Purchased landing is not predictive alpha
Status: preliminary source review
No transaction-level sample has been collected. This is a research design, not a measured trading edge.
What the source review supports
Jito's low-latency documentation describes bundle submission, atomic sequential execution and a tip-based auction. The browser summary reports that competing bundles are ranked using tip efficiency relative to requested compute resources. These are external documentation claims, not AGENCY measurements; implementation details still need direct verification.
Atomic execution describes how a bundle executes if included. It does not guarantee landing, profitability or a particular position relative to all other transactions. The retrieved summary used language suggesting deterministic inclusion; that conclusion is not supported here.
Quant's hypothesis
A strategy can appear predictive when its advantage is actually better submission latency or paid access to execution. Comparing only landed trades would also discard failed opportunities and bias performance upward.
Measurement specification
Record signal timestamp, submission timestamp, route, quoted output, minimum acceptable output, requested compute limit, compute price, transaction signature, landing slot, execution status and realized output. Track base fees, priority fees and Jito tips separately, using actual charges rather than assuming every failed submission paid every cost. Attribute bundle-level costs without double counting.
Compare routes under matched signals and market conditions. Include failed, expired and unlanded attempts. Report net outcome after execution charges and slippage, with explicit treatment of inventory and unmatched opportunities. Routing can itself be selected based on opportunity quality, so route comparisons are not automatically causal.
Open questions
- What are the current Solana priority calculation and scheduler rules? Primary Solana documentation is the next source.
- How do account conflicts constrain bundle competition?
- How should reverted or uncled-block cases be represented?
- Can a public historical dataset identify unlanded attempts? Landed on-chain transactions alone cannot reconstruct that denominator.
Bottom line
Execution quality and signal quality are different variables. Neither is measured in this notebook yet.