November 9, 2026 • 8 minute read

BAM Changelog — October 23 – November 6, 2026

BAM Team Avatar
BAM Team
Block Assembly Marketplace
BAM Changelog — October 23 – November 6, 2026

Two weeks after Alpenglow, the work turned to hardening. A mandatory upgrade, Jito-Solana v4.4.1-jito and FireBAM v26.10.6, landed ahead of a consensus change at Epoch 1079. We found and fixed an upstream Agave bug that cost validators their first leader window each epoch. BAM Node v0.4.2-rc.0 rolled out across the mainnet fleet, and we initiated testing of TEE attestations on every BAM Node. We also announced the first third-party BAM Node operators.

At a Glance

* Required now: Jito-Solana v4.4.1-jito, or FireBAM v26.10.6. Older versions fork off since Epoch 1079.

* Keep at least 2 SOL in your identity account. If it can't pay for Jito's tip setup, BAM leaders produce empty blocks.

* Under Alpenglow, keep at least 2.5 SOL in your vote account, or you silently drop out of the next epoch's stakes.

* BAM secures 164.5M SOL across 382 validators, 37.8 percent of network stake. FireBAM reached a record 21 validators.

* First third-party BAM Node operators announced: Triton One, Bitwise and SOL Strategies.

* BAM Node v0.4.2-rc.0 is live across mainnet, and TEE attestation testing has started on every BAM Node.

Validator Action Required

1. Run a supported version

Client

Minimum Version

Jito-Solana (Agave / BAM)

v4.4.1-jito

FireBAM (Firedancer)

v26.10.6

Frankendancer

Not supported

The loaderv3_programdata_size_floor feature activated at the start of Epoch 1079, on November 6 at about 13:34 UTC. Validators on earlier versions fork off, whatever their client. Jito-Solana v4.4.1-jito is based on Agave v4.4.1, and FireBAM v26.10.6 on Firedancer v26.10.6. The v4.3 line has no release with this change, so it is no longer supported.

By Epoch 1079, every FireBAM validator ran v26.10.6, and all AgaveBAM stake ran v4.4.0 or later. If you are still on v4.4.0-jito, upgrade now.

Reminder: never run a production validator built with dev-context-only-utils (DCOU).

2. Keep your identity account funded

Your identity account pays for Jito's tip program transactions:

• Once per epoch: creating that epoch's tip distribution account, about 0.0015 SOL of rent plus a 0.000005 SOL fee.

• Per leader rotation: a 0.000005 SOL fee to move the tip receiver back to you, if another leader changed it.

The rent goes to your vote account when the account expires, not back to the identity, so the identity balance only goes down.

If the identity can't pay, BAM schedules no transactions for your leader slots, so you produce empty blocks. From v4.4.1-jito, the validator reports this as an error-level metric, bam_scheduler-tip_programs_error, with the slot, the error and your identity balance. Earlier versions logged only repeated should_init_tip_distribution_account=true lines at INFO. On the non-BAM Block Engine path, bundles are skipped instead, and you lose MEV tips (tip programs error, not processing bundles).

Keep at least 2 SOL in your identity account. Set any sweep scripts to leave that amount. Never sweep the identity down to the rent-exempt minimum. See Identity Account Funding in the validator docs.

3. Keep your vote account funded (Alpenglow)

Under Alpenglow, every staked vote account pays about 0.8 SOL of vote account tax (VAT) per epoch, and the minimum balance is about 0.83 SOL. A vote account below that silently drops out of the next epoch's stakes, with no rewards and no error. A vote withdraw of "ALL" does not enforce this minimum. Keep at least 2.5 SOL in your vote account.

4. Rotating your identity? Follow the documented procedure

An identity key rotation can leave the tip distribution account set up for the wrong key. That leads to empty blocks on BAM and lost MEV rewards for the following epoch. Fund the new identity before the swap, and follow the key rotation guide. It is written for Firedancer, but the steps apply equally to Agave. From v4.4.1-jito, a rotated identity that can't set up tips also shows up in the bam_scheduler-tip_programs_error metric.

