Consumer-scale blending of commercial whiskies

Inputs

A one-sentence sensory target, intended audience, and drinking context, or a clearly labeled exploratory inventory set
A declared decision question: detect a difference, describe attributes, measure liking, or choose a preference
A participant-role statement distinguishing the formulator, selected or trained assessors, and target consumers
Candidate bottles with brand/expression, batch, proof, fill level, and acquisition notes
Clean capped sample bottles, calibrated pipettes or graduated cylinders, identical glasses, labels or random codes, water, and a tasting record
An untouched control of each base and enough stock to rebuild the winning formula
An alcohol-specific participation plan covering suitability, informed participation, serving limits, total exposure, stop conditions, water and food, non-drinking alternatives, and safe transportation

Sequence

Choose the working mode. In exploration mode, taste inventory to discover promising combinations without calling every change an improvement. Before optimization begins, name and version the target.
Define or freeze the target. Separate an invariant quality floor from the variable creative objective. State the qualities every acceptable prototype must retain, then define the desired aroma, palate, mouthfeel, finish, proof range, audience, and use occasion. If either layer changes, open a new version.
Declare the sensory question before the test. Decide whether the next session is meant to detect a difference, describe attributes and intensity, measure liking, or obtain preference. Do not treat those response families as interchangeable, even when more than one is collected in a session.
Assign participant roles and controls. Use selected or trained assessors for repeatable discrimination or description and target consumers for liking, preference, and use-context questions. A participant may appear in more than one session, but the role, instructions, and inference must remain distinct.
Approve the alcohol-specific participation plan. Record eligibility or suitability, informed participation, planned serving and total exposure, stop criteria, water and food, non-drinking alternatives, and transportation. Follow applicable law, venue policy, and professional guidance; this procedure does not prescribe a universal safe pour.
Profile, qualify, and screen components. Taste candidates separately under consistent conditions. Record brand/expression, batch, proof, age when known, production recipe or mash bill when known, useful attributes, intensity, provisional role, stock constraints, traceability, stock condition, reconstruction feasibility, and any reject/hold fault. Apply the quality gate before formula search. Treat labels and prestige as hypotheses, not verdicts. When possible, compare coded pairs that differ mainly in one variable and record remaining confounds.
Choose a provisional architecture while preserving option value. For a small formula, nominate a structural lead and role-defined supports. For a more complex formula, lock a coherent base first, then add a middle functional layer and small accents. Keep components separate until evidence earns commitment, and treat role labels and source percentages as working hypotheses; the observed blend may promote, demote, or repurpose a component.
Reduce the pool, then choose the experimental region. Remove components that lack a plausible role or fail the quality floor before formula search. Because blend proportions sum to the whole, treat ratio trials as a mixture experiment. For three plausible components, use the seven-point simplex-centroid—three pure controls, three equal binary mixtures, and one equal three-way mixture—as the default diagnostic screen. If sensory intensity, proof, cost, stock, or target roles make much of that space implausible, document minimum and maximum bounds and search only the feasible region. Expand only promising directions.
Measure exactly and preserve the design. Record every volume or mass, component proportion, and component proof. Confirm that each formula sums to the declared total. Do not use an undefined “drop” as the permanent unit, and do not silently change a design coordinate.
Scale adjustment size to sensory impact and proof. Use smaller steps as smoke, high proof, old oak, finishing influence, or other intensity rises, and reduce the increment again near a promising formula. Evaluate proof as a flavor-control variable because it may change which component character leads. Treat any legal or published maximum as a boundary, never as a sensory target. Professional examples near five percentage points for base changes and two points for intense top-dressing are starting heuristics, not universal thresholds.
Evaluate observed roles and interactions. Compare each component’s intended role with what it actually does in the mixture. Ask whether the addition contributes a distinct function, preserves the quality floor, advances the variable target, or clarifies the sequence. Record reinforcement, suppression, contrast, sensory noise, and any role reassignment rather than predicting the result from component notes alone.
Narrow and branch again. Retain the best two or three candidates, then create documented variations. Never silently overwrite a formula.
Prepare the sensory session. Standardize sample identity, glassware, volume, proof treatment, temperature, rest, coding, instructions, response scales, and recording. Randomize serving order where practical. Use a bounded attribute lexicon and references when needed, but do not mistake shared words for assessor calibration.
Control carryover and fatigue. Arrange less assertive samples before more assertive ones only when that ordering serves the declared method; otherwise randomize or counterbalance. Limit the flight, split sessions when necessary, provide recovery intervals and palate resets, and record any heat, aroma, or adaptation that may contaminate later judgments. No exact whiskey-specific recovery interval is claimed.
Use discrimination tests only for discrimination. A triangle test may be considered for sufficiently homogeneous, low-carryover pairs when the decision is whether a perceptible difference exists or sufficiently close similarity can be supported. State which question is being tested. A triangle result does not identify the changed attribute, its direction, or which sample is preferred. Use licensed procedures and statistical rules if formal ASTM or ISO conformity is claimed.
Rest and repeat independently. Record immediate, one-hour, overnight, and later observations when stock permits. Randomize codes and order across independently prepared or independently evaluated sessions. Several sips from one pour are not independent replicates.
Preserve disagreement. Retain individual judgments. When tasters disagree, identify the disputed attribute and retest, seek an adjudicating taster, revise, or hold; do not average disagreement away automatically.
Run target-consumer work separately. Define the target population, comparison products, benchmarks, blind or contextual presentation, use occasion, and questions. Record liking and preference as distinct responses. Limit conclusions to the sampled population, products, comparison set, and context.
Test real-use robustness. Retaste in the intended context—after food, with water, over ice, in a cocktail, or with a stated pairing—without letting that session replace the controlled decision round.
Confirm reconstruction and scale. Rebuild the winning small formula from the original bottles. If scaling, compare the larger batch with a retained laboratory sample and document every correction.
Apply the stopping and inference rule. Lock the formula only when it repeatedly meets the target and further controlled changes do not produce a material improvement over the prior best. Keep every prediction and interaction claim inside the tested feasible region and require adequacy or validation evidence before using a fitted response model. Record the reason to approve, hold, or reject.
Document provenance. Save component identity, batch, proof, formula, dilution, inventory limits, dates, rest conditions, sample codes, participant roles, method, individual responses, carryover notes, and final version.

