Methodology

Every tracked asset gets one of three calls: Buy (news and momentum both lean positive), Sell (both lean negative), or Hold (mixed, weak, or no signal either way). Alongside the call is a confidence score from 0-100% — how strongly the underlying evidence points that direction, not a prediction of how big the move will be.

For prediction markets specifically, Buy/Sell are relabeled "YES undervalued" / "NO undervalued" since there's no literal buy-or-sell-the-company framing — the scoring underneath is identical.

None of this is a recommendation. It's a summary of what the news and the price are currently saying, computed the same way for every asset, every time — see the disclaimer.

The pipeline is deliberately simple and, right now, keyword-based — we'd rather be upfront that it's algorithmic than let it look smarter than it is:

Ingestion: RSS feeds and other news sources are polled continuously; every new article is stored once (deduplicated by URL).

Matching:each article is matched to tracked assets by ticker symbol and company/coin name. Single-character tickers (like "V") are excluded from symbol-only matching — too ambiguous against ordinary English text — and only match by full company name instead.

Sentiment:each article-asset match gets a score from -1 (very negative) to +1 (very positive), based on the presence and density of positive/negative financial keywords near the asset's mention. This is v1 — a keyword scorer, not an LLM — chosen for speed and predictability; the interface it plugs into is provider-agnostic, so a more sophisticated scorer can replace it later without changing anything downstream.

Recency weighting:a fresh article counts more than an older one. Weight halves every 24 hours (a 24h-old article counts half as much as one published just now, a 48h-old article a quarter as much, and so on), so today's headline dominates over last week's.

Price momentum (the asset's recent percent change) is normalized and blended with the recency-weighted average sentiment into one composite score:

News carries a bit more weight than momentum (60/40) since momentum alone can be noisy and doesn't explain whya price moved. When there's no recent news for an asset at all, momentum carries the full weight on its own rather than being diluted by a missing news component.

The composite score sits somewhere on a line from -1 to +1. Cross +0.15 and the call is Buy; drop below -0.15and it's Sell; anything in between is Hold. Confidence is derived from how far the composite sits from zero, with a floor for assets with no news at all (momentum-only calls are capped lower, since there's less evidence behind them) and a ceiling of 98% (nothing is ever "certain").

An asset can show "No signal yet"instead of a call entirely — this happens when there's neither recent news nor enough price history to compute a momentum figure. Rather than manufacture a Hold from nothing, we'd rather say nothing.

Stocks are the one asset class where "No signal yet" is often about scale rather than missing data. The stock universe (loaded from the SEC's full ~10,000-company ticker list) is far larger than crypto or prediction markets, and scoring all of it every 3 hours isn't practical on news matching alone. So stock signals are computed only for a bounded, growing trackedset: any stock that's matched at least one news article recently, is on any user's watchlist, is one of the small number of featured picks, or has ever been scored before (once tracked, a stock stays tracked). Every other stock in the universe is fully searchable and viewable — its price loads live the moment you look at it — it just shows "No signal yet" until something (a news hit, a watchlist add) pulls it into the tracked set.

Every signal issued gets its price at the moment of issue recorded. 24 hours later, it's checked against the asset's current price and marked Correct or Wrong: a Buy is correct if the price went up, a Sell if it went down, and a Hold if the price stayed within ±2%. Before 24 hours have passed, a signal shows as Pending. The full history — nothing is ever deleted — and an accuracy summary (overall and per-mode) live on the signal history page.

Every external provider precandle64 pulls from:

• We don't have insider information — every input is public news and public market data.
• We don't guarantee outcomes. Signals summarize current evidence, not the future.
• We don't move markets — precandle64's scale doesn't and isn't intended to influence prices.
• We don't sell your data, to anyone, ever.

Signals across every mode recompute every 3 hours, driven by a scheduled refresh job that pulls fresh news, updates prices, and re-scores every tracked asset. Premium accounts always see that latest computation. Free accounts see picks as they stood roughly 12 hours ago — the same underlying signals, just a fixed delay behind — so upgrading doesn't just unlock more assets, it also closes that gap to near-real-time.

Being upfront about where this system is weaker matters as much as explaining where it works:

• Accuracy varies meaningfully by asset class — large-cap stocks with heavy news coverage behave differently than thinly-covered small caps or altcoins.
• Low-volume assets have less signal to work with — a stock or coin with barely any trading or press has correspondingly weaker momentum and news components.
• Some prediction markets have thin data — a newly-listed market may have only one or two price snapshots, not enough for a reliable momentum read.
• Past accuracy doesn't guarantee future results — see the accuracy numbers on the signal history page as a track record, not a promise.