My New Wick Passed One Full Candle Test — How Many Repeat Burns Should I Perform Before Approving It for Production?

Your result, made precise

The worked example, labelled: a 200 g candle at 8% (16 g of Lavender and 184 g of Luxury Soy Wax CSI 464, load as a percentage of total weight) in your own jar, measured at 8 cm inside. The Cotton Braided wick you started with sooted by hour three. You stepped down one CSI rung to Eco C2, poured one candle, cured it and burned it in 4-hour sessions to the end. It passed every limit on your sheet. You're asking how many more before you order production wicks.

Here is that one candle's log, against the limits you set before burn 1. Every figure is illustrative; the point is the last column.

On a pass/fail sheet, that's a clean pass. On a margin sheet, the pool depth in the bottom third came within 0.1 cm of the line. That single number changes the answer to your question, and it's why this post asks you to write margins down, not ticks. The method for setting those limits is the cluster's job elsewhere (the acceptance criteria for a smaller pool, what an acceptable commercial flame looks like); here they're inputs.

What one candle can and can't prove

Suppose a wick really fails some share of candles, p, because of small differences you can't see: wick to wick, pour to pour, centring, cure, the room. The chance that it passes every one of n candles is (1 − p) to the power n. That's ordinary probability, not candle science, and it explains why a single good candle is so weak as evidence.

Read the top row first. A wick at the edge of your jar, failing half of all candles, passes one candle 50.0% of the time: one pass is a coin toss. Three passes would happen by luck only 12.5% of the time; six, 1.6%. That's what replicates are for: they reliably catch a wick that is wrong often.

Now read the bottom row, honestly. A wick that fails 1 candle in 10 still passes six in a row 53.1% of the time. Turn it round: if every candle you've burned has passed, the highest failure rate you can rule out (at 95% confidence) is about 95.0% after one candle, 63.2% after three, 52.7% after four, 39.3% after six and 25.9% after ten. No small maker burns enough candles to prove a rare fault is absent. So replicates are one of three defences, not the only one:

The same logic sits behind the series' "three candles per candidate" rule (how many candidates side by side): it's the smallest number that separates a wick from one odd candle. And it's why two published posts on this exact trap are worth reading: a two-hour pass that failed after longer burns and a first burn that looks fine after a size change.

Why a passed wick can still fail: causes, ranked

Each row is something that differed little or not at all in your one test, but will differ across production. The ranking is our judgement of how often each one turns a pass into a fail for a single-wick soy container candle; signs are common observations, not CSI burn data.

Causes 1 and 2 are the wick; 3–5 are how the candle is made; 6 and 7 are how it's burned. A good replicate plan varies the making causes a little on purpose (another pour date, another pack, perhaps another lot) and holds the burning causes still, so that a fail points at the wick or the formula and not at your Tuesday afternoon. Cause 1 is the one this worked example is worried about, and it's the only one where more candles is the wrong fix: if the wick is at the edge, the replicates will say so, and the answer is the formula or the construction (the cluster hub, when neither wick size works properly, ranks those options).

How many replicates: three tiers and the margin rule

There's no industry rule for replicate counts, and we won't invent one. These tiers are our judgement for a small Indian candle business making single-wick and double-wick soy container candles; they're written so you can defend the choice to yourself later.

The worked example is Tier C: the pool depth sat at 93.3% of its limit in the bottom third, and an 8 cm jar is where C2 runs out of room above it (Cotton Braided, the next rung, is the wick that sooted). So: six more candles, three on each of two pour dates. With the original that's 7 in the decision.

The margin rule (ours, labelled): after the replicates, look at the spread as well as the verdict. If all six pass and their worst depths sit comfortably below the first candle's 1.4 cm, the first candle was the high one and the wick is approved. If they pass but cluster at 1.4–1.5 cm, the wick is passing by millimetres: approve only if you accept that, and expect the occasional fail in production, especially in summer. If any one goes over, the wick fails, and a seventh candle won't rescue it. Indian summers (roughly April–June) run warmer rooms than October; whichever tier you choose, keep one candle from the approved batch and burn it in the hottest month you sell in.

The replicate plan: candles, dates, measurements, approval

Here's what the worked example's replicates might show. It's illustrative, and it's chosen to show why the test exists.

Six of seven pass, and the wick fails. That's the rule doing its job: the first candle's thin margin was real, and a warmer afternoon in the second set pushed one candle over. The decision isn't a seventh candle; it's the formula or the wick. Options, in order of cost: test 7% or 6% (both inside Lavender's stated 6–10%) with C2; check whether the room is the whole story by burning R5's twin in a cooler room; or ask us about a construction between C2 and Cotton Braided, which we'll tell you plainly we don't list. If every replicate had sat at 1.2–1.3 cm, the first candle's 1.4 cm would have been the high one and C2 would be approved.

Wax: a 1 kg plus a 500 g (₹767.40) is cheaper than two 1 kg packs (₹1,014.80) for this test. Time is the real cost: about two and a half weeks of evenings. What's at stake if you skip them, at 2,000 candles, is the next post's question.

The CSI wick buy list and the step to production

CSI's sized, tabbed wicks are three rungs, C1 → C2 → Cotton Braided, plus one wooden width. There's no sized, tabbed wick between C1 and C2 or between C2 and Cotton Braided. If your replicates say the wick is at the edge, we can't sell you a half-size; the honest options are a pair of smaller wicks, a formula change, or a question on WhatsApp.

Argued against our own interest: buy the small pack for the replicates even if you're sure. C1, C2 and Cotton Braided cost the same per wick at every pack, so a 100-pack bought now saves nothing and leaves you holding 94 wicks of a size your replicates may reject. And if your replicates pass on a wick from a pack you already own, buy nothing: use those. (Coming down from Cotton Braided to C2 happens to save ₹2.56 a candle; a side effect, never the reason.)

The step to production quantity

Only after every replicate passes. Production wicks = candles × wicks per candle × (1 + a defect allowance). Worked example, labelled: 250 candles a month on C2 plus 2% is 255 wicks: 2 × 100 + 50 + 10 = 260 for ₹2,454.40. A 500-pack costs the same ₹9.44 a wick and ties up ₹2,265.60 more. Reorder point = daily use × your lead time + safety stock; CSI publishes no lead times, so ask on WhatsApp before you set yours. Keep twenty wicks from the first production pack, sealed, as your reference (6995), and compare every later pack against them (6994). For the specific question of a large order, the one-time purchase that makes replicates look cheap is in the 2,000-candle post, and the bulk-buying gate in published form is bulk-buying one size only after every major fragrance has passed.

What to buy, and what to skip

Five items, chosen on page statements, price and stock on 7 October 2026. CSI publishes no burn, soot or flame data for any wick, so nothing here is ranked on performance, and each card says when not to buy.

Skip for now: any production pack before the last replicate passes; a new oil "to see if it's better" while the wick is under test (one variable at a time); a larger wax bag than the 1.5 kg you need; and a second wick size "just in case", unless the margin rule tells you to test one.

Frequently asked questions