ORIGINAL MODEL-REVIEW RESEARCH

Trading news, ranked and explained by 31+ LLMs.

Instructor subject match · Review pending

Inspect the saved opinions, check the model identities, and see where the models disagree.

32distinct models represented
235saved model opinions
101source stories and public demo items

These results were generated on November 13, 2026. There are 33 participating model entries, including two names for the same Llama 2 manifest. Four embedding entries are not article-rating models. The alias is visible in the raw data but counts once in aggregate statistics.

Open the 32-model comparison ↗

Spread of saved model scores
Spread of saved model scores. Original Lyfter visualization; underlying observations and definitions are available in the linked data. Download SVG
Suggested reviewer: Scott Maragioglio · Pending

Use the original observations.

Download all observations (JSON) · Download all observations (CSV)

Each record identifies its source item, model entry, three scores, reasoning, scoring time and available input provenance. New articles show their actual review count, which can be zero.

What these scores do not establish

  • Historical reviews used differing excerpt lengths and interfaces.
  • The duplicate alias is shown but excluded from aggregates.
  • Scoring-time manifests, complete prompts and sampling settings were not recorded with every historical rating. Current manifest observations are not asserted as historical scoring-time versions.
  • Scores express model opinions, not fact verification, market forecasts or model-quality benchmarks.

The published observations and disclosed method allow readers to inspect our multi-model approach.

Read the rubric and publishing standards · Who publishes the research