Methodology & caveats
TokenIntel separates prioritization from verdict. The community votes on what gets investigated next. Votes never decide whether a project is a scam — the final verdict is produced by the investigator and supported by on-chain evidence, contract review, and wallet analysis.
Risk scoring
A weighted, additive, evidence-bound model. Every weight is positive and every awarded point must link to evidence — if there is no evidence, the point is not awarded.
A signal whose data was not collected is excluded from the score and from the score’s maximum, and is published as NOT EVALUATED rather than treated as clean. The model below defines 200 points in total, but no report is scored against all of them — each is scored only on the points it could actually evaluate, normalised to 100. Every report lists which signals it checked and which it could not, so a project is never made to look safer because we could not look.
| Signal | Weight |
|---|---|
| Unverified contract source | +15 |
| Proxy with unknown or centralised upgrade authority | +15 |
| Team wallets drained, destinations unlabelled | +18 |
| Owner withdraw or drain function | +12 |
| Hidden mint or suspicious supply control | +12 |
| Transfers to a known exchange hot wallet | +10 |
| Large flows to unlabelled destinations after the raise | +8 |
| Blacklist or freeze controls | +8 |
| Ownership not renounced | +8 |
| Transfer tax above 5 percent | +6 |
| Sale contract and token share one deployer | +4 |
| Liquidity removable by a team wallet | +12 |
| Concentrated holder distribution | +8 |
| Weak or unverifiable public team identity | +6 |
| Team wallets no longer hold a substantial share of the raise | +6 |
| No public code repository | +52 |
Risk model version 4. Reports state the version they applied.
Source repository signals
Exactly one side of this dimension is ever scored, and both sides are worth 52 points, so a project is measured against the same maximum either way. A project that publishes no repository takes the full 52 as a single signal: none of the 6 findings below can be checked, and an unpublished codebase could contain any of them undetected. A project that does publish one is assessed on the 6 findings instead — a clean repository scores zero against the same 52 points, so transparency is what earns the lower score.
| Signal | Weight |
|---|---|
| Deployed bytecode does not match the repository source | +20 |
| Contracts absent from the published repository | +10 |
| Repository created immediately before the raise | +6 |
| Single or throwaway contributor history | +6 |
| Fork with original attribution removed | +6 |
| No tests, CI, or audit artefacts | +4 |
The extractors for these six findings are still being built. Until they exist, a project that does publish a repository has them recorded as NOT EVALUATED rather than scored.
Risk labels
Coverage limits
Known limits of the pipeline itself, not findings about any project. Each run repeats only the limits its collected evidence did not resolve.
- Live owner or proxy-admin state may be unavailablehigh
The pipeline queries owner() and EIP-1967 state. This limit remains active only when a required result fails, reverts, or cannot be decoded.
- Verified implementation source may be unavailablehigh
The pipeline resolves and requests the implementation source. This limit remains active when an upgradeable proxy's implementation is unverified or unavailable.
- Exchange attribution is a nine-address allowlisthigh
Anything outside nine hardcoded exchange hot wallets defaults to UNKNOWN, which inflates the untraced signal. No match means not identified by this list, never that no exchange received funds.
- The team wallet list is a configuration assumptionmedium
Wallets come from configuration and are never proven on-chain to be the complete set the team controls. If others exist, every figure derived from the list is an undercount.
- Coverage is Ethereum mainnet onlyhigh
Cross-chain activity, card settlements, and other networks are invisible to this pipeline. Totals and any share-of-funds figure could be materially wrong if that volume is real.
Caveats
- Reports describe risk and evidence, not legal conclusions.
- Observed inflows are gross successful inbound transfers, not audited net proceeds.
- USD figures use one documented price snapshot, not the historical price at each transaction.
- Totals, balances, destination aggregation, and deterministic signals are computed from stored evidence. Qualitative signals are assessed against quoted source evidence and labelled as such in each report.
- A signal whose data was not collected is excluded from both the score and its maximum, and published as NOT EVALUATED rather than treated as clean.
- An UNKNOWN destination is unclassified, not proof that it is malicious.
- A zero sale-contract balance is the expected end state once proceeds are swept to team wallets, and is never scored on its own.
- Data snapshots are stored so each report remains reproducible.
- Project teams may submit responses or corrections.