Methodology

A current price and a price history reach us through separate paths. That redundancy is useful: if the two disagree by an order of magnitude, one of them is wrong, and the disagreement is visible without needing an outside reference to compare against.

Every few hours we compare each asset's current price against the most recent candle in its own stored history. The tolerance is wide on purpose — a volatile coin can move a long way in a day — so the check is looking for unit and mapping errors rather than for price movement.

It is not a theoretical safeguard. It is how we found that an entire class of currency pairs was being stored inverted, and how we found ten coins that had picked up a different project's price history because the two share a ticker symbol. Both were fixed; the check is what surfaced them.

Different markets move at different speeds and we do not pretend otherwise. Each value is classified by the feed that produced it, not by its category: crypto is polled every three minutes where Binance carries the pair and every fifteen otherwise; metals are front-month COMEX futures that Yahoo Finance publishes about ten minutes behind the exchange, so they are labelled delayed rather than live; foreign-exchange rates are European Central Bank reference rates published once per business day. Presenting a once-a-day rate as an intraday quote would be misleading regardless of how recently we fetched it, and so would presenting a ten-minute-old futures quote as one.

• crypto — last refreshed 23 minutes ago
• forex — last refreshed 3 hours ago
• metal — last refreshed 13 minutes ago

Prices are recorded as they happen and kept. That matters for a reason worth stating plainly: a chart assembled on demand from somewhere else disappears the day that source changes its terms, while a history we have stored does not.

Where an asset trades on a venue with a public candle feed, we store that venue's real open, high, low and close values. Those are always preferred: an exchange sees every trade, and we see the prices we happened to read.

Around eight thousand tokens in the catalogue are quoted by no venue we read. They were found on a decentralised exchange and no data provider has ever listed them, so nobody can supply their history — the provider that would does not know they exist. For these we read the live price every four hours and store each reading, and a day's readings become that day's bar: first reading the open, last the close, highest and lowest the high and low. A day with only one reading produces no bar at all, because a single point is not a candle.

Such a chart is labelled on the asset's own page, with the number of days recorded and the date recording began. It is a sampled series and we say so: a move that happens entirely between two readings is not in it. What it is, is the only record of that token's price that exists anywhere, and it grows by one day every day.

Chart patterns, candlestick patterns and indicator events are detected automatically from stored candles. Where we quote a historical success rate, it is measured against a baseline: the rate at which price moved the same way over the same period without any signal at all. A raw win rate in a trending market mostly measures the trend.

That comparison sometimes produces unflattering results, and we publish those too. Several widely followed signals — including the 50/200 moving-average crossover — have performed no better than the market did on its own across the history we hold. A statistic is only worth showing if it is allowed to say no.

Rates are withheld entirely below thirty recorded occurrences. A percentage over a handful of cases invites a conclusion the data cannot support, and a footnote does not undo that.

Chart patterns are detected geometrically from stored candles. The percentage on a pattern card labeled shape fit measures how closely the price structure matches the textbook geometry — it is not a probability of success and is never presented as one.

Success probabilities come from backtests instead: every historical occurrence of a pattern type is replayed against what price actually did next, aggregated across all assets of a class. Each published rate appears beside its sample size and a random-walk baseline — the rate a coin-flip entry with the same target and stop would achieve — because a hit rate without its baseline quietly poses as skill. Detections also pass a quality gate first: impossible levels, targets inside an asset's ordinary daily noise, and stablecoin noise are rejected before publication, with the rejection reason stored.

The replay follows four rules, each of which decides whether a published rate means anything. A pattern is entered once, at the close of the bar it confirms on, and never re-entered while it remains on the chart. Its outcome is read from the bar after entry onward, so a move that had already happened when the signal appeared cannot be counted as the signal working. A bar that touches both the target and the stop is recorded as a loss, because a daily or hourly candle cannot say which came first and the optimistic reading is the one that flatters us. And every trade is charged a 0.20% round-trip cost — roughly a taker fee at the top of the retail band plus slippage of similar size — so a target sitting a fraction of a percent from entry shows as the loss it is rather than as a win.

Two numbers are published from that replay and they answer different questions. The success rate is the share of occurrences that reached the target before the stop. Median R is what the trades returned per unit of risk, across all of them, after costs — it is negative for a pattern that lost money, which a figure taken over winning trades alone can never be.

The score (0–100) is a market-strength reading computed from six measured components: trend versus the asset's own moving averages, momentum, the accumulated edge of its recent signals, quality-gated pattern activity weighted by backtested edge, volume against its own 30-day median, and news attention. Weights are fixed and documented; when a component's data is missing it drops out and the rest renormalize — absence is never treated as neutrality.

Every stored score carries its algorithm version and full component breakdown, so any number the site has ever shown can be traced to the formula that produced it. It is not a buy or sell rating, and the interface says so wherever it appears. We have no news-sentiment feed, so the attention component measures how much the market is talking about an asset — not what it is saying — and is labeled accordingly.

Descriptive text on asset pages is generated from each asset's own recorded facts and must pass an automated QA gate before publication — checks for duplicated words, leaked formatter values, unresolved placeholders and impossible figures. A failing document never replaces previously published good content. The standards it is held to are in the editorial policy.

