What 3,971 Solana Token Outcomes Say About Rug-Risk Scores

Correction and methodology clarification (September 2, 2026): We clarified methodology and disclosure wording, including the auditability and verification limits below. The approved cohort and figures are unchanged.

Why we refreshed this outcome study

MemeAssist records a token's scores at analysis time, then checks what happened to its price over the next 24 hours. That prospective order matters: the score exists before the outcome, rather than being reconstructed after a collapse. Our first edition reported 1,464 completed outcomes and focused on the Overall Health Score. By 31 August 2026, the approved research snapshot contained 3,971 resolved outcomes with a recorded Rug Risk Score—large enough to test the separate rug-risk bands directly.

The question is deliberately narrow: did the score separate tokens that subsequently produced much worse 24-hour outcomes? It does not ask whether a token was criminally fraudulent, and it does not treat every price decline as a rug pull.

The result: High rug risk isolated the most dangerous group

We grouped each token by the Rug Risk Score captured at analysis time. A row counted as resolved when a valid 24-hour return was recorded or the tracked pool was classified dead. Dead pools count as −100%; unresolved and no_data rows are excluded.

The clearest finding is at the top of the scale. A token in the High band was about 3.9 times as likely to be classified dead as one in the Low band (53.2% versus 13.6%). More than half of High-risk observations were dead, which pulled the band's median all the way to −100%. All three bands exceed our minimum of 30 resolved outcomes, so the table is not being driven by a tiny subgroup.

The honest wrinkle: Medium and Low did not rank cleanly

A perfectly calibrated score would produce a smooth staircase: worse returns and more dead pools at every higher band. This snapshot did not. Medium and Low had the same rounded dead-pool rate, while the Low band's average return and share finishing up were slightly worse than Medium's.

That means the evidence supports a specific conclusion, not a universal one: High is a meaningful severe-risk warning. It does not show that every move from 20 to 40, or 40 to 60, predicts a proportionally worse return. Market momentum, liquidity, token age and selection effects can dominate outcomes among tokens that avoid the strongest deterministic warnings. MemeAssist therefore presents the numeric score beside its underlying evidence and plain-English verdict, rather than treating one number as a trading signal.

What the original Health Score cohort found

The original August 2026 edition used a separate Overall Health Score and a smaller, earlier cohort of 1,464 completed 24-hour outcomes. In that snapshot, tokens scoring 60+ on health averaged +5.2%, while tokens below 40 averaged −5.0%; weak health scorers were five times more likely to lose at least half their value. Those figures remain a historical prospective result, but they answer a different question. The refreshed table above is the current evidence for the dedicated Rug Risk Score and uses the newer dead-pool treatment.

Neither result says a high Health Score guarantees gains or that every High rug-risk token is a proven scam. Scores summarize observed risk conditions at one point in time; markets and pool conditions can change immediately afterward.

Methodology and limitations

Cohort: 3,971 production analyses with a recorded Rug Risk Score and a resolved 24-hour outcome in the founder-approved evidence snapshot on 31 August 2026. The High band contains scores 75–100, Medium 30–74 and Low 0–29. We count a row when its tracker status is done with a valid 24-hour return, or dead. Dead pools receive a −100% return and count in the dead-pool column. Rows marked no_data, unresolved rows and 503 resolved rows without a recorded Rug Risk Score are excluded from the band analysis.

• Dead is not a legal finding of fraud. It means the tracked market became dead or delisted under the outcome system. That is economically severe, but it does not identify the actor or intent.
• This is observational. The bands were recorded prospectively, but the study does not prove the score caused the outcome or isolate every market variable.
• The cohort is selected. It contains tokens submitted to or discovered by MemeAssist, not every Solana token launched in the period.
• Confounding remains. The bands were not adjusted for liquidity, market cap, token age, momentum or overlapping risk signals.
• Statistical uncertainty is not quantified. No confidence intervals or significance tests were calculated.
• Generalizability is limited. Results from this selected, period-specific production cohort may not apply to other periods or to all Solana launches.
• The horizon is 24 hours. A token can survive this window and fail later, or recover after an early decline.
• Returns are highly skewed. Averages are sensitive to rare large moves, which is why we also publish medians, dead-pool counts and the share finishing up.

Auditability and verification: public row-level records for this historical cohort are not available. The source facts and outcomes were generated from the internal production tracker and were not manually verified row by row. Accordingly, the aggregate tables cannot be independently row-audited from public data.

Check the current evidence for a token

The score in this study is the same Rug Risk Score shown when you paste a Solana mint address into MemeAssist. Use it as a screen, then read the authority, liquidity, holder, creator, wallet and trading evidence behind it. Re-run old reports before acting because a score is a timestamped observation, not a permanent label.