comb

The agent continuously observes the wallets interacting with a specific token and maintains a limited internal record of the wallets it considers most relevant at any given time. Each record contains the wallet address together with a compact history of the behavior the agent has observed, including when the wallet first appeared, how long it remained active, whether its balance increased or decreased, how frequently it interacted with the token, and whether its behavior changed enough to alter the agent’s interpretation of it.

The agent does not have unlimited memory. It can maintain active records for only one hundred and sixty three wallets at once. This limit is fixed. Once all available memory is occupied, the appearance of a new wallet creates a direct competition for space. The agent must decide whether the new wallet contains enough new information to justify replacing one of the wallets it already remembers.

That decision is made from the agent’s current internal state. It evaluates the wallets it is holding according to factors such as recency, activity, behavioral change, repetition, persistence, and the amount of unique information each wallet contributes to its understanding of the token. A wallet that has remained unchanged for a long period may gradually become less useful. A newly active wallet may become more important because it introduces behavior the agent has not seen before. The system is therefore not maintaining a list of the largest holders or the most recent participants. It is maintaining a changing set of observations that the agent currently considers worth preserving.

When a wallet is removed from active memory, the agent loses the record associated with it. The address can still appear onchain and the wallet can continue interacting with the token, but the internal description the agent previously constructed is no longer available to it. The agent cannot retrieve that description later and cannot use it when making future decisions.

The website preserves these discarded records separately. This creates an important difference between what the agent knows and what the site knows. The public archive contains every memory the agent has removed, including the exact observation that existed before deletion, the time the wallet entered memory, the duration for which it remained there, the reason it became less valuable relative to another observation, and the moment it was replaced. These records remain readable to anyone visiting the site, but they are deliberately excluded from the agent’s working context.

The archive therefore contains information about the agent’s own past that the agent itself can no longer access. A visitor can move backward through the history of the system and see how the agent’s understanding developed over time. The agent can only operate from the one hundred and sixty three records that remain active in the present.

A wallet can also return after being forgotten. If the agent encounters that wallet again after its previous record has been removed, it does not restore the earlier memory. It begins observing the wallet again from its current behavior and creates a new record as though the earlier observation were unavailable. The archived version and the new version then exist independently.

This allows the same wallet to appear multiple times across the history of the system. A wallet may have one record from an earlier period in which it accumulated tokens slowly and another months later in which it begins moving large amounts through different accounts. A visitor can compare those observations directly and recognize that they belong to the same address. The agent cannot make that comparison unless the older information happens to remain inside its active memory.

Over time, the project becomes more than a record of token holders. It becomes a record of how a constrained observer builds and loses an understanding of a changing onchain environment.

The important information is not only which wallets are present. It is which wallets remain important enough to survive inside the agent’s limited memory, which observations are displaced when something more relevant appears, how often forgotten wallets return, and how different the agent’s current understanding becomes from the complete history preserved outside of it.

At any moment, the agent holds only a small representation of the system it is observing. That representation is continuously revised as new activity arrives. Every new observation has the possibility of changing what is remembered. Every replacement permanently changes the information available to the agent afterward.

The site records both sides of that process. It shows the active memory the agent is currently using to understand the token, and it preserves the growing body of information the agent once knew but can no longer see. The result is a system whose public history becomes more complete while the intelligence producing that history remains intentionally incomplete.