Every candidate in the package assumes dynamics="keller-miksis", and it is the default.
On the records analysed, three separate operators beat it: Gilmore/Tait on every record (+17.1, +4.2, +1.2 nats in identified coordinates at starts=10), and both enthalpy Keller–Miksis forms on two of three.
The counterweight, from the same analysis: the operator is by far the most absorbed model axis — a Keller–Miksis → Mie-Grüneisen perturbation is 77.7 % absorbed by refitting the material, leaving under two noise units. So most of an operator change can be mimicked by refitting, and the ranking slid by ~50 nats when the prior box moved. The margin also collapses once the equilibrium radius is freed.
So this is genuinely open, not obviously "switch the default": either change it, or document why a default that three operators beat is the right one to ship.
Migrated from the deleted docs/open-work.md (item 1).
Every candidate in the package assumes
dynamics="keller-miksis", and it is the default.On the records analysed, three separate operators beat it: Gilmore/Tait on every record (+17.1, +4.2, +1.2 nats in identified coordinates at
starts=10), and both enthalpy Keller–Miksis forms on two of three.The counterweight, from the same analysis: the operator is by far the most absorbed model axis — a Keller–Miksis → Mie-Grüneisen perturbation is 77.7 % absorbed by refitting the material, leaving under two noise units. So most of an operator change can be mimicked by refitting, and the ranking slid by ~50 nats when the prior box moved. The margin also collapses once the equilibrium radius is freed.
So this is genuinely open, not obviously "switch the default": either change it, or document why a default that three operators beat is the right one to ship.
Migrated from the deleted
docs/open-work.md(item 1).