calybris1.3.0Calybris Core 1.3.0 · WebAssembly, in this tab
Lab 1.1 The losers, and the futures.
Two things the published core can be asked. First: for one request, what every candidate measured and which limit stopped it. Then: what the same policy does across a quarter of a million versions of the same day.
00 · What the core can take
Measured here, on this machine
Nothing on this panel was typed in. The core runs in this tab and the page holds the stopwatch, so the numbers are yours: a phone and a desktop will disagree, and both are right.
01 · Why the others lost
Every candidate, and the limit it met
The core returns a verdict per candidate, not only a winner. A refused candidate names the gate it failed first and the two numbers that gate compared, in that gate's own unit.
···
02 · A quarter of a million futures
The same policy, under a moving world
Every trial jitters the prices, the budget, the value, the risk and the latency, then asks the core again. The trials run in Rust; this page only starts and stops the clock.
Research build · 0.7
A learner that cannot cross the line
The earlier lab is still here: a Monte-Carlo AIXI planner (Veness, Ng, Hutter, Uther and Silver, 2011), Thompson sampling over a supplier catalogue, and a fragility sweep, all gated by the Calybris kernel. The learner proposes, the core decides, and a receipt lets anyone check the run.
Open the research lab →