Reading Catalayer AI News Analysis: Relevance Scores, Predictions, and Confidence

What You See on Each Story

When you open a news article in the Catalayer workspace with Market Intelligence enabled, you get (besides the headline + body):

1. Relevance score (0-10)
2. One-line summary (2 sentences extract)
3. Why it matters (context: why this story is material)
4. Affected assets (tickers, commodities, or sectors likely moved)
5. Predicted impact (direction + magnitude, 1-hour and 1-day)
6. Follow-up watch (what to monitor next based on this)

Each element has specific meaning and specific limitations.

Relevance Score (0-10)

What it measures

How likely this story is to cause price movement in the affected assets.

Not: "how important to you personally" (that's in Monitor rules). Not: "how interesting" (that's subjective). Exactly: "how often has a story with these characteristics been followed by meaningful price action in the next hour".

Scale interpretation

• 9-10: Rare. Earnings surprise, M&A announcement, FDA decision. Expect immediate price move.
• 7-8: Material. New contract, guidance change, major partnership. Expect moderate move.
• 5-6: Context. Industry report, analyst day summary. Minor or delayed move possible.
• 3-4: Noise. Analyst reactions, opinion pieces, recycled news. Usually no action needed.
• 0-2: Ignore. Filler content.

How it's scored

Catalayer AI scores each story on multiple factors:

• Keywords and named entities
• Source type (breaking wire vs. opinion blog)
• Time-of-day (earnings after hours weighted higher)
• Cross-reference with similar past stories and their price outcomes
• User feedback signal (are users clicking through or dismissing)

The score is NOT a crystal ball. It's a probabilistic indicator based on how similar stories have behaved.

Confidence calibration

Over 500 stories you've seen:

• If 9-10 scores consistently produced 3%+ moves, trust those
• If 9-10 scores routinely underperformed, the system is miscalibrated for your style (common with niche sectors)

Catalayer AI adapts as you click/ignore. But initial calibration takes 2-4 weeks.

One-Line Summary

2-sentence extract of what happened. Useful for:

• Triaging 50+ stories quickly (read headlines + summaries, skip bodies)
• Sharing with a team member quickly
• Confirming a story is what the headline suggests (sometimes headlines mislead)

Limits

Summary is extractive, not generative — it pulls verbatim text. If the original article is badly written or has buried the lede, the summary reflects that.

Why It Matters

A 2-3 sentence explanation of causation: not just what happened but why it moves a stock.

Example:

• Story: "TSMC reports Q1 revenue miss"
• Why it matters: "Lower TSMC revenue signals weak demand in the smartphone and PC ends of the chip market. Expected downstream impact on AAPL iPhone demand forecasts and NVDA data center capex timing."

This section is Catalayer AI's most valuable output for active traders. Headlines alone don't tell you why they matter.

When it's wrong

Sometimes the "why it matters" explanation is generic or misses the actual market interpretation. This happens when:

• News is unusual / hasn't happened before in similar pattern
• Interpretation depends on insider context (analyst day knowledge, management history)
• Multiple plausible explanations exist and market hasn't decided yet

Trust it as hypothesis, verify with price action.

Predicted Impact

Direction (up/down/mixed) + magnitude (minor/moderate/major) over 1-hour and 1-day horizons.

What "predicted" actually means

NOT: a guarantee of direction. INSTEAD: weighted average of past similar stories' price outcomes.

If 20 similar past stories moved target ticker down 2-5% in the first hour, the prediction says "down, moderate" — that means the base rate suggests that outcome, not certainty.

How to use it

• If prediction + your analysis agree: high conviction trade setup
• If prediction + your analysis disagree: think harder. Either the system is wrong or you are
• If prediction says "mixed" or very low confidence: likely uninteresting — move on

How to NOT use it

• As an auto-execute signal (you still need to think)
• As sole basis for trades (use as one input)
• As validation of speculation (predictions for low-confidence stories are low-value)

Affected Assets

List of tickers, commodities, or sectors likely moved. Catalayer AI extracts these from:

• Direct ticker mentions in story
• Named company references mapped to tickers
• Supply chain / sector implications

Where it's weak

• Obscure supply chain effects (Tier 3 suppliers)
• Complex derivative exposures (options Greeks)
• Macro spillovers (USD move → EM stocks)

Use as starting list, not comprehensive.

Follow-Up Watch

Things to watch in the next hours/days based on this story. Example:

• "Watch for Apple's official response (not yet issued)"
• "Earnings in similar names due next week likely pressured"
• "Policy makers likely to comment within 48 hours"

Acting on follow-ups

You can often create new Monitor rules directly from follow-up watch items. E.g., if story says "watch for Apple's response", create a monitor: AAPL AND (statement OR response OR official OR comment) active for next 48 hours.

Calibrating Trust Over Time

The most valuable thing you can do: track how Catalayer AI's predictions perform against reality.

Light approach

Mental note each time AI predicts a 3%+ move. Check price action 1 hour later. Note matches and misses.

Rigorous approach

Track in a spreadsheet:

• Story timestamp
• Prediction direction + magnitude
• Actual 1h, 1day price change
• Delta between predicted and actual

After 50 stories, compute accuracy by score tier:

• 9-10 predictions: what % accurate?
• 7-8 predictions: what %?

Use this to trust / distrust specific score ranges for your trading style.

Market Intelligence Tier

All of this requires Market Intelligence or All Access subscription. Free tier sees headlines only, no analysis.

The reasoning: analysis costs compute (both the AI inference and ongoing model refinement). Free tier keeps costs sustainable while analysis is a paid capability.

When AI Analysis Fails

Catalayer AI is honest about failure modes:

• Novel events — things that have no historical parallel (first ever AI company IPO, first geopolitical event of its type). AI can only score based on precedent; novel stories get low-confidence scores.
• Correlated clusters — 20 stories all about the same event. Each gets scored but collectively they represent 1 event. Use the first story, ignore repeats.
• Narrative-driven moves — stocks moving on sentiment alone (meme stocks, reddit-driven). AI scoring doesn't capture social momentum well.

In these cases, your judgment beats the AI.

FAQ

A: Click "details" on any analyzed story. You'll see the top factors that contributed to the score. Useful for understanding when to trust vs. second-guess.

A: Context changes. A story 5 minutes after breaking is "new" (high relevance). 3 hours later, it's "stale" (market already reacted, lower relevance). Scores reflect current actionability.

A: Regularly. AI is one input. Your judgment + AI + prior analysis + market context = decision. Disagreement is healthy.