What you can check
Each figure is generated from a tracked file and links to the page that says how it was made and how to run it again.- Pubnet ledgers 2 to 65,505,999, every one, checked. 65,505,998 ledgers and 10,869,626,939 transactions were read from the public data lake, decoded, re-encoded byte for byte and hash-checked: 0 ledgers refused and 0 mismatches over 65,505,997 header hash-chain links. The scan computes shapes and these checks; no pipeline or SDK was replayed over all of it. Mainnet scan
- Entries a real stellar-core emitted. 446 corpus entries, each tagged with its CAP and protocol; 138 of them were produced by a local, offline stellar-core under both the testnet and the mainnet passphrase, 143 were captured from network history and 53 come from stellar-core’s own goldens. The 111 hand-made entries are marked synthetic. How to read a corpus entry
- Public incidents reproduced before and after their fixes. Three wallet-backend incidents and three
js-stellar-sdk incidents fail on their pre-fix versions and pass on the fixed ones (
make verify-replay,make verify-conformance). 6 of the 18 public incidents on that page are reproduced this way; the others are run, checked or held as corpus input. Known incidents - One corpus across the SDKs. 21 columns (releases of the Go, JavaScript, Python and Rust SDKs, and stellar-core) in one conformance matrix, so a defect found in one SDK is checked in the others. SDK conformance matrix
- Findings filed upstream. 92 findings with a public record, 48 of them filed upstream. Findings
Results
Mainnet scan
What was scanned and checked, the shape census, prevalence per entry.
SDK conformance matrix
Agrees, diverges and not run for every SDK column.
Replay results
The corpus through wallet-backend and stellar-etl, before and after fixes.
Known incidents
Each public incident: reproduced, run, checked or held as input.
Findings
Every finding with severity, component, entries and upstream issue.
CAP coverage
Each CAP’s entries, their source ledgers and their mainnet prevalence.
Start by role
I maintain a Stellar SDK or XDR library
Put every corpus entry through your decoder in your own CI.
I write or run an indexer
Replay the corpus through your indexer with an adapter, then on every pull request.
My team writes Go ingestion code
Run
xdrlint locally and in CI.I ship an SDK for a new protocol
Test the next protocol’s shapes before the SDK release.
I add shapes for a new CAP
Write corpus entries and check them against stellar-core.
I want to try it first
Read the corpus offline from a fresh clone, then run the demo.
What the kit is
Every part reads the same corpus manifest (Manifest and index format). The diagram of how
they connect is on Architecture overview.
Conventions on this site
- The command-line tools are
xdrcheck,xdrcheck-corpus,xdrlint,xdrcheck-judgeandxdrcheck-tagger(Command-line reference). OWNER/REPOstands for the XDRCheck repository on GitHub, in URLs,uses:lines and Go module paths alike.- Commands run from the root of a checkout of the repository unless a step says otherwise. After
make build, the tools are in./bin; add them to yourPATHwithexport PATH="$PWD/bin:$PATH". - A corpus release is named by the protocol it covers:
30.0.0covers Protocol 30. See Release and channels. XDRCheck is in active development toward its first release; until it is published, run everything from a checkout of the repository, and read<version>and<commit>in commands as the release or commit you pin. - The figures on this site, the README’s results and the dashboard (
apps/dashboard) are generated from the same tracked files undercorpus/andevidence/.