Memoir

Methodology

Facts first. Signals sourced. Interpretation cited.

Memoir separates what a source directly returned from what can be interpreted about it. Onchain and public-index records become Evidence; named third-party claims and reproducible metrics become Signals; optional AI explanations remain Interpretations that must cite the supplied IDs and preserve uncertainty.

Method: source-cited address investigationLast reviewed: September 3, 2026AI is optional

Three information layers

EvidenceA transaction, trace, log, position observation, or other returned public record with a stable ID and source.
SignalA named source's claim or a deterministic metric tied to its subject, observation time, verification level, and cited evidence.
InterpretationAn AI-produced explanation or hypothesis that cites only evidence and signals supplied in the request.
CoverageThe chains, time window, sources, statuses, retention bounds, and limitations that constrain every other layer.

Selection and analysis

Ordinary browsing never requires an LLM. A user can select timeline ranges, individual records, assets, positions, relationships, account facts, or public signals. Memoir freezes that cross-module selection before analysis begins.

For a one-click address explanation, Memoir uses current workspace signals plus a bounded recent activity episode. It does not claim to send a large address's entire indexed lifetime.

If selected input exceeds one safe model pass, Memoir partitions the evidence into bounded passes and synthesizes the structured results. Model context metadata and a conservative output allowance determine the budget; an oversized selection is rejected rather than launching unbounded paid calls.

AI output rules

  • Direct observations and interpretations remain separate.
  • Claims cite the Evidence or Signal IDs provided to the model.
  • Unknowns, source failures, spam-token possibilities, disputed labels, and alternative explanations remain visible.
  • A real person or organization appears only as a source-attributed candidate when supplied public evidence explicitly makes that attribution.
  • Behavioral resemblance cannot establish a real identity, sensitive trait, criminal intent, physical location, or common ownership.
  • Structured output is validated before it becomes a report; invalid responses become archived failures with their evidence snapshot preserved.

Identity, wallet screening, KYC, and KYT

Memoir performs public-address investigation and wallet screening; it does not perform KYC. KYC verifies a customer through identity and institutional records that are not established by public address activity.

Memoir is also not currently a continuous KYT monitoring system. It does not provide always-on transaction monitoring, configurable compliance rules, real-time alerts, case disposition, or a regulatory decision trail.

A source-attributed address label is a lead with provenance—not proof that the current user, controller, or beneficial owner has been verified.

Grouping and relationships

Reliably adjacent events may be grouped for readability, while uncertain records remain ungrouped. Grouping organizes facts; it does not infer purpose. Relationships are calculated from loaded direct interactions and per-asset flows. Memoir does not automatically infer common control or expand an ownership graph.

Read source-specific boundaries and the current coverage matrix before relying on an absence or label.

Inspect facts before interpretation.Investigate address