v0.20.0

Added

• kac export carries the corpora your corpus consumes. Every type a producer exported travels, so a consumer
  receives types it never adopted and a citation into them resolves. Their parts merge into one flat file per type,
  and their records are filed under the shortcode of the corpus that wrote them. An inherited line carries that
  shortcode on id, on record and on every seeAlso value, and again under shortcode; a line with none is your
  own. The manifest gains sources, one entry per corpus inherited, each holding the publishing block its producer
  wrote, because a record of theirs is read at their commit in their repository. kac pack seals all of it, so a
  third corpus inherits the chain.
• kac export refuses rather than writing a hole. It stops with the reason and no files where a declared import
  has not been restored, where a consumed corpus is at an export format this build does not read, where two corpora
  export one type at different shapes or section fidelities, and where one corpus arrives twice at two versions.

Changed

• The breadcrumb names the corpus each count belongs to. A merged export holds several corpora's records under one
  type, so a type now gets a line for your own records and another for each corpus you consume, reading
  standards (from eng). 37 entries across 12 records: …. A type you wrote none of gets no line under your own name.
  The closing line warns against answering from memory rather than naming what a word means, because a corpus may ship
  any of the three lookup skills or none of them.

• glossary-lookup, policy-lookup and standards-lookup read sources. Each builds a link and a fetch from the
  publishing block of the corpus that wrote the line, reached through the line's shortcode, rather than from yours.
  Each says what a prefixed id means and where an inherited record's file sits. standards-lookup states that one file
  already holds the whole union, and reads what stayed behind off types in the manifest rather than naming a fixed
  four. kac update brings the three down.

• A type's manifest entry names two more of its part line's keys. idKey and seeAlsoKey join recordKey and
  partKey, so a corpus merging that type stamps the keys the producing type actually named. formatVersion moves to
  4, so run kac export again before kac bundle, which refuses an export built to another shape.

• kac validate reports a Covers line that names nothing. mirrors-citations now reports a labelled footnote
  gathering no citation the field could carry, against the line. A line naming only ids of types the field does not
  point at gathers nothing either. A section carrying no line at all stays silent, which is how a rule discharging no
  clause is written.

• kac validate reports a Covers line a space left out of italic. Markdown will not read an emphasis mark with a
  space against it, so _**Covers:** [pol-SCRT].EMBED _ is not the form and the marks reach the page.
  mirrors-citations reports the line, and its citations count as before, so a standard covering six clauses gets one
  finding rather than six against its frontmatter. A line that is bold alone is still the labelled prose form and is
  passed over.

• kac validate tells a Covers line naming a record whole to name the part. Where the field declares
  part-required:, mirrors-citations reports the line in the words ref-resolves uses, naming the target type's own
  word for a part: this 'Covers' line names 'pol-EVER' whole, and 'implements' names a clause. It used to report that
  the field did not list the id, which sent an author to put a bare policy id there and meet ref-resolves refusing it
  on the next run. A field that admits a bare id keeps the message it had.