Multi-leg transaction undercount | SmartFlow Observatory

What it measures

Publication limit: The source note leaves its pre-publication red team open and states that the extrapolated dollar figure must not be cited. This page documents the method only; the result is not cleared for publication.

The frequency and value of additional qualifying USDC Transfer logs in transactions already present in the database, using per-transaction sampling, plus whether the missed legs map to payout destinations in the endpoint catalog. It does not measure transactions omitted entirely by ingestion.

How it is produced

The first script samples recorded transactions and measures multi-log incidence. The second quantifies missing legs by month. The third joins missed-leg recipients to catalog payout destinations. The database schema uses tx_hash as the primary key with INSERT OR IGNORE, so only one qualifying leg can be stored for a transaction hash.

Source scripts

rpc_sample.py
h6_quant.py
h6_join_catalog.py

The RPC sample requires a User-Agent header.

Limits and assumptions

Sampling starts from transactions already present in the database, so it estimates undercount relative to stored transactions.
The monthly sample is equally weighted at 400 transactions per month rather than weighted by ledger volume. A ledger-wide dollar extrapolation requires volume weighting and was not performed.
A catalog hit means the payout destination of a missed leg appears among the 575 catalog addresses. A payment to an uncataloged destination remains a miss, just as it does for stored rows.
The catalog join does not evaluate facilitator-mediated classification through tx_sender. A separate batch-settlement test is therefore required before concluding that the attributed clean core is unaffected.
RPC data reflects the chain's current state. Reorganizations at the sampled Base depth are assumed negligible.

How to refute this

Check whether the missed legs cluster on batch-settlement senders rather than on end recipients; if they do, the "stored core is unaffected" conclusion does not hold.
Weight the sample by monthly ledger volume before citing any dollar-level undercount; the unweighted average understates it if the highest legs-per-transaction month dominates volume.
Investigate the highest-volume recipient of missed legs. If it is an uncataloged x402 merchant with a live 402 endpoint, the reported catalog-hit rate is an artifact of catalog incompleteness.
Re-run both scripts on a fresh window. Incidence and catalog hit rate should reproduce within sampling bounds.

Last validated

2026-07-24