

Three things to buy for this job

Frequently asked questions