Re-geocodes each address with the BC service and reports how far its answer sits from the one already obtained, in metres. It is the way to answer a question [geocode()] cannot answer about itself – whether a match is right at all – and where the two differ, **BC's answer is the more reliable**: it is a parcel-level provincial authority, while NAR is a national compilation of what provinces and municipalities supplied.
Value
`g` with `bc_match_method`, `bc_score`, `bc_precision` and `bc_dist_m` appended. `bc_dist_m` is `NA` where either side has no point.
What a disagreement does and does not prove
The two sources are **not independent** – BC's geocoder and NAR's BC records plausibly share upstream data – so `bc_dist_m` is not a benchmark of NAR's accuracy. Small distances can be two views of one underlying record agreeing with itself, and the distribution is a lower bound on how far apart genuinely independent sources would sit. Use it to find suspect rows, which it does well, rather than to estimate the error `uncertainty_m` excludes.
Rows outside British Columbia are skipped rather than sent, since the service does not cover them and would answer about a BC place of the same name.