Validator Health Under Alpenglow

Vote credits and tower vote lag no longer measure health. Since Epoch 1064, votes are not in the ledger, and epochCredits tracks reward lamports, which mostly reflects your stake. Watch these instead:

1. Skip rate per leader window. On v4.3.2-jito.0 or later, the first-window loss should be gone. Escalate if windows keep failing.

2. Revenue and CU per leader slot. Count skipped slots as zero.

3. Votor log lines. Finalized ... fast: and Block Notarized are healthy. Block notar-fallback, Voting skip and TimeoutCrashedLeader need a look.

4. Vote and identity balances, as above.

5. Vote delivery from peers. Staked validators report consensus_vote_metrics. You can vote every slot and still reach only part of the cluster.

6. Rewards ratio. Compare your reward lamports to your expected stake share of inflation. Jito Steward now scores validators this way.

Median slot duration held at exactly 200ms in every epoch since Alpenglow activated.

Releases

Jito-Solana

• v4.5.0-beta.0-jito (November 9, testnet): the first release on the v4.5 line, based on Agave v4.5.0-beta.0. For testnet only; on mainnet, stay on v4.4.1-jito.

• v4.4.1-jito (November 5): required. Based on Agave v4.4.1, with the ProgramData size floor that activated at Epoch 1079. It also reports BAM tip program failures as an error-level metric, rejects BAM TPU ports above 65529, and fixes Tip Router snapshot finalization.

• v4.4.0-jito (October 29): ready for mainnet adoption. Includes the 200ms-slot scheduler pacing fix. v4.4.0-rc.1-jito (October 27) was the 10 percent canary.

• v4.3.2-jito.0 (October 26): fixed the upstream Agave bug that under-packed blocks at epoch rollover. Also frees pruned banks in PoH, moves webpki crates off distrusted roots, and adds an account-size limit to simulateBundle.

FireBAM

• v26.10.6 (November 5): required. Based on Firedancer v26.10.6, with support for the loaderv3_programdata_size_floor feature.

• v26.10.5 (October 28): fixes a rare crash at the epoch boundary, makes voting connections more reliable, and packs blocks faster. The default shred_tile_count is now 2.

BAM Node

• The BAM Node moved to the 4.4 line. v0.4.2-rc.0 (TEE v0.24.3) rolled out to every mainnet region between November 4 and 6.

Performance

• Small blocks: our write-up, BAM vs. kcompactd: How We Fixed Small Blocks With Two sysctls, explains how host memory compaction was slowing PoH hashing by 7–17 percent. Two host sysctls fixed it. Small blocks fell about 77 percent, from 1 in 800 to 1 in 3,600, and PoH jitter dropped from 9–12 percent to about 1 percent.

• Faster execution: BAM Node v0.4.2-rc.0 pins worker threads to CPU cache domains and adds a background bundle pre-simulation filter, which drops bundles that can't succeed before they reach the auction.

• Storage: all mainnet BAM Nodes now run on XFS ledger and accounts disks.

• FireBAM vs AgaveBAM: across Epochs 1065–1079, FireBAM blocks averaged 95 percent of AgaveBAM's CU per block (block-weighted).

Reliability

• Agave epoch-rollover bug: Jito found it, sent the fix upstream to Agave, and shipped it in v4.3.2-jito.0. Volunteer operators confirmed the fix at the Epoch 1069 boundary before release.

• Rolled-back rounds and bank replacement: the BAM Node now suppresses stale commits from rolled-back auction rounds. It also handles nonce checks correctly when a bank is replaced within the same slot.

Network Adoption

Metric (Epoch 1079)

Value

BAM stake

164.5M SOL (37.8 percent of network)

BAM validators

382

AgaveBAM

361 validators, 156.0M SOL

FireBAM

21 validators, 8.4M SOL

BAM stake rose from about 161.5M SOL in our last update, and the validator count reached 382. FireBAM more than doubled to 21 validators, a new high.

