the vertical slice of the phase-1 root chain — 2026-10-16

One block from genesis to a second machine, locally: the consensus kernel (#189), the reference miner (#186) and the wire (#187) in one binary, three nodes on one machine, every block mined by soft3 mine --backend cpu against a node's /template + /submit, every header, body and transaction crossing the radio wire to the other nodes and validated there by the kernel. Requirements: specs/chain.md, specs/chain-wire.md, specs/chain-conformance.md. Earlier receipts: audit/chain/net-2026-10-15.md (the wire on its stand-in), audit/chain/mine-2026-10-15.md, audit/chain/conformance-2026-10-16.md.

what changed

• main: #186 merged on the kernel (4527df7, merge of feat/chain-mine-on-kernel = #186 + the swap f1f4e94: the miner asks chain::{header, pow, daa}, its local ASERT copy is gone).
• #187 brought onto main as one squashed commit (the branch carried a merge of main whose own changes a commit-by-commit rebase would drop), then:
  • chain/net/devchain.rs deleted; chain/net/kernel.rs implements the wire's port ChainView over the kernel's ChainApi: header-first acceptance, NeedBodies (the wire fetches the named bodies from the peer and resubmits, gossip::apply_fetched; IBD takes the gap by ranges), FutureTime held and retried each sync tick, DeepFork / Checkpoint logged and never scored, ChainError::Block scores the serving peer +10, tip-moved events carry the reorged-out bodies, a main-chain txid index for GET /tx/{txid}. Headers and u256 cross as their wire bytes; a test pins header ids, txids, bodies and tx roots on both codecs.
  • mempool admission asks the kernel (validate_tx with the neuron's lower pending nonces); a nonce gap inside the window is not scored, a nonce beyond it is (+10). The kernel's validate_tx now checks the signature before nonce, fee and balance — chain-wire §4's order; the adversarial tx flood (unsigned junk at fee 0) was going unscored without it.
  • HTTP §7 on the kernel: /tip and /template rendered by Tip::json / Template::json; /nonce {neuron, nonce, confirmed}; POST /tx {txid, kind, size, fee, account, nonce} (already present = 200); GET /tx/{txid} {txid, kind, tx, height|"mempool", block, index}, 404 {"error":"unknown"}; POST /submit 200 / 409 stale / 422 with the kernel's reason. /status genesis is the chain id (what the wallet signs for), genesis-block, next-bits from ASERT, finalized.
  • the node no longer mines (--mine/--solo/--miner, /dev/mine gone): mining is soft3 mine. soft3 chain genesis-dev writes a real GenesisFile (kernel-checked, launch names refused), soft3 chain key|tx sign with soft3::wallet::Signer.
  • gossip topic loops survive a lagged receiver instead of ending.
  • [profile.dev.package.cyber-hemera] opt-level = 3: tests mine real blocks at the floor target on debug builds.
  • scripts/chain-fleet.nu: miners are soft3 mine processes; every pid it spawns is recorded in $work/pids and killed by pid — the pkill -f is gone.

Test chain: dev, bits = POW_LIMIT_BITS (0x1e100000, 2^20 expected hashes, the easiest target the kernel accepts outside cfg(test)), genesis time two hours back so ASERT sits at the floor for the first ~240 blocks. No launch genesis is touched; genesis-dev refuses pussy, bostrom, pussy-rc.

Hash: hemera 0.3 (hemera d5f0a09). Every hash on the slice goes through the hemera API (hemera::hash, the kernel's hash32, the miner's sponge view); the move to hemera 0.4 (profile #15) re-blesses PoW vectors, the genesis particle and chain.snap in its own sweep.

stand

~/cyber/.stands/N2, Apple M4 Max, macOS 26.4.1. soft3 on slice/net over main 4527df7. Siblings at the kernel receipt's integration base (bbg 1d39975, cybergraph 0eae7ae, foculus 9e32aa6, mudra 4bfe3e8, zheng 0c882e7, joy 06de02c), lens e11eaa8 (release/0.2), nox 2f09ca3 / trident a3cef15f (release/0.4), hemera d5f0a09, radio 50600cc1, tade c7d404a. Deviation from the requested pins: zheng release/0.4 and bbg master do not build with soft3 main (bbg master lacks the vendored fjall the chain store uses; bbg 1d39975 needs zheng's feat/public-certificate-v3 API), so both stay at the kernel receipt's base.

