This explorer shows testnet-12 (R-core+, ADR-0152 v3.1), launched 2026-09-25 about 23:15 JST from release 0e8ec984e. Since the DAA-3,600 flag day (2026-10-02) a node on this network prints params fingerprint 254509533bb693ced0fed823a4c25e166ba2542d576e4021b0e4b4d6fe4079e1 and fence schedule 750, 1000, 1300, 1700, 2000, 3600 (schedule id 1e39c738…). The fleet runs node-only updates of that ruleset (int-10.5: panel throughput fixes; the fingerprint is unchanged).

Next flag day: DAA 5,300 (≈ 2026-10-05 afternoon JST). It arms the generative model classes, model self-improvement, layer-sharded panels, bundled verification certificates, the inference-surface additions, the claim-capacity step-up, and the useful-work transition (verifier silence no longer slashes an honest producer; PALW-BASE-0 becomes a reserve; real-model work carries the clock). The release and its new fingerprint will be announced here before DAA ~5,200; a node that has not updated is refused by upgraded peers from DAA 5,300 — if it stays connected across the fence it does not see the refusal: it stops at DAA 5,299 until it restarts, so check that your node is past 5,300 after the flag day. Genesis a27f8f44fe4d91a5…; the eight genesis bonds are outputs 0–7 of premine txid 5e0d5f1b….

Join. Build origin/main (0865f1dec or later) and run kaspad --testnet --netsuffix=12; the CLI is misaka --network testnet-12 (its default). Peers come from the DNS seeders seeder1–4.misakascan.com; the P2P port is 26311. A datadir from an earlier testnet-12 genesis is refused at startup — move it aside. A main build no longer joins testnet-11 (the previous network; build 1f98d3bf4 for it). The faucet has no testnet-12 funding yet. Producing or verifying: docs/testnet12-join-mining.md.

Known issues — details in the launch note (in Japanese). The two CRITICAL ones were fixed by the activation fences at DAA 750 (release c3dbaee3c).

• Fixed at DAA 750 — heartbeat transparency (was CRITICAL): a heartbeat miner with no bond could pull published attempts into a private chain and reorg within the merge depth — a double spend.
• Fixed at DAA 750 — panel-seed grinding (was CRITICAL): the panel seed is now the anchor attempt's execution commitment. As a stopgap, only the eight genesis bonds' attempts anchor a panel, so the panel draw trusts the operator for now; per-operator panel skew is monitored.
• Fixed at DAA 750 — full seats: Valid locks after Final filled the genesis seats' 500‰ room from about DAA 624, and claims whose panel could not be drawn were voided (their escrowed worker reward is destroyed, not paid; no collateral was taken). From DAA 750 those locks count against the whole collateral, and a lock and its slashable window end at Final + 1,000 DAA for claims licensed from DAA 750 (earlier locks keep licence + 3,000).
• Also at DAA 750: registry resilience (a claim waits for its bind deadline instead of voiding), bond maturity for non-genesis bonds, model sinks bound to objects, a strict economic win for reorgs (ties within 2 DAA ticks follow GHOSTDAG order), the same rule for pruning-proof IBD, mergesets accepted parents first, slashing evidence that cannot be verified no longer spends stake, and an operator bond at its exposure ceiling can still anchor.
• Fix at DAA 1,300 (release 587cab2b0): a claim whose panel draw the eligible-stake floor refuses is re-anchored at its next slot instead of voided (DAA 600–749 lost 540,943 MSK of escrowed reward to such voids), and seat locks written before DAA 750 take the F + 1,000 life, so the seats' room does not run out again.
• Fixed in the node (c3dbaee3c, kept in 587cab2b0): IBD past DAA 316 (older builds cannot sync a fresh node), unlicensed claims (V04: collectors now license in bind-age order), a remote header crash and pruning-proof IBD hardening.
• DNS finality is in Bootstrap: validators are funded after launch; until then the DNS reorg gate does not act, and the EVM bridge's deposit claims wait for it.
• 8k readiness: a stalled 8k seat can drop the class to Held, which voids its unbound claims. Partly fixed; the rest follows by fence and node update.
• Scheduled fixes: the bridge audit (BR-1–BR-7), position fixes.

When is a payment final? Never count blocks, blue-score depth, DAA or heartbeats — heartbeats can inflate all of them, and a stretch with only heartbeats confirms nothing. Small amounts: on your node's selected chain, a blue attempt block at least 30 blue score above the block carrying the transaction, with attempts still following. Large amounts: finality depth 600 blue (about 4 hours), or a Final PALW anchor (getPalwSettlement / misaka palw settlement).