Multi-Agent Review: How to Filter Noise Through Disagreement

From 516 to 2: Today's Snapshot of the Coin Selection Funnel

The latest full-market scan (as of 2026-10-27 10:00) shows that Binance USDT spot trading pairs totaled 516, of which 267 had market caps within the research scope, and 69 passed all hard filters. Among the reasons for failing hard filters: market cap outside research scope 208, risk tags 168, not yet in a bottom area 38, insufficient or unstable liquidity 13, listing time too short 2. Figure 1 illustrates how the funnel narrows: from 516 trading pairs, to 267, then to 69.

Note that this is a snapshot of the latest scan, not a cumulative total for the past 24 hours. The past 24 hours' review council data is counted separately: 2 sent to review, 1 approved, 0 red-flag vetoes, 1 with insufficient approval votes, 0 under review. From the 69 that passed hard filters to only 2 eventually sent to review, this shows the product is very vertical, and most coins are filtered out at the rule layer.

The Potential Board Is Only a Starting Point: AI Reviewers Make the Decision

Figure 2 shows the potential score and below-peak decline for the top 10 on today's potential board. Ranked first, ARKM has a potential score of 65 and is 80% below its peak; second YGG has 63 and 91%; ILV has 63 and 82%; CITY has 61 and 74%; SPELL has 61 and 84%; EIGEN has 60 and 91%; FIDA has 60 and 80%; GALA has 60 and 90%; CHR has 59 and 82%; MAV has 57 and 87%.

The potential board is scored by a rule engine, not AI judgment. A high score only means these coins meet system criteria on dimensions such as price bottom, decline, and volume, but whether they can enter actual operation must be decided by the AI review council. Among the 69 coins that passed hard filters today, only 2 entered review, indicating there is a strict AI review between the potential board and final signals.

Four Reviewers: The Informational Value of Division of Labor and Disagreement

The four AI reviewers each have their roles: Fundamentals analyst looks at project quality, Technicals at price structure, Risk officer at risk boundaries, and a dedicated Skeptic plays devil's advocate. Figure 3 shows their stance distribution over the past 24 hours: Fundamentals analyst buy 1, wait 1, avoid 0; Skeptic buy 2, wait 0, avoid 0; Technicals buy 2, wait 0, avoid 0; Risk officer buy 1, wait 1, avoid 0.

With only 2 cases sent to review today, the sample is very small, so the stance distribution only reflects a narrow range of judgments. Interestingly, Skeptic and Technicals voted buy in both cases, while Fundamentals analyst and Risk officer each had one wait. This disagreement itself is the value of multi-agent review: if all reviewers always passed unanimously, either the system is too loose or the reviewers are redundant. A dedicated Skeptic is set up so that someone proactively finds flaws before groupthink emerges.

Approved Signals and Track Record: Sample Too Small to Draw Conclusions

Among publicly disclosed review-approved signals, EGLD received 4 buy votes, with all 4 reviewers agreeing. In terms of public track record, there is only 1 closed signal, with a win rate of 100%, average 9.89% per trade, median 9.89%, maximum drawdown 0%, average holding period 4.4 days; 13 signals are currently open. In the last 7 days, 0 closed and 0 profitable. For on-chain large transfers, in the past 24 hours there was 1 transfer into exchanges, amount 0.1 million USD, involving 1 coin; and 78 transfers among large holders, amount 79890.09 million USD, involving 13 coins.

It must be cautioned that there is only 1 closed signal, an extremely small sample, so it cannot be used to judge strategy stability, let alone extrapolate returns. The last 7 days with 0 closed further indicates this is still very early and data accumulation is insufficient. For users, what this product shows today is how the review mechanism operates, not any return promise. Until the sample is sufficient, all conclusions must remain prudent.