I’ve been instrumenting the MEV-Share event stream for the past week
as part of scoping a research proposal on searcher route choice and
hint quality. The first pass surfaced something I didn’t expect: after selector decoding, the stream looks far more strategy relevant than a naive parse suggests.
Posting the full FRP draft here for community feedback before formal submission.
FRP: How Informative Are MEV-Share Hints? Searcher Route Choice, Blind Bidding, and Retained Surplus on Flashbots Surfaces
Author
Vishal Subbu
Summary
54.6% of 6,843 observed MEV-Share hint events reclassified as DEX- or aggregator-related after selector decoding, versus 0.9% under a naive surface parse. The stream is substantially more strategy-relevant than it first appears, and the searcher decision problem that follows is the subject of this proposal.
MEV-Share gives searchers access to selectively revealed orderflow through an event stream and bundle workflow, but that access comes with a routing decision: when should an opportunity be routed through MEV-Share, and when should it remain private? That decision appears to trade off three linked factors: information leakage, refund mechanics, and bid competitiveness. Flashbots’ docs explicitly frame MEV-Share as a searcher surface built around hinted transactions and partial-bundle workflows, where hidden information changes how searchers can reason about opportunities.
Recent operator discussion suggests that refund-related settings can reduce competitiveness and, in some cases, create silent-discard conditions when refund-related transaction cost is not covered. The same discussion suggests that refunds are treated as trusted or permissioned rather than guaranteed, and that the real leverage is on the bid side: knowing true profit floor and not overbidding. At least some searchers appear to rely on rough bribing heuristics rather than explicit dynamic floor models.
This proposal studies searcher execution quality under route choice.
The central question is:
For which opportunity classes should a searcher route through MEV-Share versus private builder paths, and how should bid and refund parameters be set to maximize retained surplus under each route?
Motivation
Existing Flashbots research has explored adjacent mechanism questions around orderflow auctions and contingent fees. What appears less understood is the workflow-level searcher problem inside those mechanisms: how a searcher should actually behave when faced with incomplete information, trusted refund semantics, and route-dependent leakage risk.
The motivation for this work is the gap between:
• mechanism-level discussion of orderflow auctions and redistribution, and
• the day-to-day operational decision a searcher makes when choosing a route and setting a bid.
Initial instrumentation suggests that the MEV-Share stream is far more strategy-relevant than a naive parser would imply. Across 6,843 events:
• 54.6% were classified as aggregator-routed flow,
• 31.1% as ERC-20 transfer flow,
• only 7.5% remained true non-DEX residuals,
• and total DEX / aggregator-related flow rose to 56.5% after reclassification.
This suggests that much of the stream is not merely noisy background flow, but real opportunity-bearing traffic whose usability depends on how much information is revealed and how searchers price route-specific risk.
This also connects directly to Flashbots’ current research agenda. Flashbots has recently argued that MEV is becoming a dominant limit to blockchain scaling because searchers, operating under poor information and private mempool conditions, submit large volumes of blind speculative transactions that waste blockspace. In their latest writing, Flashbots reports spam bots consuming more than 50% of gas while paying less than 10% of fees on some rollups, with Base throughput gains largely absorbed by spam search. If better information quality on Flashbots-native surfaces changes how searchers route and price opportunities, that is not just a searcher optimization question; it is potentially a blockspace-efficiency question as well.
At the same time, preliminary operator discussion suggests:
• refund settings can directly lower competitiveness and may induce silent discard under some conditions,
• not every builder is assumed to support refunds consistently, and refund settlement is viewed as a trusted or permissioned mechanism,
• searchers believe the real leverage is on the bid side rather than in hoping for rebates after the fact,
• and at least some searchers are still operating with rough bid heuristics instead of explicit route-aware floor models.
These observations suggest that searchers may be losing retained surplus not because opportunities are absent, but because route choice and bid-setting are under-modeled.
Research Questions
1. How informative are MEV-Share hints by opportunity class?
2. For which opportunity classes does routing through MEV-Share improve retained surplus for searchers, and for which does private routing dominate?
3. How should searcher bid / max-floor logic change as hint richness and leakage risk vary?
4. When do refund-related settings meaningfully alter competitiveness or induce discard risk?
Hypothesis
Based on initial instrumentation, I expect to find that aggregator-routed flow represents the largest underserved opportunity class on MEV-Share: extractable, but requiring strategy adaptation beyond standard AMM simulation.
I expect ERC-20 transfer flow to be structurally poor for MEV-Share routing regardless of hint richness, because it will often reveal too little strategy-relevant structure while still exposing searchers to route-dependent competition.
I also expect refund friction to matter most for mid-sized opportunities, where refund transaction cost and reduced competitiveness represent a meaningful fraction of bundle value. This aligns with recent operator discussion describing refund settings as a source of both trust risk and silent failure under certain conditions.
Methodology
Passive instrumentation
I will continue instrumenting the MEV-Share event stream and recording:
• raw hinted events,
• revealed fields,
• selector richness,
• address and topic structure,
• and opportunity-class guesses.
Selector decoding and reclassification
I will decode function selectors and use known address and calldata patterns to separate:
• aggregator-routed DEX flow,
• ERC-20 transfer flow,
• direct swap calls,
• NFT / bridge / safe / sequencer / other non-DEX classes,
• and true unknowns.
Comparative analysis
For each opportunity class, I will compare:
• hint richness,
• route-attractiveness proxies,
• likely simulation viability,
• leakage-sensitive vs leakage-tolerant classes,
• and refund-friction-sensitive classes.
Where feasible, I will also compare:
• private route vs MEV-Share route,
• refund-on vs refund-off behavior,
• heuristic bidding vs modeled bidding,
• expected value vs realized retained surplus.
Searcher interviews
Searcher interviews will be used as supporting evidence, not the empirical core. They will help clarify:
• how searchers currently choose between routes,
• how they interpret refund mechanics,
• how they estimate bid floor,
• and which outcomes they currently track or ignore.
Expected Contributions
1. A reusable public dataset and methodology for analyzing MEV-Share hint quality and searcher-side route choice
2. An empirical taxonomy of opportunity classes on MEV-Share
3. A framework for route-aware bid setting under heterogeneous information quality
4. Evidence on how refund friction affects execution quality in practice
5. A clearer link between information quality on Flashbots surfaces and blind-bidding pressure in searcher workflows
Deliverables
• A public Flashbots forum post or paper presenting the findings
• A presentation at a Flashbots or community event
• Public code and methodology for instrumentation and analysis where feasible
Feedback I’d especially value:
1.Does the route-choice framing feel right, or should this be scoped more narrowly as an execution-quality question?
2.Is the current opportunity-class taxonomy missing an important category?
3.What would be the most convincing next empirical test before formal submission?
Appreciate any pushback, especially from people actively using MEV-Share or thinking about searcher-side workflow design.