Examples

Twenty-four runnable notebooks, from a first simulation to multi-supplier, perishable, and production workflows, with real forecasts from smooth. Each page here is the executed notebook; use the download button to run it yourself.

Running the notebooks

Clone the repository, install Stockcast, then pip install jupyterlab "smooth>=1.0.7" and open examples/notebooks. Notebook 04e uses smooth's simulated intervals, so its simulated targets can differ slightly between runs.

Foundations

Start here if you are new. These follow the same ideas as the Walkthrough, with more pictures.

Notebook	You will learn
01 · Inventory flow	how demand, an order, and the lead-time pipeline change the state
02 · First engine simulation	what a complete, checked run produces: event table, windows, manifest, metrics
02b · Decision schedules	periodic, delayed, one-time, and irregular ordering calendars
02c · Synthetic demand	demand from any distribution: one sampler for the panel or one per SKU
03 · Your own loop	the primitives behind the engine, and what the engine adds

Forecasts to orders

How forecasts become targets, and targets become orders.

Notebook	You will learn
04 · Weekly forecast to order	a one-week smooth forecast becomes a weekly order: one step, then a loop through time, then the simulator
04a · Forecasting from sales	how stockouts hide demand from a forecaster that learns from sales, and how to refit the forecast inside a run
04b · Daily newsvendor	price, cost, and salvage choose one daily purchase
04c · Cumulative protection target	lead time turns a forecast into a multi-day protection target
04d · Rolling cumulative targets	dated forecast snapshots refitted at every review
04e · Cumulative target methods	independent-normal, approximate, and simulated targets on the same replay
04f · Scheduled forecast simulation	an irregular schedule with a forecast for each coverage window

Extending Stockcast

Custom rules and richer operations, one building block at a time.

Notebook	You will learn
05 · Write your own ordering rule	write an \((s, Q)\) rule yourself, tune it on costs, and extend it
05b · How often to check stock	where to put the reorder point and how often to check stock when checks happen at set times, from one SKU to 100
05c · Real-world rules and disruptions	order days, whole-case and chiller-space rules, a cancelled supplier day and a stock count, added one at a time on the same demand
05d · Several suppliers and late deliveries	buy from a local roaster and a cheaper importer: random delivery times, deliveries in two parts, and a rule for who gets each order
05e · Inventory processes: expiry, discards and returns	the built-in ShelfLife process, a custom outflow (after_demand) and inflow (before_demand), and the stock balance that ties every flow together
05f · Supplier delivery outcomes: delays, shortfalls and caps	DeliveryOutcome for delayed and partial deliveries, an OrderingConstraint that caps orders, and a SupplierAllocation that sends the excess to a second supplier
05g · Reorder levels per item	the ways ReorderPointPolicy.fit accepts s and S (one pair, per-item values, a dated quantile forecast, a custom provider) and what per-item levels change

Experiments and operations

Complete studies and a production pattern.

Notebook	You will learn
06 · Controlled comparisons with run_comparison	targets from an external forecaster, one run_comparison with five branches, and the controlled contrasts it gives: forecast probability, forecast model, and policy rule
07 · Shelf life on real demand (M5)	ShelfLife with dated opening deliveries on 20 M5 items, forecast and history targets across three shelf lives, a custom waste-rate metric and the cost components
08 · Callbacks: changing orders and stock during a run	on_after_prediction and on_after_demand callbacks, the audit table, why callback order matters, and how constraints apply after callbacks
09 · A full experiment: forecasts, policies and extensions	three forecasts refitted at every review and passed as a policy_schedule on 30 M5 items; supplier rules, a callback, an unreliable supplier and a shelf life added one at a time and measured; then run_comparison of (R,S) and (s,S) with each forecast, priced with the evaluator
10 · A production job: plan, close and event triggers	a live job built up step by step: one day with every table (restore with from_observed and with_open_orders, receive, refit, predict, supplier rules, fulfill_demand), pure plan and close functions, a store with idempotent phases and an order outbox, an event stream that triggers the phases, a shadow backtest with SimulationEngine for monitoring, and a morning rebuilt from a stock count and a late delivery