SybilGambleyyu/presentation-change-assurance-benchmark

Presentation Change Assurance Benchmark (PCAB)

This is the dataset mirror for PCAB 0.2.0, an open, deterministic corpus for tools that review stored PowerPoint OOXML changes.

What is in the dataset

• —12 paired baseline/candidate PresentationML packages.
• —One public, target-free truth.json per pair.
• —A root manifest.jsonl catalogue.

Every pair has fixed visible DrawingML text and exactly one changed ZIP member. The pairs cover click, hover, and mouse-over action changes; external hyperlink retargeting; macro, program, file, presentation, OLE-verb, media, navigation, and reserved actions; missing relationship bindings; opaque VBA-project bytes; and opaque embedded-object bytes.

The packages are synthetic and generated from source. They are not recovered business decks. example.invalid targets are non-routable. VBA and embedded members are inert marker bytes, not executable samples or real OLE documents.

Intended use

Use PCAB to evaluate a static presentation-diff, review gate, or analyzer:

1. 1.Inspect each baseline/candidate pair locally.
2. 2.Emit an observation report using the PCAB protocol.
3. 3.Score it with the accompanying pcab CLI.

The corpus records stored package facts only. It does not support claims about target reachability, macro validity or execution, OLE activation, media playback, PowerPoint rendering, trust decisions, exploitability, or runtime Office behavior.

Provenance and reproducibility

The canonical source, validator, scorer, checksums, and release assets are at `SybilGambleyyu/presentation-change-benchmark`. Run pcab validate --fixtures fixtures from the source checkout to verify archive integrity, deterministic generation, package-member boundaries, stable visible text, private action/relationship shape, and public truth without opening a presentation client or following any target.

The dataset is MIT licensed. Please cite the canonical GitHub release and pin the revision used in an evaluation.