Scoring QA Update: Fewer False Alarms, Sharper Scam Detection

Running a scoring system at scale means you will occasionally catch something in your own QA that feels wrong — a well-known token with a suspicious-sounding alert, or a sketchy token that somehow sneaked past a check. Both types of errors matter, and we track them constantly.

Today we are shipping a set of targeted fixes that came directly out of that internal QA work. The goal across all of them is the same: make the score more truthful, not just more favorable or more alarming.

What we improved

• Fewer false “can't sell” warnings on established tokens. When BarryGuard cannot fully simulate whether a token can be sold — because the check requires features that are not available for that specific token type — it previously could still produce a vague warning. That was unfair. Now, if we genuinely cannot run the simulation in a meaningful way, we say that clearly instead of translating uncertainty into a risk flag. The warning only fires when there is real evidence of a sell block, not when the check is simply out of reach.
• Old tokens with unresolvable deployers are handled more fairly. On some older tokens, especially on BSC and Base, the original deployer address is no longer traceable through standard on-chain methods — blockchain history gets pruned over time. Previously, a missing deployer could push an old token into an “unknown creator” penalty, making it look young and suspicious even when it had been running for years with a large verified holder base. Now, when the age evidence, the holder count, and the market data all point to a genuinely established token, BarryGuard gives it a neutral score and flags the deployer gap as an information limit rather than a risk signal.
• Scam patterns can no longer hide behind a single positive signal. A locked liquidity pool is a good sign in most cases. But if a token shows locked liquidity alongside multiple other scam indicators at the same time — unusual contract behavior, concentrated ownership, suspicious bytecode — those positive signals are now capped. They can no longer pull the total score into the moderate range on their own. Multiple red flags still mean a high-risk outcome, even if one individual signal looks clean.
• Less score drift when price data is momentarily unavailable. Price and market-cap APIs occasionally return errors when they are under load. When that happened, checks that depend on price context could score lower than they should, creating a temporary dip that had nothing to do with the token itself. We now retry once after a short pause before treating a pricing failure as a real data gap, which smooths out those transient fluctuations.
• Clearer internal diagnostics for faster future fixes. We improved how our QA process surfaces details about unexpected results so the team can immediately see whether a surprising score comes from a data provider gap, a calibration issue, or a genuine bug. This does not change any scores today, but it means that when something does drift in the future, we can diagnose and address it faster — and that benefits you directly.

What is still open

Scoring calibration is never fully done. Some edge cases that surfaced in this QA round are still under investigation — mainly around very old contracts on BSC and Base where the deployer data gap is large and the on-chain age signals are ambiguous. We know the patterns and are working toward better detection, but we will not ship a fix until we are confident it does not create a new problem somewhere else.

We also keep a close eye on how scam-pattern caps hold up as new token structures emerge. Scammers adapt, and so do we.

Better, not perfect

No scoring system eliminates all false positives and all false negatives at the same time. The goal is to keep making the tradeoff more honest — and to keep explaining our reasoning openly so you can decide how much weight to put on any given result.

If you see a score that does not match what you know about a token, we genuinely want to hear about it. The feedback that reaches us through the check page is one of the main ways these QA rounds get started. It helps more than you might think.

Check a token now →