BAM Registry and Attestations

Attestations. BAM Nodes are now able to serve TEE attestation reports over HTTPS, bound to each node's TLS key. The endpoint ran on testnet first, was switched on for the Dallas, London and Salt Lake City mainnet BAM Nodes on November 4, and reached the rest of mainnet with BAM Node v0.4.2-rc.0 by November 6, serving to test the attestation service across the entire BAM Node fleet.

BAM Registry. The registry client is deployed on every mainnet BAM Node and is off by default, and the registry service has moved into Jito's production infrastructure. Mainnet enrollment will follow attestations. FireBAM's registry support is in review.

Third-Party BAM Node Operators

We announced the first cohort of third-party BAM Node operators: Triton One, Bitwise and SOL Strategies. Read the post: First Third-Party Node Operators Join BAM.

• Phase 1, testnet: underway. All three have hardware in place, and Bitwise's testnet node is online.

• Phase 2, early mainnet: operators' nodes serve only their own validators. Targeted before the end of the year.

• Phase 3, full mainnet: open to any validator. Targeted for early Q1 2027, once the BAM Registry and attestations are live.

Want to run a BAM Node? Join the waitlist at bam.dev/node-operators.

Ecosystem

• Direct Bundle Ingest: onboarding continues by application. A Block Engine UUID or pubkey does not grant DBI access. New controls let us cap Block Engine intake while DBI scales, enabling additional searcher onboarding and increased rate limits based on performance.

• Preconfirmations: the revenue-distribution keeper is live. Validators are opted in automatically. Revenue arrives as a priority fee on a memo transaction, separate from BAM Boost.

BAM Boost

2,152.4 SOL was distributed to BAM validators over Epochs 1065–1079, about 143 SOL per epoch. Epochs 1078 and 1079 paid about 197 SOL each, the most since Epoch 1064. Check your claims on the JIP-31 Subsidy Program Rewards page.

Operator FAQ

My validator forked off at Epoch 1079. What happened?

The loaderv3_programdata_size_floor feature activated at the start of Epoch 1079. Validators not on Jito-Solana v4.4.1-jito, FireBAM v26.10.6 or the matching Agave or Firedancer releases can't follow the cluster. Upgrade and restart.

My BAM validator produced empty blocks. What's wrong?

Check your identity balance first. If the identity can't pay for that epoch's tip distribution account, BAM schedules nothing for your leader slots. On v4.4.1-jito this appears as the bam_scheduler-tip_programs_error metric, with your identity balance. Earlier versions logged only should_init_tip_distribution_account=true at INFO. Keep at least 2 SOL in the identity.

Vote credits look strange since Alpenglow. How do I tell if my validator is healthy?

Credits now track reward lamports, not votes. Use skip rate per leader window, revenue per leader slot, Votor log lines, and your vote and identity balances. See "Validator Health Under Alpenglow" above.

On BAM, my TPU and forwarding counters are near zero. Is that expected?

Yes. While you're connected to a BAM Node, your advertised TPU is the BAM Node's TPU, so user transactions go there. Traffic that still reaches your own TPU is not passed on to BAM.

I'm moving my validator, or a failover, to another region. What changes for BAM?

Update --bam-url, --shred-receiver-address and, if you use a regional one, --block-engine-url. Your TPU updates automatically once you connect. Regional endpoints are listed at bam.dev/validators.

Can I switch back to FireBAM?

Yes, on v26.10.6. Frankendancer does not support Alpenglow and is no longer supported for BAM.

(Searchers) The Block Engine says my bundle failed simulation in every region, but it landed. Why?

When the upcoming leader runs BAM, the Block Engine forwards your bundle to that leader's BAM Node right away, in addition to running its own simulation. The Block Engine simulates against its own view of the chain, which can lag the leader, so it can report a failure while the BAM copy lands. One regional submission is enough to reach a BAM leader.

(Searchers) Does my Block Engine UUID give me DBI access?

No. DBI is a separate application, and onboarding is gradual. Until you have access, Block Engine bundles for BAM leaders are still forwarded straight to the BAM Node.