For tokens with a published contract address we read what the contract itself permits — whether supply can still be minted, whether an authority can freeze or seize balances, whether the owner is hidden, whether trading can be blocked. This describes code, not conduct. It cannot tell you whether a team will deliver or whether reserves exist.

The scanner returns the same finding for a regulated stablecoin as for an anonymous token: an authority can freeze balances. For one of those that is the exit, and for the other it is the disclosed design — Circle can freeze USDC because law requires it to, and minting is how redemption works at all. Reporting “USDC: high risk” would be both wrong and unfair to a real issuer.

So powers that are standard for an asset class are listed separately as issuer controls and excluded from the verdict, while the facts stay on the page — a holder genuinely should know who can freeze their balance. Anything outside that list is counted in full: a hidden owner, a creator with a history of abandoned tokens, or a modifiable transaction fee is explained by no asset class and is reported as found.

A token with no report shows no panel. Absence of a scan is not a clean result, and an empty box in that position would be read as one.

A page shows what exists rather than filling gaps with placeholders. If we hold no project description, there is no description section; if an asset has too little history for a chart, the chart is absent rather than drawn from two points.

Pages that do not yet rest on enough asset-specific data are published and browsable but are not submitted to search engines. They become eligible automatically once enough material accumulates — nothing is manually promoted.

Nine measures calculated by Heriux from its own price history. Each describes what happened or what is; none promises anything. Shared definitions: a day is a completed UTC daily candle; candles with a non-positive close are ignored; if more than 2% of days between an asset's first candle and today are missing, candle-based measures show a dash with the reason. The stablecoin flag comes only from the data provider's Stablecoins category or a manual list — never from an asset's price. RSI(14) is Wilder's. The point-in-time percentile of a value compares it with every earlier value of the same series and never uses future data.

RSI(14) zones: Oversold <30, Weak 30–45, Neutral 45–55, Strong 55–70, Overbought ≥70. An episode is a maximal run of consecutive days in one zone, sampled on the day the zone was entered; today's ongoing run is excluded. For a horizon of 7, 30 or 90 days, an episode counts only if the close that many days later exists. Shown only with at least 10 episodes. The base rate is the share of all days whose close was higher after the same horizon. The one-line takeaway compares the episode rate with the base rate against its sampling error. Samples overlap and are not independent; small differences can be chance. Historical statistics, not a prediction.

Beta is the ordinary-least-squares slope of the coin's last 30 completed daily returns on the benchmark's (at least 25 paired days). The benchmark is Bitcoin; for Bitcoin itself it is the median daily return of the top-20 non-flagged coins. The market-driven part of a 24h move is beta × the benchmark's 24h change; the independent part is the remainder, in percentage points. The home-page board lists coins with market cap ≥ $10M and 24h volume ≥ $1M.

Peak = highest daily close in Heriux data (forex: last 5 years). Drawdown is the distance below it; the required gain is peak / price − 1, which is why −75% needs +300%. Within 0.5% of the peak the asset is shown as at its record close.

The share of max supply (or total supply where no maximum exists) not yet in circulation: <10% Low, 10–40% Moderate, 40–70% High, ≥70% Very high. A descriptive fact about potential future supply, with no unlock dates and no price prediction. Provider supply data may be incomplete.

Over the top-150 universe: a = share of coins up in 24h; b = share of 24h volume in rising coins; c = 50 + 10 × median 24h change, clamped to 0–100. Pulse is the rounded mean of the three. Bands: 0–19 Heavy selling, 20–39 Selling, 40–59 Balanced, 60–79 Buying, 80–100 Heavy buying. Snapshotted hourly.

For windows of 20, 50 and 200 days: the share of top-150 members whose last close is above their simple moving average of that window; members with fewer candles than the window are excluded from that window's denominator. The headline is the 50-day value. Zones: <20 Very weak, 20–40 Weak, 40–60 Neutral, 60–80 Strong, ≥80 Very strong. Values before the first stored snapshot are reconstructed from the current membership and labelled approximate.

Three numbers over the top 250 by market cap: the provider's standard Bitcoin dominance; Bitcoin's share after excluding stablecoins, wrapped/staked/bridged and commodity-backed tokens; and the stablecoin share of the whole top 250.

The step is the round number (1, 2, 2.5 or 5 × a power of ten) closest to 5% of the live price for crypto, 1% for forex and metals. The two nearest multiples above and below the price are examined within a band of ±15% of the step. A test begins on a touch day with no touch in the prior five days and resolves within ten days: a break is two consecutive closes beyond the far edge, a bounce is a close at least half a step away on the approach side; whichever happens first wins, otherwise the test is unclear. Unfinished tests are not counted.

Five components, each a point-in-time percentile of the asset's own series: distance from the 200-day average, 90-day change, distance from the running peak, RSI(14) taken directly, and the 30-day / 365-day volume ratio where volume exists. The temperature is their equal-weighted mean, 0–100: 0–19 Freezing, 20–39 Cool, 40–59 Mild, 60–79 Hot, 80–100 Boiling. Needs 400 daily candles. The market temperature is the median across the top-150 universe. It describes how stretched the price is against its own history — Boiling does not mean it will fall, and Freezing does not mean it will rise.

If a figure looks wrong, it may well be. Reach us through the contact page with the asset and what you expected to see. Data errors are worth more to us than they are to you, and the integrity check above exists because they happen.