How Kinetiq Brief works

1. Relevance ranking

Each news item receives a transparent score from nine weighted components. The components are stored with every selection, so the system can always explain why an item appeared in your Brief. Position weight can never fully suppress a severe event affecting a small position — high-materiality events on owned assets have a score floor. The default weights are shown below (portfolio-type affinity is 0 by default and activates when you pick a portfolio type — see the next paragraphs).

Source quality follows a fixed hierarchy: regulatory filings, then company investor relations, government and central-bank sources, exchange announcements, reputable financial news, trade publications, secondary commentary, and finally social media — which is never treated as confirmed fact without corroboration.

Four further adjustments shape the final ranking — all deterministic, none of which change facts or add advice:

• Portfolio type

  The type you choose (long-term, income, swing, short-term…) tilts these weights and boosts a set of news categories that reader cares about (the "portfolio-type affinity" component) — so which stories reach the top adapts to how you invest. Neutral types leave ranking unchanged.

• Per-holding diversity

  A cap on how many top stories can come from a single holding, so one prolific feed (crypto especially) can't crowd out news about the rest of your portfolio.

• Crypto-noise damping

  Low-materiality crypto-only items (price predictions, influencer takes) are down-weighted; genuine, material crypto news is not.

• Macro backdrop

  Macro stories (see §2) are ranked separately, by how exposed your portfolio is to each theme, and shown in their own section rather than competing with holding-specific news.

2. Where the news comes from (and what's still missing)

Kinetiq pulls news two ways. For every asset you hold, it fetches that asset's own news feed. Separately, it always pulls a small set of macro feeds— interest rates, the US dollar, oil, broad equities, market volatility and gold — which populate the "Market backdrop" section regardless of what you own. The AI never goes looking for news on its own; it only classifies and explains the items these feeds return, always from a cited source.

We believe in being straight about the limits of the current beta, which runs on free, publicly-syndicated feeds:

• Thin European coverage

  European-listed funds (tickers ending in .AS, .DE, .L) return little or no news on the free feeds we use in beta. As a stopgap we source news from a US-listed equivalent (e.g. a MSCI World or S&P 500 fund) so the holding isn't silent — but it's a bridge, not the final answer.

• Interests you don't own

  An Interest you follow but don't own yet now pulls in its own stories, sourced from a representative fund for that theme (e.g. a clean-energy or cybersecurity ETF) and shown in "Watchlist & interests", clearly marked as something you don't hold. The proxy is a bridge on free feeds, not a licensed thematic feed.

What's coming.Before Kinetiq leaves beta and charges for anything, we're moving to a licensed news and fundamentals feed with proper European coverage and news sourced by theme (so an Interest you follow surfaces stories even when you don't own it). The current gaps are a consequence of building the beta on free data — not the intended destination. (Kinetiq is a weekly look-back that explains what happened, so we deliberately don't run a forward economic-events calendar — that's a trader's tool, not ours.)

3. Portfolio calculations

Metrics that need history you don't have yet (volatility, drawdown, correlation, risk contribution) are shown as unavailable — never estimated from insufficient data.

Risk Radar attention thresholds

The Risk Radar in your Brief marks each concentration bar with an "attention threshold" — the level where the bar turns amber to say "worth a second look." These are Kinetiq's default awareness levels: common rules of thumb, not regulatory limits and not a recommendation to act.They're a starting convention, and you can set your own levels in Settings → Risk Radar thresholds.

Crossing a threshold is never presented as good or bad, and never triggers advice — it only surfaces the number so you can decide what, if anything, it means for you.

4. Risk Lab position sizing

risk budget        = account equity × risk %
stop distance      = |entry − stop|
per-unit risk      = stop distance + est. slippage + est. fees
risk quantity      = risk budget ÷ per-unit risk (FX-converted)
final quantity     = min(risk qty, allocation cap, cash cap),
                     rounded DOWN to instrument precision
planned stop loss  = final quantity × per-unit risk
1R                 = stop distance;  target at nR = entry ± n × 1R
break-even winrate = 1 ÷ (1 + reward-to-risk), before costs

The calculator always names which constraint determined the final size. Volatility-based stop comparison uses ATR with Wilder smoothing over 14 periods. Planned risk is not guaranteed: gaps and slippage can produce larger real losses, and a stop order may not execute at the stop price.

5. Kinetiq Pulse

Pulse keeps equity and crypto regimes separate — they are different markets and are never combined into one pseudo-scientific score. Each component shows its score, direction, inputs, last update and limitations. Components without licensed underlying data are labeled as demonstration data and excluded from any live claim. The current component set (trend, breadth, volatility structure for equities; trend, dominance, funding and leverage conditions for crypto) will activate as licensed data providers are connected, and this page will be versioned accordingly.

6. AI grounding rules

• ·The model only sees retrieved article evidence, deterministic portfolio metrics, your theses and the product methodology.
• ·Every external factual statement requires a citation; briefs are verified before publication and items failing checks are dropped.
• ·The model never calculates portfolio numbers, never predicts prices, and never produces buy/sell/hold instructions — such output is rejected by a language filter.
• ·Article content is treated as untrusted data: instructions embedded in sources are ignored by design and covered by tests.
• ·When evidence is weak or sources disagree, the Brief says so and lowers its stated confidence.