Every number traces back to a stage and a provider.
Five stages run in order. The early ones collect and verify evidence; the last two turn it into the risk score and the confidence score that always travels with it. Select a stage to read what it checks, which sources serve it, and — the part that matters most — what it cannot see.
The pipeline · select a stage
Input
Resolves what was pasted into one scannable token. An EVM address, a Solana mint, a DexScreener pair URL or a free-text search all land here; the chain is identified — explicitly or from market listings — and the scan is dispatched at the deepest depth that chain supports.
What this stage checks
Data sources
Chain coverage
Other chains with public market listings receive the universal market scan — market categories only, with capped confidence.
What it cannot see
A stage that returns nothing lowers confidence. It never lowers risk.
How the two scores combine
Risk score
An ordinal summary of the warning signs that were observed, weighted by severity. It is not a probability, a rating or a prediction, and two tokens with the same score are not equivalent — their findings may differ entirely.
BAND NAMES ARE OBSERVATIONS. LOW IS NOT A CLEARANCE AND CARRIES NO TICK.
Confidence score
How complete the evidence behind the risk score actually is. Every deep scan starts at 100 and loses points for each unavailable check, unverified source or provider conflict. It is reported on every scan and is the figure that tells you how much the risk score is worth.
A low score at confidence 51 means key checks were unavailable — not that little was wrong. Reading the risk figure alone is the most common way to misuse this product, which is why the two never appear apart.
Baseline market scans use the universal engine (u1.0): confidence starts at 92 and is capped at 88, so a market-only scan can never present deep-scan certainty.
Category weights · hard caps
Each category's observed points are capped at its weight; the risk score is the sum of capped categories (0–100). Narrative/slop analysis is scored separately and shown on reports without affecting the risk score.
The universal market engine (u1.0) has its own weight table over six market categories and can reach at most 85 of 100 — a scan that never ran contract checks cannot reach full marks.
Why this score
Example factors only — every real report carries its own breakdown.
Blacklist function confirmed - 8 points - HIGH
Example report preview: transfer restrictions would require further verification before relying on this token.
Category: CONTRACT_PERMISSIONS. Source: example-contract-analysis. Status: OBSERVED. Timestamp: 2026-06-30T09:00:00.000Z. Confidence impact: 0.
Liquidity dropped over 30% - 9 points - HIGH
Example report preview: liquidity moved materially compared with the last snapshot.
Category: LIQUIDITY_RISK. Source: example-liquidity-history. Status: OBSERVED. Timestamp: 2026-06-30T09:00:00.000Z. Confidence impact: 0.
Holder data unavailable - 0 points - INFO
Example report preview: holder checks were unavailable, lowering confidence.
Category: HOLDER_CONCENTRATION. Source: example-holder-provider. Status: UNAVAILABLE. Timestamp: 2026-06-30T09:00:00.000Z. Confidence impact: -12.
Narrative quality / saturation analysis
Narrative analysis estimates whether a token's theme appears fresh, saturated, derivative, or artificially hyped. It does not predict profit. It is scored separately from technical scam risk: by default it never changes the risk score, and an optional "include narrative in total risk" setting only adds a small, clearly labelled adjustment to a separate combined score.
What is scored (0–100 each)
- Novelty — how new/uncommon the theme looks. New is information, not an endorsement: a fresh narrative can still be a rug.
- Saturation risk — how heavily recycled the theme is (dog/frog memes, baby forks, moon naming…).
- Momentum — market activity (volume vs liquidity); social momentum is added once social providers are configured.
- Originality — how derivative the name/theme is, including near-duplicates of well-known tokens.
- Manipulation risk — patterns consistent with artificial hype (e.g. heavy volume on thin liquidity or brand-new pairs).
- ScanZX Slop Score — a combined copycat/saturation read; higher = more slop-like.
- Narrative confidence — completeness of narrative data only. Missing social providers lower this score, never the technical confidence.
Categories and honesty rules
- Categories: original/emerging, fresh derivative, overused meme, copycat/slop, artificial hype signals, insufficient narrative data.
- An overused meme is not automatically a scam, and an original narrative is not automatically quality.
- Similarity currently uses string matching against known tokens and the ScanZX scan history; unavailable checks (social providers, external saturation) are always listed as not checked.
- The sniper profile highlights fresher narratives but never presents freshness as safety.
Change notice
The deep methodology in force: eight severity-capped categories, confidence starting at 100, and the INSUFFICIENT DATA override below 35 confidence.
The baseline market methodology for chains without deep coverage: six market categories reaching at most 85, with confidence capped at 88.
REPORTS ARE NOT SILENTLY RESCORED WHEN THE METHOD CHANGES — EVERY REPORT KEEPS THE VERSION THAT PRODUCED IT.