Skip to main content
The 1.39 simulation runner executes typed commerce and replenishment scenarios through the real embedded engine in fresh private stores. It does not read your live store, contact suppliers, or transfer funds. Results are conditional projections from supplied assumptions. This surface is different from legacy A2A demos, which have separate --scenario/--live options.

​
Discover the contract

The core MCP profile includes six tools: The JavaScript entrypoint is @stateset/cli/simulation; discovery and simulation tools are read-permission tools because business mutations are confined to private episode stores. They allocate local resources, so read permission does not mean zero cost.

​
Run a replenishment plan

Install @stateset/cli@1.39.0, save this as inventory-plan.json, then run the command below:
Inspect success, run outcomes and reconciliation, not only the process exit. Reports retain per-SKU daily inventory, first unmet-demand stockout, daily cash, and first minimum cash. An ending zero balance alone is not a stockout. Independent replay checks physical availability before sales, receipt timing, order sizing and final native balances.

​
State the assumptions

  • Demand can be seeded ranges/history or an explicit dailyUnits schedule of exactly days entries. Supplied seeds do not alter explicit demand schedules.
  • Supplier lead-time/first-fill schedules use order-placement days 0 through days and therefore need days + 1 entries. Split/delayed receipts retain prepaid inbound value.
  • Receipts precede sales, which precede new purchasing. Day-zero purchasing counts toward cash.
  • Minimum order sizes, case packs and inventory-position ceilings apply before whole-order cash constraints. cashReserve withholds shared spending capacity; blocked orders are not shrunk.
  • reviewIntervalDays defaults to daily. Reviews occur on day zero and interval multiples; deferred attempts make no PO or cash movement, and quantities are recomputed at the next review.

​
Compare policies and goals

Experiments use two to eight distinct shared seeds with no per-policy seed. Compare exact financial deltas and comparisonToBaseline.changedConditions: opening capital/stock, prices/costs, purchase priority, reserve, horizon and realized paths may differ. Path-diversity counts expose repeated trajectories; different seeds do not prove independence. Money means are exact numerator/denominator fractions. Fill rate is exact sold/demanded; zero demand is null, and service goals are not evaluable. Goals include overall/per-product fill rates, ending/minimum cash, gross margin and purchasing spend. All goals must pass on the same seed. A goal miss is a valid result and need not make execution fail. A minimum-cash goal evaluates policy; cashReserve changes spending behavior.

​
Limits and interpretation

UUIDs and timestamps are not deterministic; aliases and financial outcomes are normalized. Episodes may span native transactions: inspect state after refusals. Empirical percentiles summarize selected runs, not calibrated confidence intervals. Taxes, fees, holding costs, permanent supplier loss and automatic seasonal fitting are outside the reviewed model. Use governed execution to turn a reviewed decision into an authorized live operation; a simulation result grants no production authority.
Last modified on October 20, 2026