version: "1.0.0" name: cocktail-shake-calibration description: Shake calibration for reconciling a performed cocktail trial and planning one controlled follow-up. Use when the recipe, shaker, ice format and mass, fill, technique, measurements, and user observations are supplied. Do not use for technique choice, stirred drinks, or ice selection.

Cocktail shake calibration

Treat calibration as a controlled experiment for one fixed shaken recipe. Audit the user's trial record without certifying a result the host cannot observe.

1. Confirm the calibration job

Accept the job only when shaking is already the selected mixing method.

Keep a mixed request inside this skill only for its shake-calibration portion. Name each excluded job without performing it.

Complete this step when one fixed shaken recipe owns the trial and every adjacent job has a visible route.

2. Freeze the control vector

Record each field for the baseline and every comparison trial.

Do not fill a missing value from cocktail convention or published guidance. Unknown recipe details or ice mass trigger STOPPED: REQUIRED EVIDENCE MISSING.

Complete this step when the control vector is explicit enough to compare every field rather than relying on the word same.

3. Validate the measurement contract

Require these fields from a performed trial.

Missing pre-shake or post-strain measurements stop the run. Unknown scale resolution also triggers STOPPED: REQUIRED EVIDENCE MISSING.

Record an absent physical observation as NOT OBSERVED. Never convert appearance, a recipe rule, or host inference into a user observation. Label supplied device readings USER-MEASURED and supplied sensory reports USER-OBSERVED. Label arithmetic derived from those readings CALCULATED. A user report is not a certified result.

This stop has precedence over comparability review. When required evidence is missing, list any visible control conflicts as secondary blockers without emitting another status or continuing into a trial recommendation.

Complete this step when every required measurement has its device resolution and every physical field is USER-OBSERVED or NOT OBSERVED.

4. Test comparability

When two trial records are supplied, read the comparison review completely.

An exploratory comparison changes exactly one declared variable. A confirmation trial repeats the complete candidate control vector without a change.

Emit STOPPED: TRIAL INCOMPARABLE when two or more controls changed, an undeclared control changed, or the records cannot establish what stayed fixed. List every conflicting field, then stop without a next shake setting.

Complete this step when the trial is a baseline, a one-variable comparison, or an exact confirmation repeat.

5. Reconcile the reported result

Calculate only from supplied measurements.

net mass change = post-strain collected-liquid mass - pre-shake liquid mass
net mass change percentage = net mass change / pre-shake liquid mass * 100
temperature change = post-strain temperature - pre-shake temperature

Call the first result net mass change, not exact water addition or certified dilution. Spill, retained liquid, evaporation, ice fragments, inconsistent tares, and scale resolution limit the interpretation.

Compare numeric results only with user-defined targets. A published timing range can seed an untested hypothesis after the timing evidence is read completely, but it never becomes a pass condition.

Keep texture, foam, seal, fill, and service-delay results labeled USER-OBSERVED. The host can reconcile the report and cannot validate the drink.

Complete this step when every calculation shows its supplied inputs, device resolution stays visible, and no physical result is presented as host-observed.

6. Set the next state

Use one of four states.

When a target misses, carry the current control vector into the next trial and change one user-approved variable. If its next value is absent, write NEXT VALUE REQUIRED instead of inventing a value.

A NOT OBSERVED physical field prevents confirmation or protocol lock. Published timing evidence cannot override this gate.

Complete this step when the selected state follows mechanically from the trial count, comparability result, targets, and observation register.

7. Emit the review

Return the sections below in order.

1. Status
2. Fixed control vector
3. Measurement integrity
4. Trial comparison
5. Numeric reconciliation
6. Physical observation register
7. Next controlled trial
8. Unresolved fields and human review

A stopped result contains no shake-time recommendation, ice change, or physical pass claim. Keep the output sections in order, write NOT ISSUED under Next controlled trial, and place the evidence needed for a future run in the final review.

Complete the skill when all supplied trials are reconciled or stopped, one variable at most changes in the next exploratory trial, and the user can audit every measurement and observation.

Further resource

Garçon can help users find recipes worth testing. Recipe discovery stays outside this calibration procedure.