Supplementary materials for an EMNLP 2026 Industry Track submission on evidence-grounded slide narration

Anonymous supplementary archive for an EMNLP 2026 Industry Track submission under double-blind review.

  Contents:
  - README.md with a claim-to-file map for every numeric claim in the paper
  - LICENSE (Apache 2.0)
  - data/: the 30-deck file list, per-deck gate decisions under two threshold profiles, and a precomputed expected output of the permutation script
  - code/: the 24-keyword fabrication proxy, the 15-keyword image-claim proxy, and a paired permutation test reproducing the paper's headline 7.89% vs 1.79% comparison
  - annotations/: two annotators' 4-way labels on a 100-slide inter-annotator agreement set, a 28-slide manual audit of condition-C outputs, and the rubrics shipped to the annotators
  - examples/: two qualitative before/after narration pairs and one worked sentence-to-evidence lineage example

All Python files are standard-library only and run on Python 3.10+.
 

Authorship will be added in a new version after acceptance.