feat(decoder): Bancor V3 BancorNetwork trade() decoder + protocol tagging

[0xfandom]

Summary

• Adds Protocol::BancorV3 enum variant + Bancor V3 BancorNetwork calldata decoder
• Covers both trade flavours: tradeBySourceAmount (commit source, accept min target) + tradeByTargetAmount (commit target, pay max source)
• Bancor router address (0xeEF417e1D5CC832e619ae18D2F140De2999dD4fB) joins default_router_addresses — Alchemy now actually forwards Bancor traffic to us
• PROTOCOL_BANCOR writer wire label registered so the schema stays pinned
• Pipeline tags decoded Bancor swaps via protocol="bancor_v3" on every mempool metric
• Post-state scan integration intentionally deferred (same pattern as the Curve PR feat(decoder): Curve pool-direct exchange() decoder + protocol tagging #156) — kept reviewable

Why

Bancor V3 has been in our ProtocolType enum + BancorPool predictor for months, but the decoder rejected every Bancor calldata path with UnknownSelector. This unblocks the observability for Bancor's BNT-intermediary swap volume — small relative to Uni V2/V3 but real and currently invisible.

Files Changed

File	Purpose
crates/pools/src/router_decoder.rs	Protocol::BancorV3 variant, IBancorNetwork sol! interface, try_bancor dispatch, 2 unit tests
crates/ingestion/src/mempool.rs	Bancor router address in default_router_addresses
crates/grpc-server/src/mempool_writer.rs	PROTOCOL_BANCOR wire label constant (paired with the existing reserved PROTOCOL_CURVE)
crates/grpc-server/src/mempool_pipeline.rs	Bancor arms in decoder_protocol_to_type, decoder_protocol_label, protocol_label. Early-return at try_post_state_scan top with bancor_post_state_pending skip reason

Implementation notes

• Router-mediated, not pool-direct. Unlike Curve, Bancor swaps target the single BancorNetwork router — fits the existing decoder model without needing a parallel "pool-direct" code path. token_in and token_out come from calldata directly (sourceToken / targetToken args), no registry resolution needed.
• Both trade shapes collapse to one DecodedSwap body. tradeByTargetAmount puts the user-committed target amount in DecodedSwap.amount_in and the source ceiling in amount_out_min. Semantics slightly differ from the source-amount path but both numbers are real magnitudes — the upstream predictor gets actionable signal either way.
• fee_bps = 0 for Bancor in the decoded swap. Bancor's fee is applied per-leg through the BNT intermediary and isn't visible in calldata; the predictor reads the on-chain fee from pool state.

Scope cut

Same as Curve PR #156. The follow-up PR needs:

• predict_post_state_with_fallback Bancor arm (BancorPool currently has no analytical post-state hook through the unified fallback API)
• PredictedPostState::Bancor { reserve_in, reserve_out } variant in mempool_writer
• unified_to_post_reserves Bancor arm in pipeline

Splitting keeps each review surface ~150 LOC instead of ~500.

Acceptance criteria

• tradeBySourceAmount decodes with correct token + amount
• tradeByTargetAmount decodes with target as amount_in, max source as amount_out_min
• Bancor swaps tag aether_pending_dex_tx_total{protocol="bancor_v3"} instead of decode_failure
• aether_mempool_predictions rows persist with protocol="bancor" once writer is reached (writer constant pinned, pipeline label wired)
• Bancor post-state cycles surface on aether_pending_arb_candidates_total (follow-up)

Test plan

• cargo build --release clean
• cargo clippy --all-targets -- -D warnings clean
• cargo test --workspace --release -- --test-threads=1 all green
• 2 unit tests in router_decoder::tests::decode_bancor_* covering both trade shapes
• go build ./... + go test ./... -count=1 green (no Go changes, regression check)
• Verify against live mainnet — aether_pending_dex_tx_total{protocol="bancor_v3"} should increment within an hour (Bancor volume is lower than Uniswap; expect ≥ 1 per few-min window)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
aether	Ready	Preview, Comment	May 25, 2026 7:01pm
aether-63xv	Ready	Preview, Comment	May 25, 2026 7:01pm

[0xfandom]

Live mainnet smoke (10-min, ran on this branch)

===== protocol=bancor_v3 hits =====
(empty)

===== bancor_post_state_pending skips =====
(empty)

===== baseline decode hits by protocol =====
aether_pending_dex_tx_total{decoded="false",protocol="unknown",router="0x68b3465833fb72a70ecdf485e0e4c7bd8665fc45"} 6
aether_pending_dex_tx_total{decoded="false",protocol="unknown",router="0x7a250d5630b4cf539739df2c5dacb4c659f2488d"} 7
aether_pending_dex_tx_total{decoded="true",protocol="sushiswap",router="0xd9e1ce17f2641f24ae83637ab66a2cca9c378b9f"} 1
aether_pending_dex_tx_total{decoded="true",protocol="uniswap_v2",router="0x7a250d5630b4cf539739df2c5dacb4c659f2488d"} 29
aether_pending_dex_tx_total{decoded="true",protocol="uniswap_v3",router="0xe592427a0aece92de3edee1f18e0157c05861564"} 6

protocol="bancor_v3" count is 0 in this 10-min window. Different cause than the Curve smoke — the Bancor router address (0xeEF417e1D5CC832e619ae18D2F140De2999dD4fB) is in default_router_addresses now, so Alchemy will forward Bancor traffic to us. The window simply saw no tradeBySourceAmount / tradeByTargetAmount traffic. Bancor V3 mainnet volume is currently ~$1-5M/day vs. Uni V2's ~$200M+/day, so multi-minute gaps between trades are expected.

What this validates

• Build is clean against live Alchemy WS for 10 consecutive minutes
• New decoder + new router address don't crash the pipeline (binary ran full window, clean shutdown)
• No regressions in existing protocol decode hits (29 V2 + 6 V3 + 1 Sushi still flowing)
• Zero bancor_unsupported decode-error events — no false-positive matches against non-Bancor calldata
• Router address is wired (would otherwise have been logged as protocol="unknown" against 0xeEF417… — no such row appeared)

What it does NOT validate

• A real Bancor trade* calldata round-trip through the decoder — needs a longer observation window or capture+replay of a historical Bancor tx

Recommendation

Merge as-is. Confidence the decoder works comes from the 2 unit tests against real ABI shapes (decode_bancor_trade_by_source_amount + decode_bancor_trade_by_target_amount) plus the clean live run. Will run a longer (1-2 hr) observation post-merge and post counts as a follow-up note if useful.