We Ranked 18 Solana Rug-Risk Signals by 24-Hour Outcomes

The short answer: five warnings separated from the pack

MemeAssist records deterministic risk signals when a token is analyzed, before its 24-hour outcome is known. On 1 September 2026, we froze the production signal table and ranked every warning with at least 30 resolved outcomes. Five stood out clearly against analyses where no tracked signal fired: no sell route, copycat evidence, top-10 ownership above 65%, a largest holder above 30%, and liquidity below 2% of market cap.

The result is not a list of guaranteed rugs. Several signals can fire on the same token, the cohort comes from tokens submitted to or discovered by MemeAssist, and “dead” describes a market outcome rather than proving fraud. But the ranking shows which warnings were associated with the most severe next-day outcomes—and which familiar warnings were weak on their own.

All 18 qualified rug-risk signals ranked

The table includes signals with at least 30 resolved outcomes. We rank by average 24-hour return because that captures both surviving markets and dead pools; dead pools count as −100%. Median return, share finishing down and dead-pool rate are included because memecoin returns are highly skewed. The no-signal row is a comparator, not a safety guarantee.

Why the top four are qualitatively different

The four worst rows all had a median return of −100%. In other words, at least half of each group ended with a dead tracked market—not merely a disappointing candle. A missing sell route is the most direct failure mode: the tracker could not find an executable route and all 30 qualified outcomes were dead. The sample is exactly at our publication floor, so it should be revisited as it grows, but the observed result is unambiguous.

Copycat evidence was nearly as severe across a larger sample: 121 of 134 resolved observations were dead. Extreme holder concentration also repeated at scale. When the top 10 controlled at least 65%, 445 of 559 markets were dead. When one holder controlled more than 30%, 484 of 689 were dead. Those two rows can overlap, so they should not be treated as independent votes or added together.

Liquidity showed a cliff, not a smooth gradient

The liquidity result supports a threshold effect. Tokens below 2% liquidity relative to market cap averaged −46.0%, and 46.3% were dead. The broader below-5% signal was much weaker: a −12.5% average and 10.3% dead. That does not mean 2.1% liquidity is automatically healthy. It means the most extreme mismatch separated a dangerous cohort, while the wider warning did not rank as a severe standalone predictor in this snapshot.

This is also why ratios should be read beside executable depth. Reported liquidity can change quickly, market-cap estimates can be noisy, and a pool that looks adequate for a small sale may fail under a larger position.

Developer selling mattered—but was not a death sentence

Developer selling was the most frequently resolved named signal in the table, with 733 outcomes. It averaged −27.0% and had a −15.2% median, both worse than the no-signal comparator. Yet 21.7% were dead, far below the rates for copycats or extreme holder concentration. That supports the conclusion in our focused developer-selling study: creator selling is meaningful negative evidence, but amount, timing, remaining ownership and liquidity determine whether it becomes an immediate collapse.

Some scary labels were weak by themselves

Several qualified rows performed better than the no-signal baseline on average. That does not make them bullish. Signals overlap, thresholds divide continuous values into artificial buckets, and the baseline contains plenty of ordinary low-quality memecoins. A 10–19% wallet cluster, 8–14% largest holder or 30–44% top-10 concentration may deserve investigation without independently forecasting a dead market.

Token age changed the concentration result substantially. The “established” variants of extreme holder concentration had only 34 and 40 resolved outcomes, but neither reproduced the catastrophic result seen across the unrestricted cohort. That is plausible: treasury, exchange, liquidity and custody wallets can appear concentrated on older projects. It is also a reminder that a raw ownership percentage needs wallet labels, token age and distribution context.

What we did not rank as proven

Nine tracked signal keys had fewer than 30 resolved outcomes in the frozen snapshot, so they were excluded from the numbered ranking: dangerous token extensions, dormant extension authority, freeze authority, mint authority, mint authority on established tokens, creator-history variants, wash trading and high slow-rug risk. Some looked severe—including three deaths from three high slow-rug outcomes—but samples that small are unstable. We publish them only after they cross the same minimum evidence floor.

Exclusion is not a declaration that a signal is safe. Freeze authority and dangerous extensions can grant explicit technical powers regardless of historical frequency. Deterministic honeypot and authority checks therefore remain safety controls even while their outcome cohorts mature.

Methodology and limitations

Cohort: production analyses with captured rug-signal arrays and a resolved 24-hour outcome, queried on 1 September 2026. A row resolves when the tracker records a valid 24-hour return or classifies the tracked pool as dead. Dead pools count as −100%. Unresolved and no_data rows are excluded. The numbered table requires at least 30 resolved outcomes for each signal.

• Signals overlap. One token can appear in several rows, so the table does not isolate the causal effect of each warning and row counts must not be summed.
• The cohort is selected. It reflects tokens analyzed or discovered by MemeAssist, not every Solana token launched.
• Dead does not prove fraud. It identifies a dead or delisted tracked market, not intent, identity or a legal conclusion.
• The horizon is fixed at 24 hours. Later failures and recoveries are outside this study.
• Returns are skewed. Rare large gains affect averages; medians, down-share and dead-pool rates provide necessary context.
• Thresholds are not natural laws. Market conditions, token age, wallet labels and data coverage can change how a signal behaves.

How to use the ranking before buying

Treat the first five signals as reasons to stop and investigate, not as a mechanical short signal. Confirm that a sell route exists, compare the mint and branding with earlier launches, inspect who the dominant holders actually are, and test liquidity against your intended position size. Then read the other five MemeAssist signal categories—creator behaviour, wallet activity, trading patterns, historical risk and contract controls—because no single table captures the whole token.

You can analyze a Solana mint with MemeAssist to see the same six evidence categories, overall health score, rug risk rating, plain-English AI verdict and detailed risk breakdown used by this outcome system.