Advanced continuity or daisy-chain method

Use a separate ledger for every cycle: opening volume, amount removed, retained fraction, all additions, component proofs, calculated proof, sensory changes, and correction. Define a correction limit before starting; if the target or identity materially changes, create a new version. Do not present an undocumented infinity bottle as reproducible blending.

Outputs

Component profiles with use/hold/reject decisions, production metadata, confounds, and intended versus observed roles
Versioned prototype formulas and coded evaluation records
An experimental-design record containing the objective, component bounds, formula coordinates, randomization, independent repeats, responses, and adequacy checks
A sensory-method record containing the decision question, participant role, presentation, lexicon or scales, carryover controls, and inference boundary
An alcohol-specific participation and session-safety record
A confirmed final formula with an invariant/variable target brief, provenance, stopping decision, and scale-check evidence
When multiple tasters participate, an individual-response and disagreement record
For consumer work, a target-population and context record with liking and preference retained separately
For continuity blends, a complete cycle ledger

Current evidence boundary

Practitioner sources converge on measured iteration, target-led optimization, invariant quality criteria, candidate-pool reduction, provisional roles, hierarchical architecture, threshold-aware additions, repeated blind tasting, explicit stopping judgment, recordkeeping, and scale confirmation. NIST/SEMATECH supplies the experimental structure: proportions sum to one, objectives and responses precede formula generation, feasible bounds constrain the search, simplex designs supply transparent ratio grids, and randomization, independent replication, and model checks precede interaction or optimization claims. Public ASTM and ISO catalog scopes add alcohol-specific participation, method-family separation, assessor-role selection, consumer target-group and context discipline, descriptive profiling, discrimination-test limits, carryover control, vocabulary, and general sensory-method boundaries. Only public scopes and abstracts were processed; licensed full text, exact sample-size and statistical rules, formal assessor-selection criteria, standardized serving prescriptions, and whiskey-specific fatigue intervals were not accessed. This procedure is informed by those sources and does not claim ASTM or ISO conformity.

Review gate

The sensory discovery cluster is processed. Keep Lifecycle Status at In Revision until a consumer pilot has passed reconstruction, independently repeated evaluation, carryover review, and the alcohol-specific participation gate.