The Hive-Knowledge Graph

Impactium is more than a settlement layer for impact. It is the Hive-Knowledge Graph of humanity: a durable, verifiable record of what people and organizations have actually done — a shared memory that both humans and AI can read.

The chain records; the graph projects

There are two halves to how Impactium is read:

• The chain is the single source of truth. It records Entities and proof — every registration, claim, capsule, vote, and cost, permanently and in total order.
• The graph (the product name for it is Karma) is a projection of the chain into a walkable property graph: Entities, Products, Capsules, Articles as nodes; created, spawned, minted, voted, and residual relationships as typed edges.

The clean way to say it: the chain records the entities and the proof; the graph records the walkable projection.

The graph is disposable — and that is a feature

Karma is never asked to be trusted, only to be fast. It is a disposable view: you can delete it entirely and rebuild it byte-for-byte by replaying the chain. Nothing of value lives only in the graph. This is what lets the read layer evolve freely — new questions, new indexes, new shapes — without ever putting the record itself at risk.

Verification on demand

Because the chain commits to its state with cryptographic proofs (ICS23 against a committed state root), any fact the graph shows can be escalated to a proof. A consumer — a person, a dashboard, or an AI agent — can take any answer and verify it against the chain's own commitment, without trusting the graph or a full node. Convenience by default; proof when it matters.

Memory for AI

This is where the "Hive-Knowledge Graph" name earns itself. Impactium is designed to be on-chain memory for AI — foremost for Aionima, the Impactivism oracle AI, and for other agents. A typed, self-documenting graph over a permanent, verifiable record is close to an ideal memory-retrieval interface for an AI: it can walk relationships, filter by scope and verification tier, and prove any entry it relies on.

Two design commitments make this trustworthy as memory:

• Permanence. Impact records are never pruned. An AI's memory must not silently lose entries, so full retention on impact-critical state is a requirement, not an optimization.
• Provenance. Every node and edge carries where it came from — partition root, block hash, transaction hash, and schema version — plus a mapping-confidence flag. Nothing in the memory is unattributed.

What you can ask it

The graph answers the questions the model is built around, for example:

• A member's full impact history — every capsule and cost, including visibly-failed claims (failures are never hidden).
• The downline walk — an Entity, its entire referral Chain of Impact, and the impact created across it.
• Provenance of any fact — the exact chain commitment a claim can be proven against.

Karma also serves the ecosystem's Chain Resources — the on-chain package and document content — and exposes a governance-grade view the network itself uses to weight votes by proximity to a subject. It is a read surface that is also, increasingly, an input to how Impactium governs itself.