Compliance

Sensory testing

"It tastes better" is not data. Sensory testing turns taste into numbers you can track across batches, compare against a gold standard, and defend to a buyer. None of it needs a laboratory — it needs discipline: the same questions, the same conditions, the same scale, every time.

Which test for which question

Panel conditions — the boring part that decides everything

1. Neutral environment Quiet, odour-free room; no cooking smells, perfume or ambient music. Natural light or neutral white for colour judgement.
2. Coded samples Three-digit random codes, never names; randomised presentation order (different order per assessor).
3. Controlled portions Equal sample sizes at room temperature — warm samples taste sweeter and softer; define the temperature and keep it.
4. Cleansers Plain water and unsalted crackers between samples; 30–60 s pauses.
5. Spit-out and ethics Provide spit cups for trained panels; only taste food that passed the safety checks — sensory panels never sample suspect product.
6. No discussion Assessors record independently before any conversation; discussion contaminates scores.

Difference tests: is it actually different?

Triangle test

Each assessor receives three samples — two identical, one different — and picks the odd one. With 8–12 assessors, the count of correct answers is compared to a standard table (for n = 9, 6 correct answers is significant at p < 0.05). If the panel can't reliably pick the odd sample, your "improvement" is in your head — which is exactly the kind of thing worth knowing before you change the process.

Paired comparison

Two samples, one question: "which is sweeter?" Fast, decisive, ideal for A/B testing a process change (e.g. dipped vs undipped colour). Count preference; significance tables decide.

Descriptive profiling: what changed

A trained panel scores defined attributes on a line or 0–10 scale. For dried food, a good generic profile covers appearance, texture, aroma and taste:

Score each sample in triplicate across sessions, average, and compare. A radar-chart of these attributes against your gold-standard batch is the single most useful quality picture a small producer can own.

Hedonic testing: do people like it?

• The 9-point scale: dislike extremely (1) → neither like nor dislike (5) → like extremely (9). Report the mean and the distribution.
• Panel = target market: 30+ consumers is the practical minimum for stable means; farmers-market customers are a fine start but they are self-selected fans — note the bias.
• Add purchase intent ("definitely/probably would buy") and one open question ("what would you change?").
• Watch the middle: a 5.8 mean with all scores at 5–7 is a safe, boring product; a 5.5 mean split between 2s and 9s is a love-it-or-hate-it product — different strategies follow.

Sensory shelf-life: the endpoint test

1. Store dated samples from one batch under real conditions (your storage, your packaging).
2. Test at intervals — 0, 1, 3, 6, 12 months — using the difference test against the fresh reference plus key attribute scores.
3. Define the endpoint in advance — e.g. "off-flavour ≥ 3, or significant difference from reference" — before you start, or you will argue yourself into extra months.
4. Set the best-before from the endpoint, with margin. This is the evidence behind the date on your label.

Printable score sheet

Five ways panels lie

• Non-blind samples — the brand or process name on the plate biases every score. Code everything.
• Untrained "trained" panel — without anchors and practice, scores drift. Train with known references (e.g. deliberately scorched vs ideal mango).
• Wrong temperature — warm fruit reads sweeter; define and control it.
• Discussion before scoring — one confident voice moves the whole panel.
• Testing only fresh product — the interesting data is at 6 and 12 months; schedule the storage tests.