Under Alpenglow, Solana votes are off chain: validators send them to each other outside the block instead of as vote transactions. Blocks have carried no vote transactions from slot 459,653,000, about 13:53 UTC on 21 October 2026. Vote filters in your decoder now remove nothing, and any metric that counted votes changed meaning at that slot.
Vote filtering code that is now dead weight#
Before the switch, most transactions in a block were votes. In slot 459,652,999, the last block that still held them, 649 of its 1,051 transactions were votes, according to getBlock reads around the boundary. Decoders built in that era usually drop votes as the first step after extracting transactions, because decoding them was wasted work.
On current data that step finds nothing. It does no harm, but it is worth knowing what it is now doing, for three reasons:
- Benchmarks. If you sized your decoder by "transactions per second including votes", the input it sees today is a different mix. Re-measure on current traffic before you trust the old numbers.
- Counters. A metric such as "votes dropped per slot" now reads zero forever. An alert built on it, for example one that fires when the vote share falls, will either never fire or fire constantly.
- Replays. If you replay old captures through the same decoder, the filter suddenly matters again. Keep it, test it on an old capture, and label it as a filter for historical data.
Vote messages are also gone from gossip in an Alpenglow epoch, so a tool that watched votes through gossip has lost that source too.
Metrics that quietly changed meaning#
These counts move at the boundary even though nothing about user activity changed on that day.
| Metric | Before slot 459,653,000 | From slot 459,653,000 |
|---|---|---|
| Transactions per block | Mostly votes, plus user transactions | User transactions only |
| Fees collected per block | Included the base fees paid by vote transactions | User fees only |
| Signatures per block | Inflated by one or more signatures per vote | User signatures only |
| Share of failed transactions | Diluted by a large number of votes | Measured against user transactions only |
| "Non-vote transactions" filter | Removed most of each block | Removes nothing |
If a chart in your reports spans the boundary, a step change at 21 October 2026 is the network changing format, not your users changing behaviour. Annotate it, or plot non-vote counts only on both sides so the series stays comparable.
Historical data: before and after the switch#
The cleanest way to think about it is that you now have two eras in one table. Rows from before the boundary include votes unless you filtered them at ingest; rows after it never do.
Two details make the boundary messier than one slot number suggests. First, the slots from 459,653,000 to 459,653,007 produced no block, and the first block after the gap, slot 459,653,008, already held zero votes. Second, the feature that turned on Alpenglow activated earlier, at slot 459,648,000. That is the start of epoch 1064, and the activation slot is not the moment votes left the block. Use 459,653,000 as the boundary for vote counts; use the activation slot only when you mean the protocol change itself.
If your dataset mixes captures taken from shreds with rows taken from an RPC node, check that both sides apply the same vote rule. An RPC backfill that excludes votes joined to an old shred capture that includes them produces numbers that are wrong in both directions.
How to mark the boundary in your dataset#
- Add a column, or a derived view, that labels each row with its era:
slot < 459653000is the vote era, everything else is the current era. - Store a constant for the boundary slot in one place in your code and reference it, rather than repeating the number in queries.
- For every metric that counts transactions, signatures or fees, decide whether it should include votes in the old era. Most should not; write that decision next to the metric's definition.
- Recompute historical aggregates with the rule you chose, so the series is consistent across the boundary.
- Keep the vote filter in your decoder, and add a test that runs it against a capture from before the boundary.
- Annotate dashboards and reports at 21 October 2026, with the slot number, so readers do not mistake the drop for a change in usage.
Questions about votes and the stream#
Are Solana vote transactions still in blocks?#
No. From slot 459,653,000 onward, blocks contain no vote transactions. Validators still vote under Alpenglow, but those votes travel between validators instead of being recorded as transactions in the block.
Does this make the shred stream smaller?#
The stream no longer carries vote transactions, so blocks hold only user activity. How that changes the packet count on your receiver depends on user traffic, so measure it on your own server rather than relying on a general figure.
Can I still tell when a block is final?#
Yes, but not from the shreds. Under Alpenglow, finality comes from certificates. Check the status of a transaction or slot through RPC at the commitment you need; confirmed and finalized now mean the same thing.
What this change does not mean#
Votes leaving the block does not make shred data any more final than it was. A transaction seen in shreds can still fail, and its slot can still be skipped, so your provisional and confirmed states are as necessary as before. It also does not change how you decode user transactions: the formats, the account keys and the instructions are the same on both sides of the boundary.
Related reading#
Alpenglow for indexers covers the other changes that arrived with the new consensus. The transaction structure page explains what a user transaction looks like once the votes are out of the way, and slot numbering explains why the slot, not the block height, is the right key for an era column. If you are building a dataset that spans both eras, building a research dataset and versioning event schemas help you keep it honest, and the decoding pipeline shows where the vote filter sits. To see what a current block looks like in raw form, compare the plans and start with Standard.
Sources#
- SIMD-0326, Alpenglow (votes move off chain) (checked 2026-11-18)
- Agave v4.4.1 gossip CRDS filter (vote messages dropped in an Alpenglow epoch) (checked 2026-11-18)
- Solana RPC getBlock (used to read the blocks around the boundary) (checked 2026-11-18)
- Anza docs: commitment levels under Alpenglow (checked 2026-11-18)
Raw shreds on your own server
Ultra Low Latency Shreds is $200 and Standard Shreds is $100. You pay in crypto, tell us which IPv4 address and UDP port should receive the stream, and we wire it up by hand.