The P&L Panel Was Wrong, and Daily Trends Is Back

Rebuilding Lodestar's Daily Trends chart meant defining an indexer's daily rewards carefully, and doing that showed the P&L panel had been getting them wrong. Over the 30 days to 13 September it misstated 54 of the 58 indexers that were paid. Both are fixed today. Here is exactly what was wrong, how we checked the new figures, and what is still to come.

2026-09-14 cargopete Infrastructure

Written 2026-09-14 Status A dated snapshot, left as written

This is a dispatch, not an entry. It records what was true and what its author thought on the day above, and it is not revised as the protocol moves. For facts that are kept current, with the date somebody last checked them, use the registry and the entries. Links in a dispatch are left as the author wrote them, so some no longer resolve. That is what a snapshot looks like after a while, and quietly rewriting them would make the date on this page a lie.

When Lodestar stopped holding a Graph API key on 5 September, the indexer page lost three things: Query Performance, the QoS Quality panel, and Daily Trends. Daily Trends is back today. So is a P&L panel that shows the right numbers, which is the part of this post we would rather not have had to write.

The P&L panel was wrong

The P&L panel on each indexer page shows what the indexer earned over a window, in rewards and query fees, against a cost model. Three things were wrong with how it counted, and all three had been wrong in production.

Rewards were dated by when an allocation closed. Under Horizon an indexer collects indexing rewards whenever it presents a proof, including on allocations that stay open. The panel only counted an allocation’s rewards once the allocation closed, and put them all on that day. Rewards collected on open allocations were either late or missing.

The delegators’ share was counted as the indexer’s revenue. An indexing reward is split between the indexer and its delegators, and the panel showed the whole thing.

Query fees were shown gross. Before a fee reaches the indexer, 1% goes to the protocol, a share goes to curators, and the delegators take their cut. The panel showed the amount before any of that.

Over the 30 days to 13 September, 54 of the 58 indexers that were paid saw a P&L off by more than 1 GRT, with 7.2 million GRT of error across them. Of 24.1 million GRT of rewards collected in that window, the panel had booked 15.7 million. The errors went both ways: one indexer was shown 259% above what it actually received, another at half of it.

If you made a decision based on that panel, we are sorry. It should not have been possible for it to be this far out without anyone noticing, and the fact that it was is a large part of why we now check every figure against the chain before shipping it.

What it shows now

The panel now shows what the indexer actually received, dated by the collection that paid it. Rewards come from the reward events on Arbitrum, split the way the contracts split them. Fees are what is left after the protocol cut (rounded up, as the contract rounds it), the curators’ share and the delegators’ cut. We checked that arithmetic against the chain’s own payment records for 1,990 collections, and it matches every one to the wei.

The panel carries a note saying it was corrected, and when.

Daily Trends shows an indexer’s rewards (indexer and delegator, stacked) and query fees per day, with a running total. It used to read a community subgraph through a gateway key; it now reads the same Arbitrum events through our own nest.

Before shipping it we compared the new daily view, running on the production nest, against that community subgraph for every indexer and every day from 24 July to 22 August: 327 indexer-days, seven figures each (total, indexer and delegator rewards, the number of reward collections, gross fees, the curators’ share, and the number of fee collections). Every one of them matches to the wei.

Fees are now labelled gross, with a separate net-to-indexer series beside them, because the old chart never said which it was showing.

What is still to come

Query Performance and the QoS Quality panel are not back yet. They are rebuilt, and rebuilding them turned up problems with the old versions too, which will get their own post when they ship. The short version: the old charts averaged averages, hid outages inside daily figures, and were built on a subgraph that had silently dropped whole five-minute buckets of data. We would rather ship them late and right.

Checking our work

Both panels read graph-allocations-nest, which is public, and anyone can run it and derive the same numbers:

nuthatch init --from https://github.com/nightswatchhq/graph-allocations-nest

An indexer’s daily figures are lodestar_indexer_daily, and the same figures per deployment are lodestar_indexer_deployment_daily, whose sum is the daily view:

SELECT day_start, indexer_rewards, delegator_rewards, fees_gross, fees_net
FROM lodestar_indexer_daily
WHERE indexer = '0x…' AND day_start >= epoch(current_date - INTERVAL 30 DAY)
ORDER BY day_start;

The work is in graph-allocations-nest#23, kittiwake#140 and #142, and lodestar#229 and #230.

If your numbers and ours disagree, tell us. This time we would like to hear about it first.

lodestarnuthatchindexersrewardsquery-feeshorizoncorrection