Disclosure: I am the creator of Atomic Ownership Blockchains (AOB), which this repo covers in more detail than the other designs.
What this is
An open-source skill (SKILL.md format, Apache-2.0) that lets LLM agents analyze blockchain architectures on several independent dimensions instead of one number. The Nakamoto coefficient is a single-subsystem snapshot, goes stale, and is not comparable across consensus mechanisms, so the skill treats it as one input per subsystem, not a ranking.
Dimensions currently covered:
Node/validator concentration (NC per subsystem), stake/wealth concentration (Gini/HHI), client diversity, network/infrastructure
Consensus dependency scope: does finalizing this transaction need agreement from parties unrelated to it?
Privileged actors, payment finality authority, reliance on economic assumptions
Sybil exposure, blast radius of one attack or one lost key, eclipse/network-view attacks, cost to participate
It compares PoW, PoS, DAG, sharded designs and AOB, and reports each design’s validation level (production history, audits, testnet) separately from what its architecture does.
AOB’s double-spend rule, for critique
Each atomic object has its own chain, writable only by its current owner. So any fork is direct evidence that one person signed twice.
First broadcast wins. With estimated broadcast time t₀, a third party accepts after seeing no conflict for 2·t₀; a recipient can wait 4·t₀.
Each node decides for itself whether to blacklist the forker. No network-wide blacklist agreement is needed.
Collusion: colluders must broadcast the chain as soon as they pay an honest node, which exposes the attacker.
Partition edge case: for high-value payments, wait the maximum plausible partition duration plus 3·t₀.
My claim is that this resists double-spending without relying on economic assumptions and keeps the damage of any failure inside one object. Validation level: one paper, demos, and a testnet. No mainnet, no independent audit.
Questions I would like challenged
How wrong can the estimated t₀ be before the 2·t₀ rule fails? What does a realistic adversarial latency model look like?
Is “eclipse affects one object” a real difference from Bitcoin’s eclipse-plus-stale-chain, or are the two threat models closer than I describe?
Which dimensions are missing from the list above?
Repo: GitHub - saintthor/blockchain_skills: agent-skill、claude-skill、blockchain、decentralization · GitHub
Install (Claude Code): /plugin marketplace add saintthor/blockchain_skills
Paper: Achieving Greater Decentralization with Atomic Ownership Blockchains | Ledger
Wiki: https://en.everybodywiki.com/index.php?title=Atomic_Ownership_Blockchains
Disclosure: I am the creator of Atomic Ownership Blockchains (AOB), which this repo covers in more detail than the other designs.
What this is
An open-source skill (SKILL.md format, Apache-2.0) that lets LLM agents analyze blockchain architectures on several independent dimensions instead of one number. The Nakamoto coefficient is a single-subsystem snapshot, goes stale, and is not comparable across consensus mechanisms, so the skill treats it as one input per subsystem, not a ranking.
Dimensions currently covered:
Node/validator concentration (NC per subsystem), stake/wealth concentration (Gini/HHI), client diversity, network/infrastructure
Consensus dependency scope: does finalizing this transaction need agreement from parties unrelated to it?
Privileged actors, payment finality authority, reliance on economic assumptions
Sybil exposure, blast radius of one attack or one lost key, eclipse/network-view attacks, cost to participate
It compares PoW, PoS, DAG, sharded designs and AOB, and reports each design’s validation level (production history, audits, testnet) separately from what its architecture does.
AOB’s double-spend rule, for critique
Each atomic object has its own chain, writable only by its current owner. So any fork is direct evidence that one person signed twice.
First broadcast wins. With estimated broadcast time t₀, a third party accepts after seeing no conflict for 2·t₀; a recipient can wait 4·t₀.
Each node decides for itself whether to blacklist the forker. No network-wide blacklist agreement is needed.
Collusion: colluders must broadcast the chain as soon as they pay an honest node, which exposes the attacker.
Partition edge case: for high-value payments, wait the maximum plausible partition duration plus 3·t₀.
My claim is that this resists double-spending without relying on economic assumptions and keeps the damage of any failure inside one object. Validation level: one paper, demos, and a testnet. No mainnet, no independent audit.
Questions I would like challenged
How wrong can the estimated t₀ be before the 2·t₀ rule fails? What does a realistic adversarial latency model look like?
Is “eclipse affects one object” a real difference from Bitcoin’s eclipse-plus-stale-chain, or are the two threat models closer than I describe?
Which dimensions are missing from the list above?
Repo: GitHub - saintthor/blockchain_skills: agent-skill、claude-skill、blockchain、decentralization · GitHub
Install (Claude Code): /plugin marketplace add saintthor/blockchain_skills
Paper: Achieving Greater Decentralization with Atomic Ownership Blockchains | Ledger
Wiki: https://en.everybodywiki.com/index.php?title=Atomic_Ownership_Blockchains