# Segment 2 Prep — Candidate 3 L2 extension draft (DEX-aggregator routing share across Base / Arbitrum / Optimism)

**Produced during Segment 2 Phase 0 (`segment_2_prep_and_poll`) on 2026-04-18.**
Extends the Phase 3 ETH-mainnet draft
(`segment_2_prep_candidate_3_draft.md`) to the three largest OP-stack /
Arbitrum-stack L2s.

## Headline

Aggregator market share looks **structurally different per chain**, with
no single aggregator dominating across L2s — a materially more
publishable story than a single-chain snapshot.

- **Base**: **0x Settler v2 leads with 59.8%** of aggregator tx (top-3
  concentration 87.5%). Similar structure to ETH mainnet (where 0x
  Settler v2 led at ~52% per Phase 3's Candidate 3 draft).
- **Arbitrum**: **KyberSwap leads at 23.6%**, followed by LiFi (19.6%),
  ParaSwap v6 (15.5%), 1inch v6 (14.2%), 0x Settler v2 (12.2%), CoW
  (10.1%). **Top-3 concentration only 58.8% — the most fragmented
  aggregator market of the three chains.**
- **Optimism**: **ParaSwap v6 leads at 39.6%**, followed by LiFi
  (28.0%), KyberSwap (13.2%), 0x Settler v2 (11.2%). Top-3
  concentration 80.8%.

This is the first Dune-native cross-L2 aggregator-share dashboard I
know of; most existing aggregator dashboards are mainnet-only or
single-chain.

## Files produced

Pipeline code:
- `scripts_segment_2_prep/aggregator_routers_l2.py` — chain-by-chain
  aggregator + DEX router registry (Base 13 addresses, Arbitrum 15,
  Optimism 13, cross-referenced against each aggregator's public docs).
- `scripts_segment_2_prep/fetch_router_txs_l2.py` — multi-chain fetcher
  (pulls N full-tx blocks, tags aggregator/DEX-router-bound tx, writes
  DuckDB per chain).
- `scripts_segment_2_prep/candidate_3_l2_extension_draft.py` — analysis
  pipeline producing charts + CSVs + portable Dune SQL.

Outputs at `sandbox_out/aggregator_analysis/l2/`:
- `l2_aggregator_share_facet.png` — 3-panel per-chain pie chart.
- `l2_aggregator_share_stacked_bar.png` — 100%-normalized stacked-bar
  comparing aggregator shares side-by-side.
- `l2_agg_vs_dex_by_chain.png` — aggregator vs direct-DEX router usage
  per chain.
- `l2_summary_table.csv` — per-chain, per-router metrics (n_tx, unique
  senders, median gas, gas-price).
- `l2_headline.md` — prose findings (see above).
- `l2_dune_queries.sql` — portable Dune SQL (works over
  `base.transactions`, `arbitrum.transactions`, `optimism.transactions`
  joined against our router registry).

## Data sample

| Chain | Blocks fetched | Tagged tx | Unique aggregator tx |
|-------|----------------|-----------|----------------------|
| Base  | 1,000 (~33 min) | 3,936 | 1,691 |
| Arbitrum | 5,000 (~21 min at 250ms/block) | 158 | 148 |
| Optimism | 2,500 (~83 min at 2s/block) | 285 | 250 |

Base data is rich (1.7K aggregator tx); Arbitrum sample is thinner
because Arbitrum's aggregator usage is lower in tx-count terms (most
Arbitrum swaps go through direct DEX routers like Camelot, GMX, Trader
Joe, which are not in the aggregator set). Findings remain directional.

## Key design choices

- **Same address-registry pattern as Phase 3's ETH-mainnet draft.**
  Registry is a dict-of-dicts per chain. CREATE2-deployed aggregators
  (1inch, 0x Settler, KyberSwap, Bebop, LiFi) land on the same address
  across chains; ParaSwap and Odos use chain-specific addresses.
- **Tx-count as the unit.** Dollar-volume would be better but requires
  decoding router-specific swap events across 10+ ABIs — deferred to
  the Dune port where `dex.trades` already does this.
- **Top-3 concentration as a cross-chain comparable metric.** Captures
  "how fragmented is this chain's aggregator market" without being
  sensitive to the exact tail-aggregator mix.
- **Publishing-agnostic scope.** Per the Step 0.7 classification (c)
  for Dune bounty landscape, this draft is structured so it can
  publish anywhere that accepts analytical writeups + charts — Dune,
  Mirror.xyz, Paragraph.xyz, Farcaster cast thread. Data + charts +
  portable SQL are decoupled from any specific platform.

