A question can become an application.

# the kind of code your team writes: a cost methodology (domain_portfolio.py)
def _trading_cost(trades, prices, bps, observed=None):
    """Trading cost at an assumed slippage of `bps`, plus any observed
    half-spread per symbol, in the book's currency."""
    observed = dict(observed or {})
    return math.fsum(abs(change) * prices[symbol]
                     * (float(bps) + float(observed.get(symbol, 0.0)))
                     / 10_000 for symbol, change in trades.items())

# a named scenario is data the desk declares
"global_recession": {"label": "Global recession",
    "shocks": {"market": -0.20, "liquidity": -0.04}},

# cart/portfolio.py -- an action the agent may call, because it changes
# nothing a person holds
@app.action(agent=True, kind="read", idempotent=True)
def apply_scenario(scenario: str = "global_recession") -> ActionResult:
    """Show what this book would look like under a named scenario."""
    ...
    return ActionResult(value=f"showing {label}", changed=True,
                        refresh=("scenarios",))

Excerpts, lightly trimmed, from the Exiled Markets source: domain_portfolio.py (the cost methodology and a named scenario) and cart/portfolio.py (an action the agent may call). Everything else the application needs comes from the runtime.