DEV Community: Toby Water

I Mapped My Entire One-Person Business on One Page — Here's the System

Toby Water — Fri, 23 Oct 2026 03:20:05 +0000

I Mapped My Entire One-Person Business on One Page — Here's the System

I run a small digital-products store by myself. No servers, no staff, no investors — just me, a stack of free tiers, and a set of automations I set up months ago and have been trusting ever since.

Here's the thing I couldn't honestly answer until recently: if you asked me to show you how the whole business fits together — how a stranger on the internet becomes a person holding one of my files — I would have had to wave my hands. The pieces live on five different platforms. dev.to brings in readers. A static page on Cloudflare Pages is the storefront. NOWPayments collects the money through one hosted link per product. An Outlook inbox handles receipts and support. The products themselves are PDFs and ZIPs sitting in local folders.

I knew every piece. I had just never seen the whole thing at once. Neither had anyone else — which is a polite way of saying the system's documentation was "it's all in my head," and my head had never actually assembled it in one place.

One page, two views

So I mapped it. The result is a single page with two views of the same business.

The architecture view is four horizontal lanes. Acquisition: dev.to articles in three languages and a free sample pack that trades 20 of my 100 phrase chunks for an email address. Storefront: the one-page catalogue — eight products and three services — plus three landing pages, one per Thai consulate, for my visa guide. Payments: the NOWPayments links and the custody wallet they settle into. Fulfilment & support: one inbox, one operator (me, AI-assisted), and delivery by email within 24 hours.

Every box on the map is clickable. Clicking it tells you what the component does, what it runs on, and — this turned out to be the important part — where it can break.

The flow view walks a single order through six steps: Discover, Evaluate, Pay, Prove, Verify & deliver, Support & upsell. The full clickable version lives on my site (tobywater.pages.dev); it's the actual sample I now show potential clients, drawn from the real store, so every box can be checked against the live thing.

I expected the map to be a tidy summary of what I already knew. Instead, three boxes started staring back at me.

Problem 1: One inbox is my entire fulfilment department

Every order receipt, every sample request, every support question lands in a single Outlook alias: toby.water@outlook.com. There is no queue system, no second mailbox, no fallback. If that account ever locks me out, orders don't bounce — they silently stall, and buyers experience it as a store that took their money and vanished.

I won't pretend I solved this. What I did was stop letting it be invisible. The weakness is now written on the map, in the box, where I have to look at it. Around it I've put the two mitigations that honestly exist: the public alias is separate from my login identity, so the address I print everywhere isn't the credential that matters, and an automated watch polls the unread mail every five minutes so nothing sits unseen for hours. Real redundancy is still on the to-do list. The difference the map made is that it's a known item on a list, instead of a surprise waiting for a bad day.

Problem 2: There's no webhook — reconciliation is me

NOWPayments settles a payment, and then… nothing pings my side of the fence. There is no webhook, no order database, no automatic "payment received, ship it" trigger. The join key between a payment and a person is an email: the buyer forwards their receipt to my inbox, and I match it against the payment record by hand before sending the files.

Drawn as a diagram, that's a dashed arrow labelled "buyer emails receipt" doing the job an API should do. It looks primitive because it is. At my current volume it's also deliberate — manual matching costs minutes a week, and the receipt email opens a human support thread from minute one, which a faceless download portal wouldn't. But I handle it with rules, not vibes: buyer email is required at checkout precisely so the match is possible, and if a receipt doesn't match a payment, I ask for the transaction ID instead of guessing.

The map's real contribution here: when order volume grows, I now know exactly which arrow gets automated first, instead of vaguely feeling that "fulfilment should probably be automated someday."

Problem 3: The catalogue has a ceiling, and I can count it

The entire store is one static HTML page, deployed as a single zip upload. At eight products and three services, that's a feature — the whole business is one artefact I can redeploy in a minute. The map forced me to write down what I'd been avoiding quantifying: this does not survive past roughly fifteen products. Beyond that, a hand-edited page becomes the bottleneck and the store needs a generator.

So I did the counterintuitive thing and shrank the catalogue. A cross-audience bundle I'd been carrying — products for English learners bundled with products for sellers — underperformed and added exactly the kind of page bulk that brings the ceiling closer. It's gone; the catalogue now sells per-audience only. The generator is deferred, but deferred knowingly, with a number attached.

A note on method, because it matters

The map was drafted by AI from my system notes, then checked by me — every box, every arrow — against the real setup, line by line. That second step is not optional garnish. Language models produce plausible diagrams: confidently drawn boxes for components that don't exist, arrows pointing in the direction that reads well rather than the direction data actually flows, a database materialising because diagrams are supposed to have one. For your own system you can catch every one of those — but only by reading, not by generating. A map nobody verified is decoration. A map somebody verified is an audit you can see.

