Sensory Testing

People can say they'd buy your product and still spit out the first bite. For anything tasted, smelled, felt, or applied to the body, opinion surveys and interviews aren't enough - you need sensory evidence, gathered under conditions that stop the result from lying to you. This is the discipline food, beverage, and personal-care companies use, scaled down so a first-time founder can run it. Proof before hype: a blind, controlled test beats a founder's own palate every time.

The method

Built on the Founder OS scaffold (High tier - this drives formulation and launch decisions). Full protocols, scales, and sample sizes in references/sensory-methods.md.

Step 1 - Decide what question you're actually asking

Match the test to the decision. The three families (see references for protocols):

Discrimination and descriptive use small trained/screened panels; acceptance needs untrained target consumers (your [ICP]) - never mix the two jobs.

Step 2 - Control the bias, or don't bother

A sloppy taste test produces confident nonsense. Non-negotiables (details in references):

• Blind & coded - samples labelled with random 3-digit codes, never "A/B" or brand names.
• Randomised & balanced serving order (counterbalance to cancel first-sample bias).
• Consistent conditions - same temperature, portion, vessel, lighting; neutral palate cleanser (water/cracker) between samples; separated tasting so people don't influence each other.
• Blind to your hope - whoever runs it shouldn't telegraph the "right" answer.

Step 3 - Size it and set the rule before you taste

Pick the sample size and the pass/fail rule up front (references give the tables):

• Discrimination: enough assessors to reach significance (e.g. a triangle test needs a set number of correct identifications for n tasters - use the reference table, don't eyeball).
• Acceptance: enough target consumers for a stable liking score (a screening read often starts around 50-100 target consumers; treat small panels as directional, not proof).

Step 4 - Run it, then read it honestly

Collect scores, compare against the pre-set rule, and report the result, not the hope. "No significant difference" is a real, useful answer (e.g. your cheaper formula is sensorily equivalent). A liking score only means something against a benchmark - your current product or the market leader, tested blind in the same session.

Output

• Test design: the question, chosen method, sample type (trained vs target consumer), sample size, controls, and the pre-set decision rule.
• Run sheet: coded samples, serving order, score sheet/scale, conditions.
• Result read: what the data says, significance/benchmark, and the formulation or go/no-go decision it supports - with assumptions flagged.

Constraints

• Acceptance ≠ discrimination. Don't ask untrained consumers to describe; don't ask a trained panel whether they "like" it. Different jobs, different people.
• Always blind, always benchmarked. An unblinded or un-benchmarked score is marketing, not data.
• Small panels are directional. Don't over-claim from 8 tasters; say so.
• Safety first. For anything ingested or applied to skin, confirm it's safe to test and meet the relevant food-safety / cosmetic-safety and consent requirements before anyone tries it - verify with a qualified expert; this skill is method guidance, not safety or regulatory advice.
• Stay theme-agnostic across consumable/applied products; the founder supplies the product.

Copy-paste version

Act as a sensory scientist helping a first-time founder test a consumer product people taste,
smell, or apply: [PRODUCT] for [ICP]. The decision I'm trying to make: [GOAL].
1. Tell me which test family fits - discrimination (are they different?), descriptive (how?), or
   acceptance (do target consumers like/prefer it?) - and the specific method.
2. Design it to kill bias: blind 3-digit codes, randomised/counterbalanced order, consistent
   conditions, palate cleanser, run blind to my hope.
3. Tell me the sample size and the pass/fail rule to set BEFORE I taste, and what benchmark to
   test against blind in the same session.
4. After I give you the results, read them honestly - significance vs benchmark - and the
   formulation or go/no-go decision they support. Flag anything that's only directional.
Remind me to confirm the product is safe and consented to test before anyone tries it.