What the evidence can support.

Source-field agreement is not independent classification precision. Neither measures laboratory closure or purchase intent.

One actual bundled denominator

The bundled public snapshot contains 3671 rows. Its SHA256 is 6ced10c6382761b50f35c976ad4164425be42e9a0b8bdf210f9641eb60c1163e. These counts are recomputed from that exact file, not taken from a separate summary.

The older summary claimed 3,705 open events while its public file contains 3,671 rows. This page does not silently equate the two. The excluded or missing 34 rows have not been reconciled here; the actual public-file denominator above is the only one used for this bundle. A later validated publication must carry its own manifest and hash.

No independent precision percentage is established

The legacy detector's resolved labels and source-field checks are not an independently adjudicated classification study. No 99.5% precision claim is made. A defensible evaluation needs a fixed sample, unit, dates, denominator, independent ground truth, unresolved cases and confidence intervals.

Unknowns are part of the record

The legacy bundle lacks engine event IDs, per-record observation cutoffs, detector-rule versions and independently accepted classification evidence. Its generation label is not a successful source check. Missing values stay Unknown rather than becoming zero or a fresh date.

Read the source definitions and method. Use a private correction case to report a specific cited record.