Canonical: https://solanashreds.xyz/blog/alpenglow-for-indexers

# Alpenglow, seen from an indexer's chair

Alpenglow changed how Solana agrees on blocks, and an indexer feels it in three places: its status column, its vote handling and its block parser. Here is what to change, and what you can leave alone.

For an Alpenglow indexer, three things changed. Vote transactions no longer appear in blocks from slot 459,653,000, finality comes from certificates and makes confirmed equal finalized, and a deshredded block is now a sequence of block components, and each block carries a single tick. Transaction formats, entries and Turbine delivery of shreds stayed the same.

## Two dates an indexer should keep apart

Alpenglow's feature gate activated at slot 459,648,000, the first slot of epoch 1064, on 21 October 2026 (about 13:36 UTC). The new consensus, Votor, took over roughly 5,000 slots later. Vote transactions stop appearing in blocks from slot 459,653,000, at about 13:53 UTC.

That gap matters when you compare history. Slots between the activation and slot 459,653,000 still contain vote transactions. The last block holding votes is slot 459,652,999, the next few slots have no block, and the first block after the change, slot 459,653,008, holds none. Use the vote boundary, not the activation slot, when you split statistics into before and after.

## Finality: what your status column means now

Under Alpenglow, a block becomes final through certificates that validators produce off chain, and the commitment levels collapse: confirmed and finalized now mean the same thing. If your schema distinguishes those two states, the distinction no longer adds information for new data.

Keep the column, but change what you rely on:

- **New rows:** treat "confirmed" as final. Waiting for a separate "finalized" step only delays the upgrade.
- **Historic rows:** keep their original labels. A row marked confirmed in a Tower-era block was not the same guarantee.
- **Skips:** a slot can still be skipped, so a row decoded early from shreds still needs its own "seen" state until RPC confirms it.

## No votes in blocks: volumes, filters and history

Before the change, vote transactions made up a large share of every block. The last block with votes, slot 459,652,999, carried 649 of them among 1,051 transactions. Now they are absent, which changes several things an indexer computes:

- **Transaction counts and rates** drop sharply at the boundary. Any chart that counts all transactions needs a note or a split series.
- **Vote filters** in your decoder now do nothing on new data. Keep them for replays of older captures, where they still matter.
- **Throughput sizing** based on old blocks overstates what you receive today, because part of that volume is gone.

## The deshredded block has a new shape

Alpenglow also retired proof of history as a clock. Each block now closes on one tick. What deshredding hands you is a list of block components: a header marker carrying the parent slot and block id, entry batches that each begin with an 8-byte count, and a footer marker. A count of zero signals that a marker follows.

An indexer that parsed deshredded bytes as a plain run of entries needs one change: read the 8-byte count, branch on zero, and parse the markers. Inside each batch, entries look exactly as before.

## What did not change

Most of your pipeline is untouched:

| Part of the pipeline | Changed by Alpenglow? |
|---|---|
| Shred headers and erasure sets | no |
| Delivery of shreds through Turbine | no |
| Entry structure (hash count, hash, transactions) | no |
| Transaction formats (legacy, v0, v1) | no, v1 has its own activation |
| Lookup table resolution | no |
| Instruction decoding by program | no |

Transaction v1 arrived separately, at epoch 1035, and the move to 200 ms slots took effect on 9 October 2026. Both landed in the same autumn, so check which change explains a shift before you blame Alpenglow.

## Rotor is not on mainnet yet

The Alpenglow proposal pairs Votor with a new block propagation layer, Rotor. Rotor is not live. Shreds still travel through Turbine, so receivers and decoders need no change for it today. When Rotor does ship, expect the delivery path and possibly the shred handling to change, and re-test your receiver against it before trusting old assumptions.

## A checklist for an indexer built before Alpenglow

1. Store the vote boundary (slot 459,653,000) and split vote-sensitive metrics there.
2. Treat confirmed as final for rows from after the change; leave historic labels alone.
3. Teach the block parser about block components and zero-count markers.
4. Keep vote filters for old captures, but stop counting on them for new data.
5. Recheck capacity numbers that were sized on pre-change blocks.
6. Keep the early "seen" state for rows decoded from shreds, since skips still happen.

## Caveats on this summary

This page covers what an indexer sees in data, not how Alpenglow reaches agreement. Block times quoted here are second-resolution estimates, so treat "about 13:53 UTC" as approximate. Protocol details can change with later feature gates, and Rotor in particular will need a fresh look when it ships. Check the linked proposals and the Anza docs before you change production code.

## Related reading

To see what votes leaving the block does to your stream, read [votes leaving the block](https://solanashreds.xyz/blog/votes-left-the-block). For the status model this page assumes, see [provisional rows](https://solanashreds.xyz/blog/provisional-rows), and for keys that survive skipped slots, [slots and block height](https://solanashreds.xyz/learn/slot-vs-block-height). [Reconciling decoded rows](https://solanashreds.xyz/docs/pipeline/reconcile) turns "seen" into "confirmed", and [200 ms slots and database writes](https://solanashreds.xyz/blog/200ms-slots-and-your-database) covers the other autumn change. If timing is critical for your pipeline, compare that with [Ultra Low Latency Shreds](https://solanashreds.xyz/ultra-low-latency-shreds).
