MRI-002
Data / Code
Source material, derived outputs and reproducibility files.
MRI-002 uses a contemporary 2026 attention snapshot alongside a historical New York Times comments corpus. The sources, derived outputs, audits and analysis code used in the investigation are collected here.
Historical Corpus
- SourceNew York Times Comments
Aashita Kesarwani ยท Kaggle
2017-2018
CC BY-NC-SA 4.0
Source dataset →
The historical source contains more than two million New York Times comments together with article metadata. MRI-002 normalized and joined the source files locally; the original source dataset was not modified.
The complete raw historical corpus is not re-hosted here.
2026 Attention Snapshot
30 high-attention New York Times Opinion discussions from August 17-25, 2026.
297 captured Reader Picks.
Participation means displayed comment count. Recognition means the sum of recommendation counts among the captured Reader Picks.
The captured Reader Picks are not every comment and not every recommendation action in the full discussion.
- 2026 article metadataattention_articles_2026.csv
- Sanitized Reader Picksattention_reader_picks_public_2026.csv
Historical Validation
120 discussions were used to develop the procedure. The method was then frozen and applied to 813 unseen discussions.
Keep the detailed explanation on the Method / Data / Limitations page.
- Holdout summaryHoldout summary →
- Discussion-level resultsDiscussion-level results →
- MethodMethod / Data / Limitations →
Woke Lexical Audit
The audit reviewed exact word-boundary occurrences of “woke” and “wokeness,” excluding ordinary wake/sleep uses. It retained 391 categorical comment-level uses. The evidence showed mixed categorical usage from the beginning rather than a clean linear change in meaning.
- Audit summaryAudit summary →
- Monthly countsMonthly counts →
- Categorical usesNot exposed here because the table contains row-level comment text.
Analysis Code
- Analysis codeAnalysis code →