Steal the exercise

If you run a small product and couldn't draw its whole system on one page from memory, that gap is worth an afternoon. Mine turned a vague sense of "it's fine, mostly" into three specific, written-down weaknesses — one mitigated, one deliberately manual with rules, one deferred with a number on it.

If you'd rather not do the drawing yourself, this is a service I sell as Codebase Maps (tobywater.pages.dev): your codebase or system, AI-drafted and human-checked line by line, delivered as one self-contained clickable HTML file, in three tiers from $39 to $279 depending on depth. Questions welcome by email — it goes to the famous single inbox, which is polled every five minutes.

Disclosure: this post, like the map it describes, was produced with AI assistance and reviewed by me line by line — which is rather the point of the article.

Vibe-Trading Hit 36K Stars in Six Months. I Read the Repo So You Don't Have To.

Toby Water — Fri, 23 Oct 2026 03:09:25 +0000

Disclosure: This article was researched and drafted with AI assistance, then reviewed and edited by a human. All figures below are taken from the project's public GitHub page as of October 23, 2026.

Every few months, a GitHub repo shows up that makes the whole "AI trading" corner of the internet lose its mind. This time it's Vibe-Trading (github.com/HKUDS/Vibe-Trading), from HKUDS, the data science lab at the University of Hong Kong, billed as "your personal trading agent."

The numbers are not hype-sized, they're verifiable: created on April 1, 2026, the repo has passed 36,300 stars and 5,800 forks, with 3,324 commits from 33+ contributors and a release cadence that is close to daily — v0.1.18 shipped on October 22. In a space where the typical "AI hedge fund" repo has 40 commits and a README written like a movie trailer, that maintenance profile alone is worth a closer look.

So I read it properly. Here's what it actually is, where the real value sits, and why the most interesting part of the project is the part its own authors tell you to be careful with.

What It Actually Is

Vibe-Trading is an open-source (MIT) research and trading-agent framework in Python. Strip away the branding and you get four components:

A self-improving trading agent that runs research loops — pulling market data, forming hypotheses, testing them, and keeping what works in memory across sessions.
Multi-agent "trading teams" — specialized agents (research, analysis, risk, execution support) that collaborate on a workflow instead of one model doing everything in a single chat.
A cross-market data and backtesting layer, including a built-in library of 478 alpha factors you can test, combine, and inspect.
A Shadow Account / paper trading setup — simulated execution against live-ish data, so strategies can run forward without real money on the line.

It ships with multiple install paths (PyPI one-liner, Docker, an MCP plugin so other agents can call it as a tool), connectors for various brokers and data sources, and a web UI. This is not a prompt pack with a landing page. It's a working codebase with the boring infrastructure — session memory, scheduled research, an API server, security defaults — that real tools need and demo projects skip.

Why It Grew So Fast

Three reasons, I think.

First, timing. "Let an agent watch the market for me" is the fantasy every retail trader has had since LLMs got tool use. Vibe-Trading is one of the first open-source projects to package that fantasy into something you can actually pip install.

Second, credibility by contrast. The repo's competitors in the trending lists are often one-person marketing exercises — thousands of stars, a few dozen commits, promises of fully autonomous profit. HKUDS is a university lab with a track record, the contributor graph is real, and the changelog reads like a product being built, not a token being promoted.

Third, the paper-trading-first design. The headline feature isn't "it trades for you." It's that you can run strategies in simulation, watch the agent's reasoning, and inspect the factor library. That's immediately useful to students, quant-curious developers, and small teams — a much bigger audience than people willing to hand an LLM their brokerage credentials.

Where the Real Value Is

If you strip out every trading fantasy, three things remain genuinely valuable:

The factor library as a curriculum. 478 documented alpha factors is, functionally, a free course in quantitative strategy design. Reading how factors are defined, categorized, and combined teaches you more about systematic trading than a month of YouTube. Even if you never run a single backtest for production, the repo is a reference implementation of ideas that are usually locked inside textbooks and paywalled courses.

Paper trading as a discipline tool. Most retail losses come from strategies that were never forward-tested at all. A shadow account forces the honest workflow: write the idea down, let it run against data you didn't cherry-pick, read the results. The agent layer makes that loop cheaper to run consistently — scheduled research, journaling, and cross-session memory mean the experiment actually continues instead of dying after one enthusiastic weekend.

The multi-agent workflow pattern. Dividing research, risk-checking, and reporting across cooperating agents with tool contracts is a pattern that transfers far beyond trading. If you build internal tools for any small team, the architecture — agents with scoped tools, structured handoffs, human approval gates — is the part worth stealing.

