UNDERSTAND WHAT A SCORE CAN TELL YOU

Review methodology.

Review the inputs, limitations, and subjective dimensions behind the saved model opinions.

01

Start at the source.

We collect titles, publication dates, links, and excerpts from official Federal Reserve and European Central Bank RSS feeds. Source outages are shown on the news desk. This is a saved demo edition, refreshed manually.

02

Compare model perspectives.

Installed text models evaluate a common title and up to the first 3,500 characters extracted from an original article on our private server. Gemma 3 1B and Qwen 3.5 2B also provide the existing reviews across this edition. The scorer fetches the source text; models cannot browse or independently verify its claims. The public cards show the publisher’s RSS summary, which can differ from the longer text used for scoring. Stories without enough source text are left unscored. Review coverage varies by article. The demo comparison shows each saved model rating.

03

Show the differences.

Each model rates relevance, education, and visible evidence from 0 to 100. The displayed dimensions are the mean of available model ratings; the overall score is the rounded mean of those three dimensions. No completed rating means no displayed score.

04

Keep your judgment.

Small language models can misunderstand context, inflate ratings, and disagree. A high score is not a fact check, an endorsement, a prediction, or a reason to trade. Read the original and consider what is missing.

What the numbers measure

DimensionLower scoresHigher scores
RelevanceUnrelated or narrow administrative newsBroad significance for investment research or economic policy
EducationAn announcement with little explanationDetailed explanation of a financial or economic concept
EvidenceAssertions with few supporting details in the excerptSpecific information and clear attribution visible in the excerpt
Lyfter is an independent research and education project. It is not affiliated with the Federal Reserve, the ECB, the SEC, or Trading Academy. The quiz measures self-reported process habits only; it does not assess investment suitability, risk capacity, or expected returns.

Reviews across the demo

Additional saved reviews cover Lyfter’s public research examples, curated SEC summaries, filing metadata, calendar entries, workshops, and tool explanations. Each review shows the exact text supplied, its length and SHA-256 hash, the model’s reasoning, and its scoring date.

For this content, relevance means usefulness to investor learning or research, education measures clarity and explanation, and evidence measures support visible in the supplied text. The news desk uses its news-focused relevance rubric. These scores do not evaluate investments, live charts, calculation correctness, private user inputs, or the full text behind a source link.

Every generation entry is included in the manual content-review run. Embedding-only models remain listed but cannot directly generate editorial ratings. A failed or missing output is displayed without a fabricated score. Reload demo reads saved snapshots; it does not launch model inference.