How can it be done to carry a thinbasic-Number (the float80-type known from 16Bit-times) or TBYTE in ASM without any conversion that would slow the execution down?
I have been researching a lot since Win32 officially does no support this type which were the native, by machine language used type of handling numeric data.

I 've found that the type IS present in the win32-system, camouflaged as something else.

Where to look?
Libraries/headers that provide backward-compatibility with 16-Bit-Windows of the late 1990s.
It must be there - part of the win32-Api since Win 98SE/2000/XP.
Most of the DLLs of that time are replaced in windows by DYNLINK-types / still wearing the name "xy.dll" to secure compatibility with pre-"managed"-times software or (i guess not voluntary) independent software for windows.

Win32 comes with Float80/Extended-Type built-in - I look at it but i do not recognize it - alike looking into a forest but unable to see a tree.

The provided DOUBLE-type to use is compared to Float80 - using 16 Bits more - as using the SINGLE type instead of a DOUBLE. Multiple times converting it is a nonsense that kills all advantage of it
--- but to use ten bytes was a nonsense too when it was developed --- why didn't they use 16 ? ---

Thats why its treated like an unwanted child by the developers of windows, they predicted no future for types with 10 bytes

Instead of downgrading and following the way into lack of preciseness my thoughts are to upgrade and to

use the Float0x80 / Float128 in a future version of tb as the native Numeral-type

Strings or Byte-Arrays remain as now but Strings as now are just Byte-Array-formats, but thats another topic.