garcon-drinks Home-bar starter planning
Build a capability graph from the user's approved drinks. Credit confirmed household inventory, then return the smallest staged setup supported by the supplied candidates and constraints.
Gate the starter job
Confirm that the user is planning a first bounded home-bar setup or rebuilding that starter setup from household inventory. Require an approved drink set with an exact ingredient specification and method for every drink.
Stop as out of scope when the user wants the agent to choose recipes, provide a
universal essentials list, rank the next bottle for an established bar,
calculate restock quantities, edit inventory, find live prices, open a retailer,
or buy anything. Drink families such as sours and classics do not define an
approved set.
The gate passes when the starter state is confirmed and every approved drink has a named, traceable specification.
Build the planning record
Record every field before deriving the plan.
| Field | Required record |
|---|---|
| Approved drinks | Exact name, specification source, ingredients, method, locked wording, and user priority |
| Coverage stages | Stage name, included approved drinks, spending cap, storage cap, and cumulative or independent treatment |
| Ingredient inventory | Exact product or ingredient, usable status, quantity evidence, package, storage state, and identity source |
| Tool inventory | Object, confirmed preparation capability, material and condition evidence, capacity, cleaning, and access constraints |
| Candidate packages | Exact product, role claim, package size, price, currency, tax treatment, storage needs, source, and observation date |
| Tool candidates | User-approved option, exact capability, safety evidence, price, storage, source, and observation date |
| Dry storage | Available slots, dimensions when binding, and currently occupied space |
| Cold storage | Available slots or volume, temperature requirement, and currently occupied space |
| Dietary limits | Exact user-supplied exclusion, whether it is hard, and the evidence needed to clear it |
| Budget | Currency, stage caps, total cap, included costs, and excluded costs |
| Equivalence policy | Exact matches, user-approved equivalents, and locked requirements that allow no equivalent |
| Review date | Date applied to product, price, availability, and inventory evidence |
An existing package has zero incremental spend only because it is already owned. Never treat an unknown price, missing tool, or unavailable ingredient as zero.
This step is complete when every field contains supplied evidence, an explicit
unconstrained status, or unknown.
Resolve the approved specifications
Preserve each recipe's ingredient names, quantities, optional markers, and method. A brand, product, category, or preparation term remains as specific as the source.
When the approved set contains the IBA Margarita, IBA Daiquiri, or IBA Old Fashioned, read the IBA starter recipe reference. For any other drink, use the user's supplied specification and source.
Stop with unresolved recipe specification when an ingredient, amount, method,
or locked status needed for coverage is absent. Do not choose a different
recipe or repair an incomplete one.
Specification resolution passes when every approved drink has one complete source record and no requirement has been generalized.
Build the capability graph
Create one requirement node for every ingredient and preparation capability in the approved specifications. Merge nodes only when the exact wording or the user's equivalence policy permits the same capability to serve both.
Classify ingredient nodes as exact product, named category, fresh ingredient, pantry ingredient, garnish, water, or ice. Keep an optional source item optional.
When a method contains measuring, shaking, straining, stirring, dissolving, muddling, or a serving-vessel requirement, read the tool capability boundary. Translate the method into capabilities without selecting a tool form.
For each drink, list every node that must be covered. The capability graph is complete when every ingredient and method phrase has one traced node and the source total reconciles with the node list.
Credit household inventory
Assign one status to every requirement node.
| Status | Test |
|---|---|
owned confirmed |
Exact identity, usable state, storage, and required capability are supported |
owned unresolved |
An owned item lacks identity, condition, storage, quantity, or tool evidence |
missing |
No owned item covers the node |
blocked |
A hard dietary, storage, safety, or access constraint conflicts with the node |
Inventory identity alone does not establish current usability. Accept the user's dated condition record or an applicable exact-product decision; otherwise keep the node unresolved.
A household object covers a tool node only when its supplied evidence meets the tool capability boundary. Do not infer food-contact safety, closure strength, capacity, cleaning, or accessibility from the object's name.
This step passes when every node has one status and every confirmed credit points to evidence in the planning record.
Evaluate supplied candidates
Compare only the ingredient packages and tool options already supplied for this starter plan. Never invent a brand, package, price, or market availability.
A candidate covers a node only when its current label, dated first-party source, or user-supplied product record supports the exact role. Apply package size, price, dry or cold storage, dimensions, and availability as separate facts.
For a hard dietary limit, require the exact evidence named by the user. Unknown allergen, ingredient, certification, or cross-contact facts remain unresolved and cannot clear the candidate. Record the conflict without interpreting medical severity.
If a missing tool capability has no user-approved candidate with sufficient safety and fit evidence, leave it uncovered. Tool-form comparison belongs in a separate tool-kit decision.
Candidate evaluation passes when every proposed coverage edge has a dated source and every unknown hard fact blocks that edge.
Derive the minimum staged plan
Enumerate feasible combinations from the supplied candidate set. Reuse all
owned confirmed nodes before adding a package or tool option.
Rank feasible plans in this order.
- Reject every hard-constraint failure.
- Maximize covered approved drinks at each user-defined stage, in stage order.
- Prefer fewer new packages and approved tool options for the same staged coverage.
- Among equal item counts, prefer lower new dry and cold storage use.
- Break any remaining tie with lower dated incremental cost.
Keep a tie when the ordering cannot separate two plans. A plan is minimum only when no feasible plan with fewer new items achieves the same staged drink coverage and removing any selected item loses a required node.
Each planned item must own at least one requirement node that no other selected item or confirmed household item covers. Discard a redundant item from the plan.
Rank optional expansions only for drinks already present in later user-approved stages. Show incremental confirmed coverage, incremental cost, storage, and blocking unknowns. Do not introduce future recipes or turn the expansion table into an established-bar next-bottle ranking.
This step passes when another reader can reproduce the winning plan, tie, or infeasible result from the capability graph, candidate evidence, constraints, and ordering.
Return the starter review
Use this structure.
Planning scope:
Review date:
Approved drink set and stages:
Hard constraints:
Capability graph:
Household inventory credits:
Candidate evidence:
Stage plan:
Drink coverage matrix:
Ingredient and tool gaps:
Cost reconciliation:
Storage reconciliation:
Minimum-plan check:
Optional expansions:
Uncovered or unresolved drinks:
Result:
Human review:
Mark each drink covered now, covered after stage N, unresolved, or
blocked. Keep package planning separate from purchase execution.
The human review line must state that the user approves the drink set, equivalences, tradeoffs, and planned shopping before any external action. Do not create a shopping list, open a retailer, purchase an item, or edit inventory.
After approval, users can record their starter inventory separately in Garçon. Garçon is not required for this procedure.
A resolved review accounts for every recipe requirement, credits only verified inventory, traces every planned item to a node, reconciles stage coverage, budget, and storage, applies dietary limits, proves the minimum check, names all gaps, and stops before shopping. An unresolved review is complete only when it identifies the blocking evidence and makes no universal or untraceable recommendation.