Keep raw and decoded stages separate
Raw shreds are ledger fragments. Decoded transactions are a later representation produced after enough fragments are available and parsed. Comparing a raw-packet timestamp from one provider with a decoded-transaction timestamp from another measures two different stages. Use identical representation and capture points for a delivery comparison.
Match packet identity
key = parse_protocol_shred_identity(packet)
arrival[source].setdefault(key, monotonic_ns())
# Compare only identical protocol identities.
# Log duplicates and packets unique to each source separately.The identity must include the fields needed to distinguish shred variants, such as slot, index, type and protocol version context. Do not use packet length or slot alone. Use a validated Solana shred parser rather than inventing byte offsets.
Control the receiver
- Receive both feeds on the same machine and use a consistent userspace or kernel timestamp boundary.
- Record receive ports, socket buffers, NIC settings, CPU load and packet-drop counters.
- Persist first arrival per source and identity, keeping duplicates as a separate metric.
- Report matched coverage and delta distributions together; faster arrival among a small surviving subset is incomplete evidence.
Keep the conclusion reproducible
A local result describes delivery to this host during the measured window. Routing changes, receiver loss and provider coverage can affect it. Clock synchronization matters when comparing hosts; a single-host monotonic clock avoids one class of cross-machine error but does not isolate every network effect. Publish configuration and sampling rules with the result so someone else can reproduce the experiment.