The Block Engine and 5-tx bundles did not get replaced last week. The new layer is BAM preconfirmations, live 9 September 2026, sold through Helius and Triton.
That is the update that matters if you ship Jito packs for Pump.fun. The rest of mid-September is product and treasury noise around it.
What shipped
BAM (Block Assembly Marketplace) has been on mainnet since late 2025. It is the TEE / plugin path for how a leader builds a block — not just “pay a tip, hope.”
Preconfirmations are the consumer API on that path: you get an early signal that a tx is committed in the BAM pipeline before the slot is public on gossip. Coverage at launch: about 34% of stake, ~383 / 665 validators, ~17 BAM nodes. Independent scrapes around 21 Sep still sit near 34.6%.
Latency claim from the rollout writeups: on the order of 5–10 ms p50 versus waiting for the shred. That is searcher time, not retail UI time.
Revenue split on the preconf product:
| Slice | Who |
|---|---|
| 30% | Helius / Triton (they bill you) |
| 35% | BAM validators (stake-weighted, first checks October 2026) |
| 35% | Jito DAO |
Opt-out today is manual (Discord). Self-serve comes later. Agave-BAM and FireBAM both count.
Q3 BAM subsidy (100% of protocol revenue to the adopter pool under JIP-37) hard-stops 30 September 2026. After that, value accrual is supposed to flip back on. Do not mix “preconf fees start in October” with “subsidy still pays the client.” Different pipes.
What did not change
- Bundles are still ≤5 txs, atomic, sequential, one slot.
- You still tip a current tip account.
- Block Engine still sees ~98% of stake even where BAM share is a third. Classic
sendBundleis not dead. - Maker Priority Plugin (Q2) is a separate BAM plugin for app-defined ordering. Preconfs are a data product. MPP is an ordering product.
Why a Pump.fun bundler should care
Detect vs land is now two markets.
Preconf subscribers see committed orderflow earlier. If you only watch shreds / Geyser, you are late to people who paid Helius/Triton for the BAM firehose. Same create you were going to backrun may already be “preconfirmed” into a pack you cannot reorder.Your bundle can be preconfirmed too.
If you submit on a BAM-aware path, you get a faster yes/no than waiting for the explorer. Failed packs stop burning the next tip one slot sooner. That is operational, not alpha.Leaders are not uniform.
~35% stake is BAM. The other ~65% is still classic Jito-Solana / others. A bundle tuned only for BAM nodes misses the leader schedule most of the day. Fan-out: Block Engine regions and the preconf pipe, then accept whichever slot actually fires.Subsidy cliff is 8 days after this post’s week.
When JIP-37 ends, validator economics on BAM change. Expect tip floors and preconf prices to reprice in October when validator checks actually start. Do not hardcode September tip tables.
What else moved in the last 10 days (not the bundler ABI)
-
21 Sep: JitoSOL (
J1toso1uCk3RLmjorhTtrVwY9HJ7X8V9yYac6Y7kGCPn) listed as collateral on Bullet. LST stays staked while you margin. Unrelated tosendBundle. - Same week: Rettig / Fortune “onchain finance” piece; Jito still hiring for JTX (self-custody trading app that public-launched in July).
- ~6M JTO unlock around 21 Sep. Token supply ≠ Block Engine.
- GitHub:
jito-solana,jito-tip-router,stakenetstill pushing 22 Sep. Read the client tag you actually run; do not assume preconfs changed tip-account layout.
Practical takeaway
Latest Jito update = BAM preconfs are a paid, partner-gated look at committed BAM orderflow, covering a third of stake, billed by Helius/Triton, validator paid in October.
If your Pump.fun bundler still only does:
simulate pack → tip → sendBundle → wait for signature
you are on the 2025 path. The 2026 path is: sequential curve quotes (still the thing that reverts the atom) plus knowing whether the next leader is BAM, and whether your competitors already saw the create on the preconf stream.
Bundles did not get a sixth transaction. They got a faster rumor about who already won the five.
Telegram: https://t.me/C00LBB
Site: https://coolbb.site
Top comments (0)