A buyer pulled one 8 oz sandalwood jar from a batch of 800, lit it, and called it weak. The factory pulled the batch log, found the fragrance load within spec, and defended the pour. Nobody could say what actually changed, because nobody had written down what “within spec” means past a single percentage.
That rejection is not a bad candle. It is a missing tolerance.
Batch consistency does not come from a better jar, a better blend, or more careful stirring. It comes from fixing the tolerance between variables – and measuring the few that actually move. This post is the spec sheet I hold a production line to: five numbers, each with its failure mode, its tolerance, and the measurement that proves it. Not theory – the lines where a batch gets released or reworked.
Why “within spec” is not a spec
Most candle QA docs say things like fragrance load 8% or pour at the right temperature. Those are set points, not tolerances. A set point tells you where to aim. A tolerance tells you how far you can drift before the batch stops being the same product as the last one.
The gap between the two is where batch variation lives. Same recipe, same wick, same jar – and one batch throws strongly while the next one burns clean but quiet. The difference is never the candle. It is the process spread nobody measured.
Here is the fix. Five numbers, in the order they usually fail.
1. Fragrance load tolerance – not just the percentage
The set point is easy: 8% of wax weight. The tolerance is the part most logs skip. Weighing by volume, or eyeballing the scale on a hot day, lets the actual load drift 1.5 to 2.5 points between batches that both get written down as “8%.”
That drift is not cosmetic. A 1-point change in load shifts the pour window by roughly 5 to 10 degrees, moves the cure timeline, and changes how much scent the finished wax actually carries. Two batches with the same wick and the same jar can throw completely differently – with the difference living entirely in a number nobody recorded past the decimal.
The rule: weigh fragrance by mass for every batch, log the exact figure, and hold the tolerance to ±0.5 point of the recipe. Never measure by volume. If the scale reads 1,240 g and 1,310 g across two runs of the same recipe, that is a batch difference already – and it will show up as a scent difference whether anyone notices it or not.
2. Pour temperature spread – the number that separates clean pours from cracked ones
A pour temperature is not one number. It is a range that moves as the pot cools, and a large batch spans 30 to 90 minutes of pouring. The first jar and the last jar can sit 8 to 12 degrees apart in wax temperature, even from the same pot, if nobody is watching.
That spread is the quiet source of batch-to-batch variation: the early jars set differently from the late ones, which changes how the fragrance bonds during cure, which changes throw.
The spec: log the pour temperature at the start and every 50 to 100 ml after that. A batch with a spread over 5 degrees between first and last jar gets flagged for rework consideration before it ever gets a lid. And a pot that needs more than 60 minutes to finish a batch has already started setting up inside the pot – split the pour rather than racing it.
This is the number that explains why one corner of a production run looks different from the other. It is not a mystery. It is a temperature spread that nobody logged.
3. Burn time variation – the only number the customer can feel
Burn time is the metric buyers actually experience, so it is the one your QA line must be able to reproduce. The working rule for a standard jar candle is 2.5 to 3.5 hours of burn per inch of diameter, with the melt pool reaching full width inside that window and no glass contact.
For an 8 oz jar with roughly 6.3 to 7.6 cm of inside diameter, a healthy burn hits full melt pool in about 2.5 to 3.5 hours. Below 2.5 and the pool is tunneling – the wax is not reaching the edges. Above 4 and the pool is pressing into the glass, which is a heat risk and a soot risk at the same time.
The batch test: light three jars from the batch, same room, same airflow, 4 hours each. Record the hour the melt pool reaches the glass edge. If three jars land between 2.5 and 3.5 hours with no more than 30 minutes between the fastest and slowest, the batch holds. One jar outside the band is a signal – two outside it is a rework, not a note in the log.
4. Wick trim tolerance – the 6 mm number
A trimmed wick should sit 6.35 mm, a quarter inch, above the wax surface. That is the trim spec in the standard, and it is also the number your burn tests are only as good as.
Trim a wick 2 mm short and the flame runs smaller and cooler, cutting throw and pushing the melt pool into the tunneling range. Trim it 2 mm long and the flame climbs, soot climbs with it, and the burn time compresses. The same candle, same wax, same fragrance – different burn result, driven entirely by a 4 mm difference in a piece of cotton braid.
The rule is not “trim before lighting.” It is trim to a fixed gauge: 6 mm, every burn, every test, every batch. A 6 mm trim gauge costs nothing and removes the single most common variable that makes one operator’s burn test disagree with another’s.
5. Cure time – the number that is a floor, not a target
Cure is where fragrance bonds into the wax matrix, and it is the number most often treated as a waiting room instead of a process step. The working window for soy-blend container candles is 5 to 7 days, and the floor matters more than the ceiling: a batch tested at day 3 will read weaker than the same batch at day 7, and the difference can be mistaken for a load problem.
The spec: cure a full 7 days before any burn test, and log the cure days on the batch sheet. If two batches are compared and one cured 3 days and the other 7, the comparison is invalid – the wax was not in the same state when it was measured.
The QA line: what a released batch looks like
Put the five together and a released batch is not “looks right.” It is a line item in a log:
- Fragrance load: measured by mass, within ±0.5 point of recipe
- Pour spread: start-to-end temperature under 5 degrees, logged at intervals
- Melt pool: 2.5 to 3.5 hours to full width on three jars, spread under 30 minutes
- Trim: 6 mm gauge on every burn
- Cure: 7 days, logged
That is the whole line. When a batch passes all five, the rejection rate drops – not because the candles are better than last month, but because the variables that made last month’s batch drift are now bounded and written down.
How many jars to test? The sampling answer
You do not burn every jar. The working sample is three jars per batch for the burn test, plus the full pour log, which is the real early-warning system – it catches the spread before it ever reaches a jar. Three jars tell you the batch holds. The pour log tells you why it holds, or why the next one will not.
Frequently asked questions
Why did the same recipe throw differently this batch? In almost every case I have seen it, the variable was fragrance load – measured by volume instead of mass, drifting 1 to 2 points between runs. Check the load log first, then the pour spread. Wick and wax are usually not the difference.
How do you test burn time the same way twice? Same room, no draft, wick trimmed to 6 mm before the light, and the same start-to-stop rule: time to full melt pool at the glass edge, not time to extinction. Three jars, recorded side by side.
Is 4 weeks of cure necessary? For most soy-blend container candles, no. Five to seven days is the working window, and day 7 is the floor for any test you plan to compare against another batch. Four weeks is a buffer for very high loads or slow-bonding fragrance blends – measure, don’t default to it.
What actually causes batch-to-batch throw variation? Load drift and pour spread, in that order. The wick is the third variable, and it is the easiest one to rule out because it is a part, not a process.
The bottom line
A buyer who rejects a candle for weak scent is not describing one bad jar. They are describing a process with no tolerance. Add five numbers to the batch log – load, spread, melt-pool window, trim, cure – and the next rejection gets an answer in the log instead of an argument in the meeting. That is the whole difference between a batch and a product.