CFDL

Getting Started

Ten minutes from nothing to a running model with a probability distribution around its NPV. No install needed for the first half.

Try it in the browser

Open the Playground and paste this model:

version 0.1
model "first-model"
time calendar monthly from 2026-01 for 24

entity legal company

assume growth ~ Normal(mean=0.02, stdev=0.01, clip=[0.0, 0.05])

stream legal.revenue on entity legal.company inflow currency USD {
  schedule every monthly from 2026-01 to 2027-12
  amount = 10000 * pow(1 + inputs.growth, time.t / 12.0)
}
Open in playground

Hit Run. The compiler and engine execute entirely in your browser.

What each line does:

  • time calendar monthly from 2026-01 for 24 — every stream lands on this 24-month grid.
  • entity legal company — the thing that owns the cash flow.
  • assume growth ~ Normal(...) — a stochastic assumption: a growth rate drawn from a (clipped) normal distribution. Swap ~ Normal(...) for = 0.02 and it becomes a plain constant.
  • stream ... { schedule ... amount = ... } — a monthly revenue stream whose amount is an ordinary formula; inputs.growth reads the assumption and time.t is the period index.

Run it with the CLI

Install the CLI (see Install the CLI — from a release binary, or cargo build -p cfdl-cli from a checkout). Put the model in first-model/model.cfdl, then:

cfdl compile first-model --out first-model/ir.json
cfdl run first-model/ir.json --out first-model/results.json --rate 0.08

results.json now holds the deterministic answer — per-stream cash flows plus core metrics. For this model: model.npv ≈ 227,331 USD at an 8% annual discount rate.

Add the distribution

Create first-model/run.json:

{
  "deterministic": { "annual_discount_rate": 0.08 },
  "monte_carlo": { "trial_count": 500, "seed": 42 }
}
cfdl run first-model/ir.json --config first-model/run.json --out first-model/results.json

The monte_carlo section of the results now summarizes every metric across 500 seeded trials — mean, stdev, min/max, and percentiles. Same model file, point estimate and the distribution around it. Because the seed is explicit and every assumption has its own draw stream, this output is byte-reproducible on any machine.

What you just produced

  • ir.json — the canonical compiled model (IR schema); commit it, diff it, hash it.
  • results.json — cash flows + metrics (Results schema); everything downstream (pandas, dashboards, reports) reads this.

Where to go next

Modeling a deal (analyst path):

  1. Work through the Language Guide and the five progressive tutorial examples.
  2. Pick your domain's pack guide — energy, CRE, credit, or opco — and start from its quickstart.
  3. Set up VS Code for diagnostics and hover docs.

Integrating CFDL (developer path):

  1. Install the Python SDK and open the notebooks.
  2. Or run the API server and hit POST /v1/run.
  3. Read How CFDL Works for the IR/Results contract.