To find comments on a specific document, use comment_on_id with the
hex objectId from document detail/search results (attributes.objectId,
e.g. '0900006486531e6b'). The human-readable documentId is REJECTED
here because the API silently returns 0 comments for it.

docket_id filters comments to all documents in a docket.

search_term is a full-text match, including attachment text; results
follow sort, not relevance. Quote a phrase ('"Professional Services
Council"') to match it exactly.

posted_comments (= meta.totalElements) counts comments posted publicly,
not comments received; mass-mail duplicates and unposted comments are
not in it, so the agency's received count can be much higher.

Search rows carry no organization (the API leaves it out), and many
titles are only "Comment on FR Doc # ...". For "who commented" questions
set include_organization=True (page_size up to 25): each row gets the
submitter's organization from its comment detail. Rows without one were
filed by individuals or the agency hid the field.

Page size: 5-100; page_number 1-40. Comments sorted by '-postedDate' by
default. The API serves 40 pages; when more comments match,
page_limit_note (and truncated=true on page 40) says how many are out of
reach. Split the query into posted_date_ge/le windows. Most comment text is in attachments; get_comment_detail with
include_attachments=True returns their download URLs.