Methodology
A plain-language explanation of the CoreX Score, the six-agent intelligence pipeline, and the boundaries of this information service. Read this before you read anything else we publish.
Built Honestly in Dubai
SmartCryptoRadar is built by CoreX — AI Software Innovations DMCC in Dubai, UAE. We operate in alignment with the principles of the Virtual Asset Regulatory Authority (VARA) and the broader UAE digital-asset framework. We are an information service: we publish research, briefings, and structured data. We do not custody funds, execute trades, or hold any virtual asset on behalf of users.
We disclose our data sources, name the underlying models we use (our synthesis layer for synthesis, our routing model for routing, our high-throughput model for classification), and publish the formula behind every score we put on screen. When something we publish is wrong, we want it to be auditable.
How We Score Exchanges
The CoreX Score is a single number from 0 to 10 summarizing the operational quality of a centralized exchange across five components. Each component is normalized to a 0 to 10 range and then combined with the fixed weights in the table below. The score is recomputed weekly from public data; the source set and timestamp are visible on every row.
| Component | Weight | What goes in | Why this weight |
|---|---|---|---|
| Liquidity | 30% | Spot 24h USD volume, order-book depth at 2% from mid, pair count for the user's quote currency. | Liquidity is the single largest determinant of execution quality, so it carries the biggest weight. |
| Coverage | 20% | Number of listed assets, breadth of stablecoin pairs, presence of long-tail tokens with credible volume. | An exchange that lists what users actually hold is materially more useful than one with deep BTC books and nothing else. |
| Reliability | 20% | Public uptime history, incident frequency from the venue's status page, withdrawal-pause events in the last 12 months. | An exchange you cannot reach during a drawdown is worse than one with thinner books that stays online. |
| Transparency | 15% | Proof-of-reserves cadence, public auditor, on-chain wallet disclosures, fee-schedule clarity. | Transparency is observable; it is not the same as solvency, but it correlates with operational discipline. |
| Regulatory | 15% | Jurisdictions licensed, regulator quality (VARA, MAS, FCA, etc.), KYC tier structure, sanctions-screening posture. | Regulatory standing reduces tail risk for users; we cap its weight so well-run unregistered venues are not auto-zeroed. |
Affiliate status is not an input to the CoreX Score. Whether or not an exchange has an active affiliate program with us has zero weight in any component. Sort order on every list is strictly CoreX Score descending. Affiliate rows are tagged with a small "(a)" icon so users can see the relationship at a glance.
Last methodology revision: 2026-04-29. Component definitions are versioned; older versions remain in the repository for replay.
How Our Signals Work
SmartCryptoRadar runs six specialized agents in parallel, each focused on a single domain: a Technical Analyst over OHLCV and indicator data, a Smart Money Tracker over labeled wallet flows, a Newsletter Analyst over the day's ingested research, an On-Chain Analyst over verified DeFi and protocol-level metrics, a Risk Sentinel over hack and exploit feeds, and a Pattern Analyst over historical OHLCV embeddings. Each agent runs deterministic routines first and only escalates to a model call when the rule layer is inconclusive.
Agent outputs are written to a structured store with timestamps, source URLs, and confidence ranges. A synthesis step (our synthesis layer) merges the six outputs into one briefing per cycle, with the rule that any factual claim must trace back to at least one cited source. Briefings publish twice daily; alerts fire on a separate cadence keyed to the Risk Sentinel and Smart Money Tracker.
Pattern matches are reported as historical analogues with the full distribution of past outcomes, never as point predictions. When 12 of 19 prior patterns similar to today were followed by positive returns, we report that ratio along with the range; we do not phrase it as a forecast. Calibration is reviewed quarterly using resolved predictions and isotonic regression.
What This Service Does Not Do
- We do not execute trades. SmartCryptoRadar is read-only and never holds keys, signatures, or withdrawal permissions.
- We do not custody funds. Nothing in the platform takes possession of digital assets at any layer.
- We do not produce personalized financial output. Briefings are general intelligence; your jurisdiction, tax position, and risk profile are not in scope.
- We do not guarantee outcomes. Pattern analogues are probabilistic and historical; they are not forecasts.
- We do not pay for placement. Content rankings and CoreX Scores are independent of any commercial relationship with the venues we cover.
Corrections and replay
If you find a factual error in a published score or briefing, email [email protected] with the URL and the disputed claim. We log the report, publish the correction, and keep the prior version available so the change is auditable. The methodology versions referenced on every score row are the source of truth for what was used to compute it.
Read this alongside our Terms. Both pages describe the same operating posture from different angles.