Builds per-key coverage statistics from a SAMPLE of rows that carried overflow JSON.
Each row's overflow column is a JSON object string (what M1 parked); this tallies, per key,
how many sampled rows had it non-null and collects a bounded sample of its values for type
inference. Robust to malformed/empty JSON (such rows contribute to TotalRows but no keys).
Coverage is therefore computed over the SAMPLE (bounded memory at any table size) — a
statistically sound basis for the pervasiveness decision; the exactness the threshold needs
(a key on ~half the rows vs. a junk key on one) survives sampling.
Parameters
overflowJsonStrings: string[]
the raw overflow-column value from each sampled row
sampleCap: number = DEFAULT_SAMPLE_CAP
max values retained per key (default DEFAULT_SAMPLE_CAP)
Builds per-key coverage statistics from a SAMPLE of rows that carried overflow JSON. Each row's overflow column is a JSON object string (what M1 parked); this tallies, per key, how many sampled rows had it non-null and collects a bounded sample of its values for type inference. Robust to malformed/empty JSON (such rows contribute to TotalRows but no keys).
Coverage is therefore computed over the SAMPLE (bounded memory at any table size) — a statistically sound basis for the pervasiveness decision; the exactness the threshold needs (a key on ~half the rows vs. a junk key on one) survives sampling.