usdt_transfers

One map module, map_transfers, emitting every USDT Transfer event on the streamed network as erc20.v1.Transfers (from, to, contract, amount, trx_hash and log_index).

amount is the raw uint256 as a decimal string — apply the token's decimals yourself (6 everywhere except BNB Chain, which uses 18).

Networks

The contract address is a module param and the deployment block is the module's initialBlock; both are set per network in substreams.yaml.

--network	Token	Contract	initialBlock
mainnet	USDT	0xdac17f958d2ee523a2206206994597c13d831ec7	4634748
bsc	USDT (18 decimals)	0x55d398326f99059ff775485246999027b3197955	4208826
base	USDT	0xfde4c96c8593536e31f229ea8f37b2ada2699bb2	12044917
polygon	USDT0	0xc2132d05d31c914a87c6611c10748aeb04b58e8f	4196335
arbitrum	USD₮0	0xfd086bc7cd5c481dcc9c85ebe478a1c0b69fcbb9	228105
optimism	Bridged USDT	0x94b008aa00579c1307b0ef2c499ad98a8ce58e58	0
avalanche	USDt	0x9702230a8ea53601f5cd2dc00fdbc13d4df4a8c7	4663628
unichain	USD₮0	0x9151434b16b9763660705744891fa906f660ecc5	11236573
tron-evm-mainnet	USDT (TRC-20)	0xa614f803b6fd780986a42c78ec9c7f77e6ded13c	8418292

Tron carries the most USDT transfers of any network. Use tron-evm-mainnet, which serves Tron in the Ethereum block format — the plain tron endpoint serves sf.tron.type.v1.Block, which this module cannot read, and streaming it produces no output and no error.

Tron addresses come out in EVM form. To get the base58 address Tron tooling expects, base58check-encode 0x41 followed by the 20 bytes: 0xa614f803… is TR7NHqjeKQxGTCi8q8ZY4pL8otSzgjLj6t.

Optimism's USDT predates the chain's regenesis, so it exists at block 0. Arbitrum's predates Nitro; Firehose serves those blocks.

Avalanche's USDT.e (the bridged variant, 0xc7198437980c041c805a1edcba50c1ce5db95118) is a separate token — pass it with -p if you want it instead.

Benchmark

cargo bench

Times the mapping loop against eight real mainnet blocks committed under testdata/blocks (1909 transactions, 246 USDT transfers). No endpoint, no network. The bench asserts every variant emits identical transfers before timing them, so a change that alters the output fails rather than posting a fast number.

Three groups, on an M-series laptop:

group	v0.1.0	v0.2.0
map_only — the mapping loop	466 µs	70 µs
decode_only — prost block decode, unchanged by either	14.44 ms	14.44 ms
decode_map_encode — the whole handler	14.56 ms	14.36 ms

The decode is 97% of the handler, so a mapping-loop win lands as roughly 1.5% end to end. Judge a change by decode_map_encode; use map_only to see whether the mapping change itself did anything.

map_only also carries two diagnostic variants: lazy_hash_only (142 µs) isolates the lazy transaction-hash encoding from the hand-rolled hex writer, and v0.2.0_minus_amount (48 µs) sizes the uint256 to decimal conversion, which is a third of what v0.2.0 still spends.

criterion stores a baseline, so a second cargo bench reports the delta against the previous run. To compare against a specific revision, run it on that revision first:

git stash && cargo bench && git stash pop && cargo bench

Fixtures are gzipped sf.ethereum.type.v2.Block messages. testdata/dumper is the package that captured them; its README covers pulling a different range or chain.

Run

cargo build --target wasm32-unknown-unknown --release
substreams run substreams.yaml map_transfers --network tron-evm-mainnet -s 60000000 -t +10 -o jsonl

Any other network, or a different ERC-20 entirely, works without a rebuild:

substreams run substreams.yaml map_transfers --network mainnet \
  -p map_transfers=0xa0b86991c6218b36c1d19d4a2e9eb0ce3606eb48 -s 25800000 -t +10 -o jsonl