Middleman v0.1.0

Who stands between
your quote and your fill.

Order the prints. Join on the maker. Name the middleman, per pool.

Build with CMC · API Hackathon Built on the CoinMarketCap API Markets and Trading Tools MIT · keyless

MOTO/WETH on Ethereum · #1 Uniswap v2 pair by 24 h transactions, CoinMarketCap's own ranking · 2026-09-18 22:36:19 UTC

66.2%

of the pool's volume in the last 800 prints

was 3 wallets buying back what they just sold, inside one transaction — 29 times in 3.6 hours. The pair is #1 because of them.

The problem

Three suspects.
No witness.

Rina routes a mid-cap through the pool her wallet picked and is filled under the quote. She spends the evening convinced a bot ate her.

Every tool shows her volume and 24 h transactions — exactly the two numbers a wallet trading with itself inflates. So she picks the busiest pool and sets a loose cap "to be safe".

The solution

One table names it.

The last 800 real prints of a token from CoinMarketCap's keyless feed, ordered by block and log index, joined on the maker address. One row per pool. One line under it.

round-tripthe same wallet's next print in the block is the other side, size within 5 % — removed and counted

sandwicha round-trip whose legs enclose someone else's print, and the wallet came out ahead — named

organiceverything else — what a fill really pays, p50 / p90 per pool

routethe pool with the lowest organic p90; that p90, rounded up, is the cap

middleman.edycu.dev

The page: 66.2 % headline in orange, the three-pool table with the round-trip row in orange, and the green route line — route via Uniswap v2 / WETH, cap slippage at 0.65 %

How it works · six stdlib modules, no server, no model

Order. Group. Join on the maker.

block 26006339 · Uniswap v2 / WETH · rows verbatim from the API · docs/proof/moto.json

Two rows. One division.

lgid   ma             side  a0 (MOTO)        a1 (WETH)   a1/a0
202    0x9c97c0a6a0…  buy    561,202.9606     0.699926    1.2472e-06
209    0xc9160fdab1…  sell   842,991.8538     1.043025    1.2373e-06  ◀ leg
218    0xc9160fdab1…  buy    829,640.6520     1.032594    1.2446e-06  ◀ leg
225    0xdd64c9d73e…  sell   774,494.6589     0.958588    1.2377e-06

same ma · same block 26006339 · same tx
|Δa0| / a0 = |829,640 − 842,991| / 842,991 = 0.0158 ≤ 0.05 → round-trip

That is the whole definition. Sell 842,991 MOTO, buy 829,640 back, 1.6 % apart, one transaction. Divide a1 by a0 and you have the price each leg paid.

This wallet and two others did it 29 times in 3.6 h — $127,254 of the pool's $192,275. Take the legs out and an organic fill pays 15.5 bps, not the 55.3 the raw tape implies.

A trader reading the raw tape sizes a cap three and a half times too loose.

Live · the same engine ported to the browser, parity-tested, behind a keyless proxy

Paste any token. Read the rows.

middleman.edycu.dev · UNI, live

UNI, live: 19 pools ranked by organic quote-to-fill, the green route line — route via Uniswap v3 / WETH, cap slippage at 0.20 % — and the raw-rows panel for the block that holds the one sandwich

119 pools of one token, ranked by organic p90

2the decision: venue + cap, with the rule stated

3the block that holds the sandwich — legs orange, victim red

middleman.edycu.dev · SHIB, live

The paste box after a live SHIB fetch: 800 prints in 8 pages, 6.6 s, 0 credits; the headline reads 1.7× more per fill in Uniswap v2 / WETH than in ShibaSwap / WETH

4800 prints · 8 pages · 0 credits, through /api/swaps

middleman.edycu.dev/evidence

/evidence: 108 requests to CoinMarketCap's keyless /public-api with HTTP status, UTC, milliseconds and the sha256 of each body; the engine's rules verbatim; the re-derive commands

5every call: status · UTC · sha256 of the body

Ten tokens · three chains · the same rule, whatever it said · docs/proof/census.json

Ten tokens. Whatever they said.

prints scanned

8,000

10 tokens on Ethereum, BSC and Solana, 445 pools; every receipt committed and re-derivable from its tape offline.

round-trips

45

by 14 wallets. The widest gap between one token's pools: LINK, 22.9× — 0.6 bps median on Uniswap v4 (Ethereum) / ETH against 14.2 on Uniswap v3 (Ethereum) / USDC.

sandwiches

2

both the same wallet, front-running a seller on UNI and on LINK — for $0.95 and $2.19. A rate of 0.025 %.

