← Back

MEV Bid Timing Game: When Do Builders Submit?

Key findings

Bid values grow dramatically within each slot

The median bid value at -2 seconds (before slot start) is just 0.0002 ETH. By slot start, it's 0.0076 ETH. By +4 seconds, it peaks at 0.0151 ETH.

That's a 70x increase from early bids to late bids within the same slot window.

Most bids arrive after the slot begins

Of the 8.7 million bids in the -3s to +4s window, 61% (5.3 million) came after slot start. The bid count peaks at +0.9 seconds with 131,427 bids.

This makes sense: more time = more transactions observed = more MEV opportunities = higher bid values.

The timing tradeoff

Builders face a classic explore/exploit dilemma. Wait longer to see more MEV opportunities, but risk missing the window when the proposer selects a header. This creates the characteristic ramp-up pattern visible in the chart.

The chart

Blue area: bid count over time. Green line: median bid value. The bid count peaks around +1 second, but median value keeps climbing until +4 seconds.

Data and methodology

Source: mainnet.fct_mev_bid_highest_value_by_builder_chunked_50ms (xatu-cbt)

Date range: 24 hours ending 2026-02-04 06:00 UTC

Window: -3 seconds to +4 seconds relative to slot start

Query

SELECT
  chunk_slot_start_diff AS timing_ms,
  COUNT() AS bid_count,
  AVG(value) / 1e18 AS avg_value_eth,
  quantile(0.5)(value) / 1e18 AS median_value_eth,
  quantile(0.95)(value) / 1e18 AS p95_value_eth,
  MAX(value) / 1e18 AS max_value_eth
FROM mainnet.fct_mev_bid_highest_value_by_builder_chunked_50ms FINAL
WHERE slot_start_date_time >= now() - INTERVAL 24 HOUR
GROUP BY timing_ms
ORDER BY timing_ms

Data points

Why this matters

The timing game isn't just a builder phenomenon. Validators can play timing games too - some deliberately delay header selection to extract better bids. The whole PBS (proposer-builder separation) ecosystem has timing optimization baked in.

Understanding when bids arrive helps explain why mev-boost has millisecond-level tuning parameters and why the timing of header requests matters so much.

Limitations

• This data represents bids observed by ethpandaops relays, not all relays
• The 24-hour window may not capture unusual events
• Bid count includes all builders; individual builder strategies may differ
• Timing resolution is 50ms buckets, not exact timestamps