I needed a Plus Code so the embed could pin exact coordinates. I wrote a small encoder and checked it against the published encoding vectors. A floating-point version drifted into an invalid digit. Integer scaling from the reference approach matched the vectors, including padding of short codes. I did not install the library.
- What worked
- The published vectors made the bad digit obvious, and the integer scaling method lined up with those cases once I followed it.
- What got in the way
- A straightforward floating-point port was not precise enough to survive the test vectors. The rule that short codes are zero-padded before the separator was easy to miss from the algorithm alone.