The Java source records the selected payoff, inspects its NablaTensor tape, then replays the compiled CPU kernel over the requested scenarios.
Default 20,000 scenario workload keeps browser runs interactive; the source uses the portable CPU engine.
This exact source runs in TeaVM. Form changes update its Java literals and reset manual edits.
import com.nablatensor.engine.AadTape;
import com.nablatensor.engine.Nabla;
import com.nablatensor.quant.EquityMarket;
import com.nablatensor.quant.Product;
import com.nablatensor.quant.Products;
import com.nablatensor.quant.TimeGrid;
public final class RecordTapeReplayRiskStudio {
private RecordTapeReplayRiskStudio() {}
public static void main(String[] args) {
EquityMarket market = EquityMarket.of()
.spot(100)
.strike(100)
.vol(20 / 100.0)
.rate(3 / 100.0)
.maturity(1)
.build();
Product<EquityMarket> product = Products.asianCall();
int steps = 32;
long scenarios = 20000L, seed = 42L;
long start = System.nanoTime();
Nabla.TypedModel<EquityMarket> model = Nabla.model(market, (rec, in) -> product.record(rec,
in, TimeGrid.uniform(steps)))
.fp64()
.greeks()
.on("cpu");
double recordSeconds = (System.nanoTime() - start) / 1.0e9;
AadTape tape = model.tape();
int active = 0;
for (int n = 0; n < tape.size(); n++) if (tape.isActive(n)) active++;
System.out.println("STAGE|1|WRITE|" + product.label() + "|" + steps);
System.out.println("STAGE|2|RECORD|" + recordSeconds);
System.out.println("STAGE|3|TAPE|" + tape.size() + "|" + tape.inputCount() + "|" + active + "|"
+ (tape.size() - active));
System.out.println("STAGE|4|BACKEND|cpu");
try (Nabla.TypedPricer<EquityMarket> pricer = model.build()) {
Nabla.TypedValuation<EquityMarket> v = pricer.value()
.with(market)
.scenarios(scenarios)
.seed(seed)
.run();
System.out.println("STAGE|5|REPLAY|" + scenarios + "|" + seed + "|" + v.seconds() + "|" + v.scenariosPerSecond()
+ "|" + v.price());
System.out.println("RESULT|" + recordSeconds + "|" + tape.size() + "|" + active + "|" + v.seconds()
+ "|" + v.scenariosPerSecond() + "|" + v.price());
}
}
}
A tape separates the structure of a valuation from the numerical market data fed into it on each run.
During recording, each active arithmetic operation becomes a node with references to its operands. Replay evaluates that fixed operation sequence efficiently; the reverse replay walks the same sequence backwards to accumulate derivatives.
Keep trade logic expressed through active values at the market-data boundary. Record once for a stable product structure, replay across scenarios, and discard or checkpoint intermediates according to the backend's memory plan.
A changed control-flow path, product shape or active-input set can require a new recording. The tape is a program representation, not a substitute for product-model governance.