The STAMPEDE bison: a pixel-art bison facing right, black body with grey horns and a red face-and-chest plane, mid-step

background: brand illustration · not market data

Observed wallet rotations · Robinhood Chain · read-only · source available

Follow the move. Inspect the evidence.

One address sold coin A, then bought coin B. Inspect the wallets, timestamps and transactions behind one recorded hour on Robinhood Chain.

Scrub the recorded hour Run it locally

Recorded sample · 2026-09-10 · no API keys · no wallet connection · nothing on this page is live

One observed sequence, three states

recorded sample · 2026-09-10 · grade clean

  1. 1Sold

    Wallet 0xd60c…7ab6 sells Piecoin·1a01 on the PONS curve.

    block 59,570,827 · sell tx 0x17da…a56d

  2. 2Same address verified30-minute pairing window

    Both transactions belong to the same address. Grade clean means no other sold token appears in the pairing window — so this sequence counts for the route.

    same address · window 1,800 s · grade clean

  3. 3Bought

    866 s after the sell, the same wallet buys TruffleHog·7b08. 54 distinct wallets made this move between 17:16:43 and 17:24:39 UTC — 883 sequence rows.

    block 59,579,393 · buy tx 0x20ed…4729

Choose a state with the buttons above, ← → or 1–3. The picture is a brand illustration; the facts are in the text.

recorded sample
60 min of PONS v2 trading2026-09-10 · 16:29:57–17:29:57 UTC
trades
146,329verified against receipts
wallets
33,743distinct addresses
observed sequences
12,164direct + clean · 30-minute window
release
v0.1.017 MB bundle · unpacks to 139 MB

Inspect the evidence

Pick a route. Read the transactions.

The six routes with the most distinct wallets in the recorded hour. A route is “sold A, then bought B within 30 minutes, same address”. Below: one row per wallet, and each row is two transactions you can open on Blockscout.

  1. 124TruffleHog·e085 → LUNAR·55b0130 rows
  2. 66EDEN·c9b6 → AUTON·5eab96 rows
  3. 54Piecoin·1a01 → TruffleHog·7b08883 rows
  4. 43EVERYTHING·b22f → MARIO·6819170 rows
  5. 38LIQ·0785 → MOTIF·f8dc38 rows
  6. 37PUCHATO → STEPPER37 rows
  7. wallets · route · sequence rows. Weight uses distinct wallets, never rows. Extract: data/routes.json, derived from demo/sample.json.

recorded sample · 2026-09-10 · 30-minute window

54 wallets sold Piecoin·1a01, then bought TruffleHog·7b08

883 sequence rows · buys 17:16:43–17:24:39 UTC · gaps 396–875 s · 0 direct · 883 clean

walletsold atbought atgapgradetransactions
0x989f…fb0f17:10:1017:17:42452 scleansell 0xd723…ff82 · buy 0xe14c…e631
0xae37…ba8017:10:0917:24:14845 scleansell 0x1283…ab89 · buy 0xcaa1…a047
0xf54f…989017:10:1117:24:20849 scleansell 0xa6d8…5719 · buy 0x10c5…3906
0xca17…755217:10:0917:24:28859 scleansell 0x3015…9717 · buy 0xb025…83a7
0x8aa7…8a7f17:10:0917:24:31862 scleansell 0xf8b9…a4a2 · buy 0xdac2…69f2
0xd60c…7ab617:10:0917:24:35866 scleansell 0x17da…a56d · buy 0x20ed…4729
0xd726…739717:10:0917:24:36867 scleansell 0x48e4…d4e7 · buy 0xebdc…8d0b
0xb4f5…7de417:10:0717:24:39872 scleansell 0x8dbf…0a55 · buy 0x2135…c2dd

8 of 54 wallets, one row each — the wallet’s last buy on this route — spread evenly across the buy times; the wallet from the README walk-through is included. Times are UTC. All 883 rows are in the recorded sample.

Grades: direct — sell and buy in the same transaction · clean — the wallet sold only A in the window · ambiguous — it also sold other coins; listed in the app, never counted in the weight. Extract: 97 routes · 4,222 rows · 1,757 wallets of the hour’s 12,164 direct + clean sequences.

Move through the whole hour at 1×, 20× or 60× — every row with its wallet and both transaction hashes, in the browser, from a static file.

Open the recorded sample

Four views · one dataset · one shared clock

The views rank, trace and map the same facts

A rotation is a fact about order: the same address sold A, then bought B within the pairing window. None of the views claims that the sale funded the purchase, that addresses share an owner, or that anything will happen next.

1 RADARscreenshot · recorded sample · 17:29:57 UTC
RADAR table of 146 coins at 17:29:57 UTC with row 15, TruffleHog·7b08, selected: inflow 54 wallets in 10 minutes, ×2.0 acceleration, score 70; the coin drawer on the right shows on-chain facts and the rotation source Piecoin·1a01 → 54 wallets

Which coins are wallets rotating into right now?

