Coming from Vensim

Naga integrates stocks and flows on a fixed time step, with the functions a Vensim model uses most, and reads a .mdl file for a common subset of the language. It has no sketch: a model is a JSON file, and its equations are expressions in it.

Import a model

naga sdimport model.mdl -o model.json
naga run model.json

What the reader translates and what it reports and leaves out is listed under Import from XMILE or Vensim. In short: INTEG stocks whose rate is a flow name or a sum and difference of flow names, WITH LOOKUP, DELAY1, DELAY3 and their I forms, subscript ranges, subranges and mappings, per-element equations and SUM, PROD, VMIN, VMAX. A variable name with spaces is read with underscores. A macro, a GET XLS or GET DIRECT reference, or a function the reader does not know is reported with the variable's name and left out; the rest of the model is still written. Read the report before you trust the result.

What maps to what

Vensim	Here
Level (INTEG)	an entry in levels
Rate	an entry in flows, with one source and one target
Auxiliary, constant	an entry in auxiliaries, read as AUX("name")
Lookup	an entry in tables, read as TABLE("name", x)
DELAY1, DELAY3, DELAY N, SMOOTH, SMOOTH3, DELAY FIXED	DELAY1, DELAY3, DELAYN, SMOOTH, SMOOTH3, DELAYP
STEP, PULSE, RAMP, IF THEN ELSE	STEP, PULSE, RAMP, IF
Subscripts	subscripts on levels, flows and auxiliaries, with SUM and MEAN over them
Units check	unit checking of equations, reported as problems
TIME STEP, integration type	simulation.sd: dt and euler or rk4
Reality Check	reality_checks in the model, run by naga check
Optimisation for calibration	naga calibrate, to KPIs or to a time series
Sensitivity (Monte Carlo)	naga doe with a Latin hypercube design
Loops tool	naga loops: which feedback loop dominates, and when

The function names and arities are in the expression reference.

What is different or missing

• No diagram. No sketch, no causal tracing, no graphs on the page. Results are frames; the HTML report draws time series.
• A flow has two ends. One source and one target, either of which may be outside the model. An equation that drains two stocks through one rate is refused.
• Not translated by the reader: a level whose rate is a compound expression rather than flow names, macros, external data functions, subscript equivalences (<->), and any function the engine does not have (the allocation functions, for example). Each is reported by name.
• Only Euler and fourth-order Runge-Kutta, on a fixed step.
• No Markov chain Monte Carlo and no Kalman filtering.

What you gain

Discrete entities, resources and agents in the same model as the stocks, on the same clock: a level can be drawn down by entities passing a TRANSFER block, and an agent's action can move an amount between levels. The bundled sd_*.json models are the classic structures, with their numbers in the gallery.