one block from genesis to a second machine

cargo build --release --locked --bins                           # 1 m 34 s incremental
nu scripts/chain-fleet.nu --bin .target/release/soft3 --keep

chain-fleet: 3 nodes · chain dev · binary …/.target/release/soft3
  ok    binary speaks the phase-1 flags
  ok    ports free: http 17780,17781,17782,17783 radio 17790,17791,17792,17793
  ok    homes + genesis + peers under /tmp/chain-fleet.wVWUzs; wallet aaaeb678a…
chain-fleet: launched pids 74396 74398 74400
  ok    all 3 nodes answer /status
chain-fleet: soft3 mine --backend cpu --threads 1 on node-1 (pid 74421) and node-2 (pid 74423)
  ok    peers connected: node-1=1 node-2=1 node-3=2
  ok    block 1 f9a96f3fc444d… mined by 111111111… is block 1 on every node
  ok    heights agree at 7 across 3 nodes
chain-fleet: late joiner node-4 pid 75323
  ok    late joiner synced to height 8 (IBD by ranges)
txid 7d26e8969032a03206a0996e1402f2ad1a0280b4fc8c72af2bc0425edf6e0117
  ok    tx 7d26e8969… submitted to node-1, mined at 10 by node-2's miner 222222222…, bob=5 on node-3
  ok    fork of depth 3 from 10: main 13 reorged to the heavier chain 15, tip 2e253b946… on every node
chain-fleet: node-3 killed and relaunched as pid 76024
  ok    node-3 recovered from disk: height 15 before the kill, 22 after restart (replayed, then synced), conserved
chain-fleet: artifacts kept in /tmp/chain-fleet.wVWUzs
chain-fleet: GREEN

Block 1, end to end (node-1's miner log, then node-2 and node-3):

soft3 mine · chain dev · node http://127.0.0.1:17780 · cpu (1 threads) · to 1111…1111
template height 1 prev 50453f910886… bits 0x1e100000 (2^20 expected hashes)
found nonce 0x7d717dafd5726d8b00000000000ff34b after 1048576 hashes, 3.9s
block 1 accepted: f9a96f3fc444da08fdb30839140110d8c38ce8364922fbe31a4f410de3020000
node-2  net · header f9a96f3f at 1 from 630a8dba
node-3  net · header f9a96f3f at 1 from 630a8dba

An earlier run the same day (/tmp/chain-fleet.Q7vBAc) was GREEN too: heights 11, late joiner IBD of 11 blocks in one range, tx mined at 13 by node-2's miner, depth-3 fork from 14 (main 17 → fork 19), node-3 19 → 20 after restart. After both runs no process of the fleet was left.

the gates

The pussy genesis for the conformance rows was rebuilt from the pinned snapshot with this build into a private directory: soft3 genesis build --chain pussy --snapshot ~/cyber/space-pussy/snapshot --out /tmp/n2-genesis/pussy.tade (4.4 s) → particle 2a83e54f…61b5, state root f2296834…5827, block 0 b863ce28…a304: the blessed vectors. The shared cache ~/cyber/.cache/genesis/pussy.tade was rewritten at 11:07 today by an unmerged branch with a longer params record (pool formats); main cannot decode it (74 trailing bytes in frame), so with the default HOME the two cache-reading tests fail on main and on this branch alike. It was left as it is.

what the deploy to cyberproxy, deimos and jupiter still needs

• a release carrying #187 on the kernel (cargo install soft3 --version <phase-1>), built against pins that build: bbg needs its fjall vendoring on master (or soft3 moves off bbg/rs/vendor/fjall), and the stack-chain-genesis gate needs those 3 warnings gone to turn green.
• the launch genesis files from soft3 genesis build with the final params, the hemera 0.4 re-bless done, and the ids pinned in chain::params::CHAINS; until then --genesis FILE on every machine, byte-identical (step C T−15).
• miners: the nodes do not mine. Someone runs soft3 mine --node http://127.0.0.1:7781 --address <neuron> (Metal/WebGPU/CPU) against at least one node at genesis, or block 1 never comes.
• endpoint ids (soft3 node id) collected into peers.toml on each box, ufw for 7791/udp, nginx snippets (README step A6), jupiter's root steps by the owner (no sudo).
• not exercised here: three real machines over the internet (NAT, relay off, clock skew), a replay at a real height (restart replays every body; tens of blocks took seconds here), and best_header / missing_bodies scanning the in-memory store on each call — fine at slice size, to index before long chains.