For judges

Everything on one page. No setup, no key, no account, no cookies. Built on the CoinMarketCap API for the Build with CMC: API Hackathon — Markets and Trading Tools.

The claim

A DEX trader sees a quote and a fill and cannot tell what stood between them. Middleman orders a token's real prints by block and log index, joins them on the maker address, and names it: the wallet on both sides of a block, or the pool itself. Then it says where to route and what slippage cap to set.

The 30-second path

Read the headline on the front page. 66.2 % of Uniswap v2 / WETH volume is 3 wallets buying back what they just sold, in the same transaction. CoinMarketCap ranks that pair #1 on Uniswap v2 by transactions; the join shows why.
Open the raw rows. Two orange rows: same maker, same block, same transaction hash, opposite sides, size within 5 %. Divide a1 by a0 and you have the price each leg paid — the panel does the arithmetic inline, the JSON is verbatim from the API.
Open /evidence. Every request behind the page — URL, HTTP status, UTC, sha256 of the body, the envelope's credit_count (1 per keyless call, against no account — 0 charged). 108 calls across 10 tokens.
Paste any token into the box on the front page. The browser runs the same engine on rows fetched live through a keyless proxy. Or run it from a clone — the command is below and takes about ten seconds.

Receipt — the capture behind the page, 2026-09-18 22:36:19 UTC

wall clock	8.56 s — 8 keyless pages of 100 prints + 3 labelling calls, no backoff fired
prints	800 — MOTO on Ethereum, chosen by rule: the base token of the #1 uniswap-v2 pair on ethereum by 24h transactions at capture time
API calls	11, every one HTTP 200
credits used	0 — keyless /public-api; no key exists to charge
credentials	none — every CoinMarketCap environment variable explicitly unset (none — CoinMarketCap keyless /public-api surface)
round-trips	29 · 3 wallets · 66.2 % of the pool's volume ($127,254 of $192,275), all inside one transaction
sandwiches	0 on this pair · 2 in 8,000 prints across the 10-token census, both by one wallet
quote-to-fill	organic 15.5 bps median, p90 60.9 — the raw tape says 55.3
route	Uniswap v2 / WETH · cap slippage at 0.65 % (organic p90 60.9 bps)
tests	145 tests (139 offline, 6 live) · property-based: 1,000 generated blocks, 0 violations · JS ↔ Python parity on every committed tape · the proxy's boundary pinned by test
engine latency	p50 3.3 ms per 800 prints (p95 3.7 ms, n=200)
live fetch latency	p50 1.5 s per page (p95 16.4 s, n=8 — one iteration sat through the throttle backoff)
raw receipts	moto.json · census.json · live_run.json · live_run_quiet.json · bench_live.json · bench_replay.json

The same command 24 minutes later measured 27.5 % — the wallets had gone quiet; an hour and a half after that the window was clean; two days later the rule picked a different pair and the join found 17 sandwiches in one window. A number that moves with the market is the proof it is measured, not asserted. All four transcripts are in DEMO.md.

Reproduce — the real path

git clone https://github.com/edycutjong/middleman.git && cd middleman
python3 scripts/middleman.py                                   # the hero by rule, live, keyless, ~10 s
python3 scripts/middleman.py --address 0xbd965230588eaa536de6aa45e8ebbc01638535e0 --symbol MOTO --pages 8 --json today.json
python3 scripts/verify_tape.py                                 # every receipt vs its tape, offline
make test                                                      # 139 offline tests
make test-live                                                 # 6 tests against the real contract

CI / deterministic replay — not the product: make bench times the engine over data/tape_moto.json, a recording made by scripts/seed.py. Nothing on the judged path reads a tape; scripts/middleman.py always fetches live and has no offline flag. If CoinMarketCap's anonymous tier is throttling your IP the run says so and backs off; an exhausted quota exits 75 with the way through.

Honest limitations

A run measures the last 800 prints, not 24 hours — 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 (16% here).
Uniswap v3 fee tiers of one pair merge — the feed carries no pool address. A row marked ×n is n pools; the number is per pair, not per tier.
Quote-to-fill is fee-inclusive and realised — p90 sits near 60 bps on every busy 0.30 % pool because consecutive opposite-side prints straddle the fee twice. Used comparatively, across pools of one token.
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.
The anonymous tier throttles per IP and the paste box on this deployment shares one. The page renders from committed receipts, the proxy caches 60 s, and the CLI is the primary live path.

Links

run it	DEMO.md — four real transcripts with receipts
how it works	ARCHITECTURE.md — derived from the code · docs/METHOD.md — the definitions
API feedback for CMC	FEEDBACK.md — eight dated, evidenced findings
the engine	middleman/ — six stdlib modules · scripts/middleman.py — the door
the tests	tests/ — 145 tests, each regression named for the defect it pins
live page	middleman.edycu.dev · /evidence · /api/health
repository	github.com/edycutjong/middleman · MIT
event	Build with CMC: API Hackathon — Markets and Trading Tools · @edycutjong

This page is rendered from the committed receipts by scripts/render_site.py and mirrors JUDGE.md in the repository. No number on it was typed by hand; make check fails if it drifts.