How TraceQuant AI Scoring Works

Each signal gets an impact score from 0–100, recalculated at ingest time from four weighted components:

• Source credibility (25%) — a static score per publisher, set manually and not influenced by votes or traffic.
• Velocity (25%) — how many related signals mentioned the same top instrument in the last 60 minutes.
• Sentiment magnitude (20%) — how strongly-worded the article is, from a static keyword lexicon (see below).
• Entity weight (30%) — how central the highest-weighted asset mentioned is to the story.

Scores ≥70are flagged 🔥 as high impact across the site. Every signal card has a "Why this score?" breakdown showing the exact numbers behind it — nothing is hidden.

The public Telegram channel is deliberately stricter than the 🔥 badge on the site: an alert only fires when a signal both scores well above the high-impact threshold and is corroborated — either flagged as a statistical anomaly or confirmed by at least one other signal about the same instrument in the same hour. A single high-scoring, uncorroborated article no longer alerts on its own.

When several outlets cover the same underlying event, only the first one alerts — later signals about the same event are suppressed for a few hours instead of re-sending the same news. There is also a daily ceiling as a last-resort safety net. The goal is a channel worth keeping notifications on for, not a feed of every article that crosses a number.

Bullish/bearish tone badges start from a static positive/negative keyword lexicon scored against each headline and snippet — not a poll of readers. The result is a direction from -100 (bearish) to +100 (bullish); a badge only shows when the magnitude crosses ±20.

Keyword matching alone can misfire when a negative word appears in a neutral or even positive sentence. Every result also gets a confidence score (0–100): when confidence is low, a low-traffic AI model re-checks the headline instead of the raw keyword count being trusted outright, and if confidence is still low the badge reads "Mixed/Unclear" rather than forcing a bullish or bearish label.

This describes the article's language, not a trading signal.

TraceQuant AI compares how often an instrument is mentioned right now against its own historical baseline (a Poisson-rate approximation), flagging a statistical spike with ⚡ when it clears a threshold. This needs a few days of accumulated history per instrument before it activates — it deliberately stays quiet rather than guess with too little data.

Market Mood blends two independently-sourced signals into one 0–100 score: the external Crypto Fear & Greed Index (60% weight) and TraceQuant AI's own average news sentiment direction over the trailing 24 hours (40% weight). If there isn't enough recent news to compute a sentiment average, it falls back to Fear & Greed alone rather than inventing a number for the missing part.

Several crypto news aggregators score articles by letting readers vote them up or down. That works once a platform already has a large, active voting base — but it also means the score reflects crowd mood and can be gamed by coordinated voting, and a new platform copying that model produces meaningless scores until it has one. TraceQuant AI's scores are computed the same way for the first article it ever ingests as for the millionth — every input is objective and every score is traceable back to it.

• News: public RSS across three categories — crypto, US stocks, and forex & metals (including primary sources like central bank press releases for forex & metals). Each signal's "Why this score?" breakdown names its exact source and credibility score.
• Fear & Greed Index: alternative.me.
• Prices: DEX pairs via DexScreener (canonical WBTC/WETH contract addresses — may vary slightly from centralized exchange spot price).
• Bitcoin network data: mempool.space.