Two decimals for the common case (0.25, 1.00), ∞ for the extend-to-all
sentinel — but anything BELOW the two-decimal resolution is printed in
exponential form instead. This log is the one surface that would reveal an
"empty node because nothing matched" bug, and the gsplats continuous arm is a
~1e-3…1e-5 epsilon: at toFixed(2) every such value printed 0.00,
indistinguishable from Lines' true 0, from a NaN, or from a zeroed array. A
literal 0 still prints 0.00, so the distinction is visible.
One cosmetic wart, left alone: a value just under the threshold rounds across it
at one significant digit — 0.009999 prints 1.0e-2. It is still unmistakably
non-zero, which is the whole point of the branch.
Render one tolerance for the
?debugquery log.Two decimals for the common case (
0.25,1.00),∞for the extend-to-all sentinel — but anything BELOW the two-decimal resolution is printed in exponential form instead. This log is the one surface that would reveal an "empty node because nothing matched" bug, and the gsplats continuous arm is a ~1e-3…1e-5 epsilon: attoFixed(2)every such value printed0.00, indistinguishable from Lines' true0, from a NaN, or from a zeroed array. A literal0still prints0.00, so the distinction is visible.One cosmetic wart, left alone: a value just under the threshold rounds across it at one significant digit —
0.009999prints1.0e-2. It is still unmistakably non-zero, which is the whole point of the branch.