RETRACTED · 2026-09-18

This project was decided as the sandwich rate per pool. The day-1 spike found zero same-block sandwiches in 1,200 prints; the census found two in 8,000. So the headline changed before the build, not the caveat after — same feed, same ordering, same join; the sandwich is one named case of a middleman, and the number the page leads with is the one the join actually found. On 2026-09-20 the same command found 17 sandwiches in one window on a different pair — small on average is not small everywhere.

CoinMarketCap DEX API · keyless /public-api surface · every call on /evidence with its hash

Six keyless endpoints. One engine.

endpoint	what the product takes from it
`/v1/dex/tokens/transactions`	the engine — ma h lgid tp a0 a1 tx en t0a t1a per swap, lastId cursor
`/v1/dex/token/pools`	pool address, venue, liqUsd — labels each (en, t0a, t1a) group
`/v1/dex/security/detail`	buyTax / sellTax — a transfer tax is never scored as a middleman
`/v4/dex/pairs/quotes/latest`	24 h buys + sells — 15.7 % of the pair's 4,899 transactions is what 800 prints covered
`/v4/dex/spot-pairs/latest`	the hero rule — #1 pair by no_of_transactions_24h, per chain
`/v1/dex/platform/list`	explorer URL templates for the transaction links

0

credits, every receipt

No key exists to charge. It survives the Oct 1 Basic reversion at zero credits.

5

systems to replace it

A multi-chain swap indexer with maker attribution, a pool registry, a fee-on-transfer simulator, a pair-count aggregator, an explorer map.

8

findings in FEEDBACK.md

From the missing CORS header to q rounded to two figures on Uniswap v4 rows — dated, with evidence.

Four real runs, 145 tests, one command · DEMO.md

Run it. The number moves.

2026-09-18 22:36:19 UTC · scripts/seed.py, the capture behind the page · no key · DEMO.md receipt 1

$ python3 scripts/middleman.py   hero rule → MOTO/WETH, the #1 Ethereum Uniswap v2 pair by 24h transactions
MOTO · ethereum · 800 prints · 3.6 h · blocks 26006339–26007416 · captured 2026-09-18 22:36:19 UTC · keyless

  66.2% of Uniswap v2 / WETH volume is 3 wallet(s) buying back what they just sold (29 round-trips, same transaction)

  pool                         prints organic  round-trips                sandw  q→fill p50 / p90     tax     liq
  Uniswap v2 / WETH               770     712  29 · 3 wallet(s) · 66.2%       0  15.5 / 60.9 bps      0/0   $956k  ◀ route
  Uniswap v2 / USDC                17      17  —                              0  66.5 / 137.8 bps     0/0    $39k
  Uniswap v2 / USDT                13      13  —                              0  71.1 / 135.2 bps     0/0    $34k

  ▶ route via Uniswap v2 / WETH · cap slippage at 0.65 %   (organic p90 60.9 bps, 1 candidate pool(s))
wrote docs/proof/moto.json  (8.6 s wall clock · 11 calls, every one HTTP 200 · 0 credits — keyless)

145

tests · 139 offline, 6 live

1,000

generated blocks, 0 violations

3.3 ms

engine p50 / 800 prints

8.6 s

live · 800 prints · 11 calls

2026-09-18 22:36:19 UTC

66.2 %

29 round-trips · 3 wallets — the capture behind the page

2026-09-18 23:00:18 UTC

27.5 %

8 round-trips · 1 wallet — 24 minutes later, 11.3 s

2026-09-19 00:28:45 UTC

0.0 %

0 round-trips, a quiet window — and the tool said so, 11.0 s

What this is not · the full list of six is on the README

Where it stops.

The window is 800 prints, not 24 h

3.6 h on the hero, 20 h on SHIB. The span is on every row; the receipt carries the window's share of the pool's 24 h count — 15.7 % here.

Uniswap v3 fee tiers of one pair merge

The feed carries no pool address. A row marked ×n is n pools sharing (venue, base, quote) — marked, never guessed. Filed as FEEDBACK.md §6.

Quote-to-fill is a realised, fee-inclusive move

Consecutive opposite-side prints straddle the fee twice, so p90 sits near 60 bps on every busy 0.30 % pool. It is used comparatively, across pools of one token in one window.

A round-trip is a shape, not a verdict

Same wallet, same block, opposite sides, size-matched. The rows and the definition are printed; nobody is labelled a bot. The anonymous tier throttles per IP — the tool backs off and exits 75 if the quota is spent.

The ask

Run it before
you score it.

Repository github.com/edycutjong/middleman Live page middleman.edycu.dev For judges /judge · no key, no cookie

Demo video see the BUIDL page

about ten seconds from a fresh clone · no key · no install

$ git clone https://github.com/edycutjong/middleman.git
$ cd middleman
$ python3 scripts/middleman.py
$ python3 scripts/verify_tape.py   # every receipt vs its tape, offline

Between your quote and your fill:
two rows and one division.