Break the deck
Contribute slides that force real Office artifacts to prove themselves: charts, tables, masters, fields, typography, and geometry.
Browse benchmark files →Generative Layout & OOXML Scoring System · open source
The slide benchmark should be built in the open.
Gloss is a public effort to measure whether generated PowerPoint files are actually native, editable Office artifacts. The corpus, grader, and release evidence live in the repository—where they can be challenged and improved.
01## The useful work is waiting for review
02+ Add an adversarial slide case
03+ Tighten an OOXML assertion
04+ Reproduce a rendering failure
05+ Review release evidence
06## Submit a pull request with receipts
Where to start
Pick a thread. Leave the benchmark stronger.
Make scoring stricter
Inspect the local Python grader, add a test, or improve an assertion. A score only matters if another person can reproduce it.
Read the grader →Review the evidence
Help define the conditions under which a public result deserves to be called verified—not merely generated.
Read the open spec →First contribution
Start with an issue. End with a reproducible change.
- 1Read the repository overview and current benchmark contracts.
- 2Find or open an issue describing the case you want to improve.
- 3Send a small pull request with tests or evidence attached.