The blueprint’s finish line was never “it mines a block.” It was: three or
more nodes sharing one chain; two wallets sending through the CLI; a
deliberate fork resolved by most-work; every Tier-1 security rule covered
by a passing attack test; and an AI assistant that can inspect the chain
and draft payments through MCP — with spending gated on a human running
--confirm outside the conversation.
All of it now runs on committed evidence. The attack lab encodes all fifteen blueprint scenarios — the historic Bitcoin bugs (duplicate inputs, value overflow, the CVE-2012-2459 merkle ambiguity, signature malleability, low-work blocks, coinbase overclaim, premature coinbase spends) and the network/economic attacks (eclipse and diversity limits, malformed P2P payloads, selfish-mining economics, mempool flood with bounded eviction, double-spend races, reorg double-spends, majority-work history rewrite). Every one is rejected the way Bitcoin rejects it.
The fuzz campaign ran ~26.3 million executions across four targets — transaction and block decoding, raw wire payloads, and script execution — with zero crashes. Nine property tests pin the invariants that matter: codecs are fixed points, money arithmetic is checked, merkle roots are deterministic and mutation-sensitive, HD derivation is stable, the mempool stays bounded, and disconnecting blocks restores the UTXO set exactly.
The load run is the workspace standard made literal: one million operations per scenario — codec round-trips, wire framing, full P2PKH script verification, mempool admission under eviction pressure, UTXO validation, and a million authenticated JSON-RPC requests against a live node — all past threshold, report committed with machine and commit.
And the demo is the part you can watch: scripts/devnet-demo.sh stands up
three nodes, converges 120 blocks, partitions one, extends the chain,
reconnects for headers-first catch-up, then isolates a node that mines a
55-block private fork — heavier than the public chain — and watches all
three reorg onto it by most-work. A wallet send of 10 BTC confirms in the
next block on every node. Seventeen phases, seventeen passes.
One honest caveat recorded in the evidence: an earlier demo run died on a transient localhost transport error mid-fork — the script now retries transport flakes (bounded, three attempts) while RPC-level errors still fail fast, and both transcripts are committed. Laptop throughput numbers are development observations, not capacity claims.
What is deliberately not claimed: this is a teaching system, not Bitcoin. It shares the design — UTXO model, compact-size canonical encoding, proof-of-work, difficulty retarget, BIP32/39/44 wallets — but runs its own genesis, its own magic bytes, and its own devnet. No real value, no real network, ever.