Generator Comparison Boundary

mdpr-skill may reference external PPTX generators as comparison points only. They are not dependencies, not fallback renderers, and not alternate runtimes for this repository.

MDPR remains the deterministic runtime for Markdown parsing, slide splitting, layout, validation, rendering, and editable PPTX output. The skill may use the external tools below to describe capability coverage or review vocabulary, but must not route deck generation through them.

Comparison Points Only

Tool	Useful comparison vocabulary	Boundary
PptxGenJS	PPTX object coverage, charts, tables, media, masters, and shape capabilities	Reference vocabulary only; do not add it as a runtime dependency.
python-pptx	Document inspection, placeholders, tables, notes, and low-level PPTX object model concepts	Reference vocabulary only; do not add it as a fallback renderer.
Other Markdown/PPTX generators	Template workflows, visual preview loops, and export ergonomics	Compare workflows only; do not replace MDPR validation or renderer ownership.

Allowed Uses

• Document gaps in MDPR capability language.
• Name review categories consistently, such as chart coverage, speaker notes, comments, placeholder roles, and media handling.
• Inform benchmark comparisons when explaining why MDPR needs a deterministic rule, schema, or renderer feature.
• Build evidence-only scorecards from MDPR artifacts and review findings.
• Normalize visual guidance into categories such as editability risk, decoration noise, theme fit, evidence grounding, and accessibility.

Forbidden Uses

• Adding PptxGenJS, python-pptx, or another generator as an implicit dependency.
• Treating another generator as a fallback renderer when MDPR fails validation.
• Using another generator to bypass MDPR overflow, text clipping, overline, coherence, editability, theme, or object-policy gates.
• Letting mdpr-skill choose final coordinates, colors, typography, z-order, object IDs, icon assets, or renderer objects.

External generators can help describe what good PPTX tooling supports. They do not change the runtime contract: the LLM can suggest; MDPR renders and gates.

Evidence-Based Scorecards

review-core exposes buildVisualGuidance(findings) and buildGeneratorComparisonScorecard(input) for comparison reports.

buildVisualGuidance maps existing deterministic review findings into normalized categories. It carries evidence references such as finding type, slide id, object kind, role, block ids, layout slide ids, and region ids. It must not echo raw coordinates, colors, typography, object ids, or renderer geometry back into the report.

Visual comparisons must distinguish topology from surface styling. Current MDPR evidence names horizontal card-row-3/card-row-4 geometry separately from grids and stacks, while open horizontal rows avoid counting a geometry change that still looks like the same repeated white-card treatment.

buildGeneratorComparisonScorecard compares MDPR against external generator references using measurable evidence: editable object coverage, design decision trace presence, layout validation references, overflow/density finding counts, native table/chart proof support, and whether manual review is required. Manual preference remains separate from deterministic evidence. The scorecard may say which system has stronger evidence on a dimension, but it must not claim that one output is objectively prettier or bypass MDPR validation gates.