Ignoring and Rejecting

A bounded region of an otherwise sound page MAY be left out of the output — no element, no box — only when it is:

• a scan artifact;
• an invisible source object;
• damaged content;
• an irreconcilable source error;
• an element explicitly out of scope;
• a genuinely unresolved ambiguity.

Ordinary logos, page numbers, and content that is difficult but readable MUST NOT be excluded. That is the requirement that matters. Exclusion is not a way past a hard page: a dense figure, an awkward multi-column spread and a badly kerned table are all representable, just slower. Excluding them turns a slow page into missing data, silently — the never drop failure.

A page MAY be declared unrepresentable, and left out of evaluation rather than represented in part, only when it is corrupt, badly cropped, mismatched to its document, unusable, or cannot be represented unambiguously. An unrepresentable page MUST NOT be given a partial or guessed rendering.

The test is scope. One bad region on an otherwise sound page: that region is excluded and the rest of the page represented. A page where the problem is the page: the page is declared unrepresentable. Excluding twenty regions to rescue a broken page MUST NOT be done — it is a rejection in disguise, and it produces an output that looks complete and is not.

A page with no ink at all MUST keep its position in the page sequence with zero elements. Its Markdown MUST consist of the single line <!-- blank page --> between the two --- separators that bound it, and its sidecar entry MUST list no items.

…end of page 1

---

<!-- blank page -->

---

Start of page 3…

Blank pages are ordinary in real documents — section breaks, duplex backs, deliberate spacers. A three-page file whose second page carries no ink and no text is not corrupt, and declaring it unrepresentable would misreport the document's length.

Neither exclusion fits:

• Excluding a region is for a bounded exception within a page. Here there is nothing to bound.
• Unrepresentable is for a page that cannot be represented. A blank page represents perfectly — as nothing.

The page MUST NOT be dropped — a missing page is far harder to notice than an empty one, and every later page number would shift. It MUST NOT be given an element either: an "empty" element with a box around nothing is invented. The comment says, inline and invisibly, that the page is blank on purpose, so a reader of the Markdown can tell a blank page from a lost one.

A page that prints only a notice such as "This page intentionally left blank", or only a folio or running header, has ink: it is an ordinary page whose content is that text.