Two views over recording_vectors.parquet, schema_version 21: a parallel-coordinate
view of the corpus where each line is one recording, and a recording view drawn from that
recording's compact vectors.
This page renders transcript text and detected PII in place. It reads a file you hand it
from your own disk; nothing is uploaded, fetched or stored. Open it from file://,
for your own viewing, and do not host it.
…or drop the file anywhere on this page.
recording vectors
evaluate triage
Counts over every recording in the file. A count selects its recordings as a
facet; j/k then step through them.
data dictionary
Read from this file's own footer: what each column is, how it was computed,
the code that produces it, and what null means. The i beside a column's name anywhere on
the page opens the same entry.
Null is absent, not zero. A line is broken wherever its value on an axis is null; the
dashed stubs run to the absent rail below the break, which carries a count and never a
position. A categorical axis is ordered categories, not a number line.
An m_<name> axis is a mean of m_<name>_n readings,
and its caption says how many rows fold more than one. Vectors and matrices are listed in the
axis picker but disabled, with the reason on the option.