Specification, solution and simulation of finite-horizon discrete-continuous dynamic choice models. This glossary fixes the vocabulary shared by the public API, the documentation and the engine; it is not a specification.
Subject: One unit of the population handed to a simulation or a feasibility check; a subject starts in exactly one regime at one model time coordinate. Avoid: individual, agent, row, person
Initial conditions: The subject-indexed collection of starting regime and starting states, supplied either as a mapping of arrays or as a DataFrame with one row per subject. Avoid: initial observations, initial population, starting sample
Initial states: The state part of the initial conditions, i.e. everything except the regime and the stakeholder role. Avoid: initial values, start states
Regime: A self-contained block of the model with its own states, actions, constraints and functions; a subject is in exactly one regime per period.
Action-free regime: A regime that declares no actions. Its constraints, if any, depend only on states and parameters and still bind. Avoid: absorbing regime (an absorbing regime may have actions), terminal regime
Edge: A transition from a source regime to a target regime, declared with the model's edges. Avoid: arc, link, regime transition (that is the probability of taking an edge)
Period target: A regime that a source regime may enter at the next period along an edge that period's graph retains. It is a one-step property of a source and a period. Avoid: reachable regime, reachable target
Reachable regime: A regime some subject can be in at a given period, given the regimes the subjects start in; the forward closure of period targets from the starting regimes. Avoid: period target, active regime (an active regime need not be reachable)
Regime-transition law: The law that decides which of a source regime's outgoing edges a subject takes, together with any gates, gate references and route fallbacks declared with it. It belongs to the edges, and so do its parameters; the source regime owns neither. Avoid: next-regime function, edge law
Per-target state law: A state law of a source regime that differs by target regime. It belongs to the source regime, not to the edge, even though it is evaluated only toward its target. Avoid: edge law, handoff law
Carried state:
A state a regime keeps in its state space, and so in its value function. A
regime carries a state it reads; reading the state's next-period draw counts.
It also carries a state it hands, with any law of motion including
fixed_transition, to a reachable target that carries it.
Avoid: retained variable, kept state (in user-facing text)
Transition-local draw: A next-period value of a random state (a process or a Markov state) that is drawn on an edge and consumed by the source's state laws toward that target, but not kept by the target, because the target does not carry the state. It is drawn from the source's law at the source's current value. Avoid: dropped shock, ephemeral state
Age: The subject's biological age, when the model uses an age grid as its time coordinate.
Period: One zero-based computational slot in a finite horizon, supplied directly in a period model or associated with an age in an age model. Avoid: t, time step (in user-facing text)
Economic time: The calendar date, quarter, year or other economic interval represented by a model's decisions. Several computational periods may belong to the same economic interval.
Fixed parameter: A parameter bound when the model is built and not supplied at call time.
Runtime parameter:
A parameter supplied with every solve, simulate or feasibility call, in a slot
that has not been fixed at model construction.
Avoid: free parameter, estimated parameter
Temporal parameter: A parameter whose value is labelled by model time and selected at the period of the function that consumes it.
Feasible (subject): A subject for which at least one combination of the regime's declared action-grid points satisfies every constraint jointly; in an action-free regime, a subject whose state-only constraints all hold. Avoid: valid, admissible (reserved for individual actions), allowed
Infeasible (subject): A subject that is not feasible. Infeasibility is a property of the supplied initial conditions, never of the model.
Feasibility mask:
The one-dimensional boolean array, in subject order, that is True exactly for the
feasible subjects.
Avoid: acceptance mask, validity flags
Invalid initial conditions: Initial conditions that are structurally malformed (unknown regime, missing or extra state, wrong length, off-grid age, inactive regime at that age, unknown categorical label) or that contain an infeasible subject.
Unsupported check: A feasibility question the engine cannot answer for the given model shape, e.g. a constraint that depends on an age-specialized function while subjects start at different ages. An unsupported check is a property of the model, is reported by raising, and never counts as feasible. Avoid: skipped check, unavailable check