Orphaned lines: detect and mitigate typographic widows/orphans across slide elements

[tmustier]

Orphaned lines: detect and mitigate typographic widows/orphans across slide elements

Problem

Generated slides frequently produce "orphaned" lines — a single short word or fragment dangling on the last line of a text block. This is a well-known typographic problem (widows and orphans) that makes slides look unpolished and wastes vertical space.

Examples from a real deck build:

• A title wrapping to 3 lines where the third line is a single word ("visual.")
• A cell where the last line contains only "slide" or "needed"
• A subtitle that wraps with just a preposition on the second line

Context from other domains

Publishing / LaTeX: Penalises bad breaks with \widowpenalty, \clubpenalty, \looseness. The typesetter adjusts spacing or reflows to avoid orphans. LaTeX also has \parfillskip control for minimum last-line length.

CSS / web: text-wrap: balance redistributes text evenly across lines (useful for headings). text-wrap: pretty specifically targets orphans on the last line. Both are increasingly supported and address the same visual problem.

Slide style guides (McKinsey, BCG): Typically say "reword to avoid" — which is what a human would do, and what an agent could be prompted to do if the tool detects the issue.

Proposed direction

This likely needs different treatment per element type:

Element	Tolerance	Approach
Titles	Zero — should never orphan	Reflow detection + agent guidance to reword
Subtitles	Low	Same as titles
Row headers	Low — visually prominent	Detect, suggest rewording
Column headers	Medium — usually short	Detect only
Body cells	Higher — more text, more tolerance	Detect when < N chars on final line
Sidebar paragraphs	Medium	Detect

Detection (lint/verify step)

A new check in pptx verify (or a dedicated pptx lint) that:

1. Estimates line breaks for each text element given its container width and font size
2. Flags elements where the last line is below a threshold (e.g. < 15% of available width, or < N characters)
3. Reports severity based on element type (error for titles, warning for body cells)

Mitigation options (research needed)

Several approaches worth investigating — not mutually exclusive:

• Balanced reflow: Like CSS text-wrap: balance, redistribute text so lines are roughly equal length. Could be done at the text-box level by the layout engine. Most impactful for titles and short text.
• Agent guidance: When orphans are detected, emit a structured suggestion ("Title wraps to 3 lines with 1 word on line 3 — consider rewording to fit 2 lines"). The agent can then reword.
• Minimum last-line threshold: During layout solving, if the last line of a paragraph is below a threshold, slightly adjust the available width or font size to trigger a reflow. This is closer to the LaTeX approach.
• Soft hyphenation / break hints: Less relevant for slides (hyphenation looks bad at large font sizes), but break hints could help the engine make better choices.

Research needed

• Survey how CSS text-wrap: balance is implemented (the algorithm) — is it applicable to fixed-width text boxes?
• Check if python-pptx exposes any text-fit or autofit controls that could help
• Look at how professional slide tools (think-cell, etc.) handle this
• Determine the right thresholds per element type — what "feels" orphaned at 14pt vs 24pt?

Acceptance criteria

• pptx verify (or similar) detects orphaned lines across all text elements
• Severity is element-type-aware (stricter for titles, lenient for body)
• At least one mitigation approach implemented for titles (balanced reflow or agent guidance)
• Body cell orphan detection available as a warning