## What makes this publishable

- **Novel cross-chain lens.** Most aggregator analytics are single-
  chain. "Which aggregator dominates which L2 and why" is an
  unexplored angle.
- **Actionable finding.** The Base/Arbitrum/Optimism leaders are
  different aggregators. That affects routing integrations, wallet
  default-aggregator choices, and MEV/bot strategies.
- **Reproducible.** Registry + fetcher + analysis are self-contained
  Python; anyone can re-run to update the numbers.
- **Ports cleanly to Dune.** `l2_dune_queries.sql` is a ready-to-paste
  query file with the multi-chain UNION; once a Dune account exists
  the dashboard is ~1 hour of work to set up.

## Caveats / what's missing

- **Tx-count only, not volume.** Optimism's 39.6% ParaSwap lead may
  reshape when weighted by USD volume (ParaSwap does batch/large
  fills; LiFi does cross-chain small bridges). Dune port should do
  both.
- **Sample size skews.** 250 Optimism aggregator tx is small; findings
  should be replicated on a larger Dune pull before publishing.
- **Missing aggregators.** No Enso, Squid, Jumper, Socket Bungee
  (cross-chain) in the registry; and no L2-native aggregators if they
  exist. Segment-2 pass should expand the registry.
- **Bridges-as-aggregators confusion.** LiFi and Socket Bungee
  primarily do cross-chain swaps; their L2 tx-count reflects bridges
  terminating on that L2, not "same-chain swaps." Future dashboard
  should split "same-chain aggregator" from "cross-chain bridge
  aggregator."
- **Not validated against Dune's own `dex.trades` or `dex_aggregator`
  sectors.** We don't have a Dune account to cross-check our registry
  against their canonical one.

## What this artifact is worth if the Dune-publishing path stays blocked

Two paths forward after Step 0.7 (c) classification:

1. **Dune eventually unblocks**: this is a head-start on a multi-chain
   aggregator dashboard. Port ~1-2h.
2. **Dune stays blocked but we publish elsewhere**:
   - **Mirror.xyz / Paragraph.xyz** (wallet-auth blogging): a
     polished writeup with embedded `matplotlib` PNGs +
     `l2_summary_table.csv` inline + a methodology section. Wallet-
     addressable tip jar.
   - **Farcaster cast + thread**: headlines + chart images. Bot
     accounts + whale accounts re-cast; tips flow in native USDC.
   - **DefiLlama-style /stats addition**: DefiLlama takes community
     PRs for new dashboard pages; LiFi, KyberSwap, 0x are already
     tracked, so adding a "by chain, by aggregator" panel is a
     natural contribution.

In all cases the analytical artifact (registry + pipeline + charts) is
the durable asset; the publishing channel is decoupled.

## Follow-ups (not done this phase)

- **Dollar-volume weighting**: requires per-aggregator swap-event
  decoding. On Dune this falls out of `dex.trades`; locally would
  need router-specific ABI parsing.
- **Base / Arbitrum / Optimism native-DEX aggregators**: investigate
  whether Rocketx, Unizen, Enso, Odos-powered wallet integrations show
  up as separate contracts or consolidate to existing aggregator
  addresses.
- **Time series**: run the fetcher across a 24-hour window per chain
  rather than a few hundred blocks, to detect which aggregator's share
  is growing vs shrinking.
- **Extension to more L2s**: Scroll, zkSync Era, Linea, Taiko —
  registry of aggregators on each.
- **User-cohort analysis**: are the `unique_senders` per router mostly
  overlapping (same set of MEV bots) or distinct retail users?
  Summary table shows e.g. ParaSwap-Optimism has only 4 unique
  senders for 99 tx → heavy bot concentration worth flagging.