One row per coin: distinct wallets that sold something else and bought this one in the last 10 minutes, acceleration against the previous 10, source coins, age, curve stage — and one score with every part shown next to the total. Presets are filters over observations, not recommendations.

Score parts, presets, context → docs/RADAR.md

2 FLOWscreenshot · recorded sample · 2026-09-10
FLOW diagram: on the left the coins sold first — Piecoin·1a01 with 54 wallets and TruffleHog·bd82 with 1 — a grey ribbon into the red node TruffleHog·7b08 in the middle, and an empty right column labelled no rotation outflow in this range

Where did this coin’s wallets come from, and where did they go?

Sources on the left, destinations on the right, ribbon width = distinct wallets. Clicking a ribbon opens the underlying transactions instead of a summary. When nobody has rotated out yet, the right column is empty — and says so.

Walkthrough of this screen → GETTING-STARTED §5.3

3 MAP · evidence Escreenshot · recorded sample · 2026-09-10
MAP presentation layout with the caption 54 WALLETS SOLD Piecoin·1a01 → BOUGHT TruffleHog·7b08, 883 observed sequence rows, and the evidence panel on the right listing wallets with their sold and bought transactions, blocks, times and grades

What does the whole hour look like — and what exactly are the trades?

Coins as nodes, observed rotations as directed edges, an autopilot over the strongest routes. The evidence panel is the product: for each wallet on a route, the sell and the buy transaction with block, time, grade and a Blockscout link.

Checked examples, including negative cases → docs/EXAMPLES.md

4 TRADERS experimentalscreenshot · recorded replay · 2026-09-12
TRADERS table in a paused recorded replay: 18,721 wallets ranked by estimated ETH PnL after fees, columns for trades, coins, win rate, hold, exit quality, rug avoidance, sniper share and quality; several rows carry the tags BOT, CONTRACT and FEES PARTLY ESTIMATED

Which wallets keep showing up, and what do their trades add up to?

Wallets ranked by estimated ETH result after fees, with hold times, exit quality and tags for bots, contracts, deployer-linked and sniper wallets. Where fees had to be estimated, the row says so. A research view with stated assumptions — not a track record, not a list to copy.

Definitions, fee model, caveats → docs/RESEARCH-TRADERS.md

Recorded replay · 8-second excerpt · MP4, 1.0 MBrecorded replay · 20× · not a live market feed
First frame of the recorded replay: the MAP view with the tape of observed sequences on the right

The full 44-second take (terminal intro, four autopilot stops, RADAR, FLOW) is a release asset: stampede-v5-autopilot.mp4 (16.7 MB) · frame notes in docs/DEMO-V5.md.

screenshot a still of the product from the recorded sample or a live session, with its capture time · recorded replay a video excerpt of the recorded hour, not a market feed · live backend a screen served by serve --mode live on the author’s machine with his own key — evidence that live ingestion exists, not a hosted dashboard.

Three steps · recorded demo · no keys

From an empty terminal to a rotation and its transactions

Requirements: Git, uv (it installs Python 3.13 itself) and Node 22+ for the web views. The verified path is macOS on Apple silicon; Windows and Linux are not verified. The first run downloads dependencies and the 17 MB sample; after that the demo runs from local files.

  1. 1

    Install

    Clone the repository, create the Python environment, build the web interface once. The build produces static files in web/dist; it does not publish anything.

    git clone https://github.com/Argona7/stampede && cd stampede
uv sync
cd web && npm ci && npm run build && cd ..
  2. 2

    Open the recorded demo

    The server fetches the sample from the public release on first start, then serves the web views on your own machine. It starts paused: press Play or Space. No API keys are read in this mode; external context (X, market data, holders) stays off.

    uv run stampede demo
# → http://127.0.0.1:8791/   replay 20×, paused
  3. 3

    Inspect one rotation and its transactions

    Let the clock run to 17:29:57 UTC. In RADAR find TruffleHog·7b08 (type truffle in find a coin), press Enter for FLOW, click the ribbon from Piecoin·1a01, then press E on the MAP route. Every evidence row links its sell and buy transactions on Blockscout.

    Step-by-step with screenshots → GETTING-STARTED §5

    keyboard · web interface

    • 1–4 RADAR · FLOW · MAP · TRADERS
    • Space play / pause the shared clock
    • E evidence rows of the selected route (MAP)
    • D coin drawer / wallet card
    • A autopilot on / off (MAP presentation)
    • P presentation ↔ explore layout (MAP)

Live chain data is a separate setup. uv run stampede serve --mode live tails the chain head and needs an Alchemy key in .env; keys are read from .env only and never printed. Expect first-time indexing and provider rate limits. The recorded demo is the supported path. docs/DEVELOPMENT.md

Optional terminal UI on the same clock, in a second window of 120×36 or larger: uv run stampede terminal --api-url http://127.0.0.1:8791

What this shows — and what it does not

Order of trades, verifiable on the chain. Nothing more.

