How to Read This Site in Five Minutes — and Ask an AI About Every Study

In shortThis site holds 83 studies of second-by-second order-book data and the trading bots built from them. Reading it all takes days; this page gets you through it in five minutes — where to start, the main findings in eight lines, how each article is built — and gives you the whole site as one Markdown file you can hand to any AI and ask questions. Research, not financial advice.

① What this site is

Market Flow Research is one long experiment: record the Binance order book every second for six coins — BTC, ETH, SOL, BNB, XRP and DOGE — and find out, honestly, what that data can and cannot predict. Then build trading bots from what survives and run them live.

The site holds 83 studies. Each one asks a single question, answers it with numbers taken straight from the research logs, and says plainly when the answer is no. Losing bots are published with every trade, the same as winning ones. The research is done with the help of AI and reviewed by a human.

② Your five-minute path

1. Two minutes — the result. Read Line 7: The Best of All Our Bots — Live Trading Report, Every Trade. It is where all the research led: a scoreboard of every bot we built, and the one that is ahead of buy & hold in live trading.
2. One minute — the findings. The list below is the whole site in eight lines.
3. Two minutes — one series. Pick the series closest to your question and open its newest article; each article links back to everything else.

The findings, each with its source:

• The order book tells you how far, never which way — size answers (rank correlation 0.30 at one second, still 0.11 at an hour), direction never does (−0.002 at a minute); over 30 out-of-sample days the volatility bet made +4.97 bps a trade and the direction bet lost 5.95. The Order Book Tells You How Far, Never Which Way
• Volume is the fuel, not the steering wheel — volume tells you how big the next move is — almost never which way. Volume Is the Fuel — Not the Steering Wheel
• Walls and drains speak about size, not direction — resting liquidity and liquidity being pulled both say something about what happens next — about size, not direction. The Wall That Goes Quiet: What Resting Liquidity Predicts
• The order book has an alphabet — of 6,561 possible eight-feature states only 333 ever occur; the alphabet describes the market well and predicts it poorly. The Alphabet of the Order Book: 6,561 Possible States, 333 Real Ones
• A gate buys precision with coverage — asking a signal to persist two bars raises precision from 0.209 to 0.238 and throws away 25% of the opportunities. Gates: The Part of a Trading Bot That Decides Not to Trade
• Finding the right places is not enough — Typhoon+ finds 74% of a perfect trader's zones on unseen days, fires false 86% of the time and loses 122% against 93% for buy & hold. Typhoon+: A Signal That Finds the Perfect Trader's Zones — and Still Loses Money
• Change the question, not the network — the order book cannot tell which tenth of a move you are in, but “will the leg end soon?” scores an average AUC of 0.73 across six coins. Change the Question, Not the Network — "Will the Leg End Soon?" Scores 0.73 Where "Which Tenth?" Scored Nothing
• Profitable is not the same as better than the market — 438 of 440 buy-and-sell network pairs ended the validation month in profit; the best made 23.6% while buying and holding BTC made about 29%. One Account, Two Networks — 440 Buy-and-Sell Pairs on Validation, and Why the Best Still Trails the Market
• Live result: the best of our bots is Line 7 — More than a dozen bots were built from this research. Line 7 — two small transformers that read 160 hours of order book as one picture — is the only one ahead of buy & hold in live trading: +5.44% since 5 Oct while BTC moved −5.29%. Every trade is in the report. Line 7: The Best of All Our Bots — Live Trading Report, Every Trade

③ Three series

The home page groups every article into three series, newest first:

• The road to Bot 7 — 31 studies. One question at a time: from zigzag legs to a network that reads 160 hours of order book as one canvas.
• Live bots — every result — 7 studies. Bots built from this research and run live: winners and losers, every trade.
• Order-book research — 45 studies. What second-by-second order-book data can and cannot tell you: walls, drains, volume, market states, reversals.

If you only read one research article, read Bot 7 — From an Order-Book Canvas to Live Trading: Every Stretch, Every Result, Every Real Trade: it puts the whole path to the live bot in one place.

④ How each article is built

Every article has the same shape, so you can skim any of them in a minute:

• In short — the box under the title is the whole article in one paragraph, with its key numbers.
• Numbered sections — ①, ②, ③… each answers one part of the question; the list at the top jumps straight to any of them.
• Charts — every chart has a caption that states its number in words; click a chart to open it full size.
• Video — most studies have a narrated walkthrough; it loads only when you press play.
• Reproduce this study — most articles from the bot series link the research log (a Markdown file, in Ukrainian, with the plan, every stage and every table) and a .zip with the scripts, so anyone can rerun the work on the open dataset.
• The box at the end — where all the research led, and the reminder that this is research, not financial advice.

⑤ Ask an AI: every study in five minutes

Reading all of this takes days. An AI can read it in a minute — so we publish the whole site as Markdown, in three sizes:

How to use it — five minutes:

1. Download mfr-digest.md (or copy its address: https://marketflowresear.ch/mfr-digest.md).
2. Open ChatGPT, Claude, Gemini or any assistant that accepts files, and attach it.
3. Ask your question. The file tells the AI to put an article link next to every number, so you can check each answer at its source.

Questions that work well:

• What can the order book predict, and what can it not?
• Which bot worked best live, and how sure can we be about it?
• Give me the five most important numbers on this site, each with its article link.
• What failed, and what did each failure teach?
• I have ten minutes: which three articles should I read, and in what order?

For a deep dive, attach llms-full.txt instead: it is about 115 thousand tokens, which fits assistants with a long context window. Both files are rebuilt every time an article is published.

⑥ Not financial advice

Research, not financial advice. Everything here — the studies, the bots, their trades — is published so the method and the numbers can be checked. Nothing on this site is a recommendation to buy, sell or copy a trade, and past results say little about the next one.

If you find a mistake, it matters to us: the research only works if the numbers are right. You can reach us on X or YouTube.

All studies as one .md for AI · All research