Skip to main content
For indexer operators and authors: what wallet-backend, stellar-etl and the kit’s example indexer stored for each corpus entry, which entries the pinned current versions get wrong, and how three wallet-backend defects behave before and after their fixes. Read it to see which shapes your deployment of these indexers mishandles, or as a model of the result your own indexer gets through an adapter. xdrcheck replay serves the entries’ ledgers to an unmodified indexer through the input it already reads, lets it ingest, then has an adapter read back what it stored. The harness compares those rows with the facts a reference extractor took from the ledger’s meta, never from the indexer. Inputs: rpc, the mock getLedgers; sep54-dir, the Galexie layout as a directory; sep54-s3, the Galexie layout from an object store.

How to read a row

Each row is one indexer version on one input; each class column counts entries. Scope corpus means the whole corpus ran; known-divergences means only the entries that declare a known divergence for that indexer ran. The full rules (eleven conditions, first match wins) are on Replay harness.

Every indexer version and input

The baseline is the kit’s complete example indexer (go/adapters/baseline); its refusals are its documented behaviour.

Three wallet-backend defects, before and after their fixes

The wallet-backend adapter (go/adapters/walletbackend) runs each before-and-after pair that pins.toml lists under [pipelines.wallet_backend]: the commit before the fix and the commit that fixed it, on both inputs (make verify-replay OFFLINE=1). The incident numbers refer to Known incidents. The transcripts are in evidence/wallet-backend-replay/ and evidence/wallet-backend-731/.

stellar-etl at the pinned current version: entries it gets wrong

Only the entries that declare a known divergence for stellar-etl ran (scope known-divergences), so this list is not a whole-corpus result.

wallet-backend at the pinned current version: entries it gets wrong

Each entry links to its row on CAP coverage; the findings table describes each defect. To replay one entry yourself: xdrcheck replay --adapter ADAPTER --target-version VERSION --entries ENTRY (commands).

Limits

  • These are results on corpus entries, not on all of mainnet. How often each shape occurs on pubnet is on Mainnet scan.
  • Each indexer runs at the commit pins.toml names; a later commit has no row until it is pinned.
  • Horizon has no row: its rig runs lake ledgers, not corpus entries (Horizon rig).
The data is evidence/results/replay-summary.json (make results-replay); the design is on Replay.