Order, not causality

A rotation is the observed order of trades by one address inside a window. STAMPEDE does not prove that the sale funded the purchase, does not attribute addresses to people or teams, and does not predict what happens next. The grade tells you how clean the observation is; the transactions let you check it.

docs/ALGORITHM.md · docs/EXAMPLES.md

One hour of one launchpad

The shipped sample is 60 minutes of PONS v2 trading on 2026-09-10, verified against receipts. Longer stores are being backfilled over resilient HyperSync streams; coverage, rejected cases and the block-time method are documented rather than assumed. Counts on this page from the filtered extract are lower bounds for the hour.

docs/COVERAGE.md · docs/SOURCES.md

Scores and wallet research stay labelled experimental

The RADAR score is a ranking of observed inflow with every part visible. The repository also carries an experimental verdict column and a research pipeline for trader quality; both were measured on hours of one launchpad, both are labelled experimental in the interface and the docs, and neither is advice. The research note on “runners” describes one five-hour sample — most such spikes ended below the alert price by the end of the horizon.

docs/RESEARCH-RUNNERS.md · docs/RESEARCH-EDGE.md · docs/RESEARCH-TRADERS.md

Live ingestion is a labelled mode, and a work in progress

serve --mode live tails the head of Robinhood Chain (Alchemy, public RPC or HyperSync), keeps late log fragments with their transaction and reconnects over the other endpoint when a subscription goes quiet. Live data is badged LIVE on every screen and shows a provider error instead of a frozen picture. Interface fixes for LIVE are listed only once verified. Planned, not shipped: live mode on the public RPC alone, CSV/JSON export of a route’s evidence rows, multi-hour recorded samples.

docs/ENGINE.md · README → Development, Known limits, Roadmap

RADAR · live modescreenshot · live backend · 2026-09-12 14:35 UTC · not current state
RADAR in live mode on the author's machine, badge LIVE, 14:35:15 UTC, data age 1 s: 21 coins a few minutes old ranked by rotation inflow, with launch-intel columns mostly n/a and an experimental verdict column reading WAIT or AVOID

Captured on the author’s machine with his own key. This website does not connect to any backend; there is no hosted dashboard.

Questions

Short answers

Which venues and which network?

PONS v2 bonding curves and the graduated Uniswap v4 pools their coins move into (V2MemeHook) on Robinhood Chain, chain id 4663. Other DEXes and plain transfers are outside the universe. Coins with the same ticker are told apart by the last four hex digits of the address, as in Piecoin·1a01.

Is there a hosted version I can open in the browser?

No. STAMPEDE is local software: the web views are served on your own computer at a localhost address, and nobody else sees your instance. This website is documentation and a path to the setup guide; it does not talk to any backend. The one thing you can use here is the recorded sample — a fixed, static extract of the public hour with a clock you can scrub; it is recorded data, not the application.

Do I need API keys or a wallet?

Not for the recorded demo: it replays the shipped hour from local files and never calls an API. Live mode needs your own Alchemy key in .env; keys are read from .env only and never printed. There is no wallet connection in either mode, and nothing is ever sent to the chain.

Does a rotation prove the sale funded the purchase?

No. A rotation is the observed order of trades by the same address inside a window. STAMPEDE does not prove money flow between coins, does not attribute addresses to people or teams, and does not predict what happens next. The grade tells you how clean the observation is; the transactions let you check it yourself.

How far should I trust the score, the verdict and the trader stats?

Treat them as experimental. Thresholds were calibrated on five hours of one launchpad; the shipped sample is one hour. Interpolated timestamps carry ≈, external context shows n/a when unknown (unknown, not zero), and fees in the trader view are estimated where the store lacks fee columns. Nothing here is a return, a forecast or a recommendation.

What is the moving picture behind the headline?

A brand illustration: a generated, silent 8-second camera move along a pixel hoofprint trail, made from the approved STAMPEDE hero still. It is environment only — it contains no chart, no price, no market data and no product screen. The three states of the sequence next to it are real rows from demo/sample.json. It plays once when allowed, never loops, and stays a still image under reduced motion or when you turn motion off.

Where is the source, and what is the license?

Source: github.com/Argona7/stampede. Guide: docs/GETTING-STARTED.md. A license has not been chosen yet — until a LICENSE file lands, the code is source-available for reading and running the demo, not for redistribution. Found something odd? Open an issue with the transaction hashes you looked at.

Run it on your machine.

Four commands, a recorded hour, and the transactions behind every number.

Get started View source