Order scan
make order-scan: 20 fresh runs per recipe lane record which ledgers come out in more than one byte order, and prove the content never varies.
make order-scan # recipes/order-scan.sh classic 20, then soroban 20 at P30; about 15 minutesstellar-core iterates a hash map with a per-process random salt, so the changes within an operation can come out in a different order on every run. The order scan measures this. It runs each recipe lane 20 times on fresh cores, records every byte ordering seen per ledger, and fails if any ledger's content differs once order is normalised.
The committed results
Results are in recipes/classic/order-scan/p30-mainnet-base.json and recipes/soroban/order-scan/p30-mainnet-base.json:
| Lane | Runs | Ledgers that vary in order | Content that varies |
|---|---|---|---|
| Classic | 20 | 18 of 80 recipe ledgers; 12, 78, 80 differ in every run | 0 |
| Soroban | 20 | 13 of 80 recipe ledgers | 0 |
The protocol-upgrade ledger (2) varies in both lanes as well; it is never an entry.
Corpus entries whose order varies pin an order-normalised hash, and list every raw ordering seen (change_order in the schema).
For indexers
Generate
make generate: 20 fresh standalone stellar-core v30.0.0 runs, in pairs, over both lanes, both passphrases, P30 and P28 and three flag profiles. Every pair must agree, and every shipped generated variant must come out again.
Conformance
make conformance: the same vectors through JavaScript and Go SDK builds before and after a fix, plus rs-stellar-xdr. Known bugs fail on the build before their fix and pass after.