The MCP surface. Vibe-Trading also ships as an MCP plugin, which means other agent environments can call its research and backtesting tools directly instead of treating it as a standalone app you babysit. That's a quiet but important design decision: the project is positioning itself as infrastructure inside the broader agent ecosystem, not just a destination. Tools that plug into other people's workflows tend to outlive tools that demand to be the workflow — and it's one more reason the repo's growth looks structural rather than accidental.

The Part the Authors Themselves Flag

Here's the detail that separates this project from the grift end of the AI-trading pool: its own Disclaimer is blunt. Live trading through broker connectors is described by the project as experimental and not validated on real brokerage accounts, and the codebase reflects that caution. There's a TAP mode — full credential isolation with human-approved writes — meaning the agent can prepare an action but a person has to approve it before anything touches a real account. There's a /halt switch for the live runner. These are the design choices of people who understand exactly how this can go wrong.

Take the hint. An LLM-driven agent with execution permissions is an operational risk even when the strategy is sound: malformed orders, runaway loops, costs you didn't model. The project's SECURITY file also warns that scammers have already issued fake tokens using the Vibe-Trading name — the project has no token, and anything claiming otherwise is fraud. If a "community" or "presale" attaches itself to this repo's brand, walk away.

My rule, and the one I'd suggest to anyone experimenting: paper trade until the strategy has survived months of forward data and your own skepticism, and treat any live deployment as an engineering risk decision, not a vibe.

Three Sane Ways to Use It

1. Learn factors, properly. Clone the repo, work through the factor library, reproduce a few backtests, and break them on purpose. Cheapest quant education available right now.

2. Run a shadow portfolio. Pick a simple strategy, let it paper trade, and review the agent's research notes weekly. You'll learn more about your own decision-making than about the market — which is the actual bottleneck for most people.

3. Borrow the workflow, not the trades. Use the multi-agent research pattern — data pull, hypothesis, test, report, human gate — inside your own team for market monitoring, competitor tracking, or content research. The trading is the demo; the workflow is the product.

Vibe-Trading is still pre-1.0 software (v0.1.x), moving fast, and it will break in places. But it's the rare AI-trading project where the honest summary is more impressive than the hype: a well-maintained open-source lab for systematic research, built by people who put the warnings in the README themselves.

I run Tobywater Digital, where we help small teams deploy and actually operate open-source AI tooling — including research workflows like this one — with the boring parts (setup, guardrails, human approval gates) done properly. If you'd like a guided setup or a paper-trading workflow configured for your team, that's the kind of deployment work we do. No return promises, no trade calls — just tooling that runs.

Pare de decorar palavras. Roube estes 20 chunks de inglês

Toby Water — Fri, 23 Oct 2026 01:58:13 +0000

Pare de decorar palavras. Roube estes 20 chunks de inglês

Você já sabe milhares de palavras em inglês. Lê artigo, acompanha documentação, passa em prova. Aí alguém faz uma pergunta simples numa call, e o seu cérebro começa a montar a frase igual a quem monta um móvel com parafuso solto: uma palavra de cada vez, caçando o verbo certo, a preposição certa, a ordem certa. Quando a frase finalmente existe, a conversa já andou sem você.

O problema não é o seu vocabulário. É o tamanho da sua unidade.

Nativo não constrói frase palavra por palavra. Ele pega chunks da prateleira: unidades de várias palavras prontas para uso, tipo it depends on…, get back to you e just in case. Os linguistas chamam isso de linguagem formulaica, e, pela maioria das estimativas, uma fatia enorme da fala do dia a dia é feita disso. Seu cérebro guarda um chunk do mesmo jeito que guarda uma palavra solta: como um item só. Um slot de memória, uma busca só, e sai uma frase inteira, gramaticalmente certa, com as preposições já soldadas.

É por isso que chunk ganha de lista de palavras em três frentes:

Velocidade. Uma busca em vez de cinco: você para de traduzir no meio da frase.
Gramática de graça. Ninguém que treina o chunk fala "depends from" ou "make sure to locking the door": a estrutura certa vem embutida.
Naturalidade. Chunk é o jeito que nativo fala de verdade. I'll get back to you soa como pessoa. I will respond to you later soa como termo de uso.

E tem um dividendo no listening: quando um chunk mora na sua cabeça como uma unidade, você passa a ouvir ele como unidade. O inglês rápido para de soar como um borrão de palavras soltas e começa a se separar em blocos que você reconhece. É por isso que quem aprende por chunks costuma perceber que o ouvido melhora antes da fala: você não está mais decodificando, está reconhecendo padrão — igualzinho ao que seu cérebro faz em português.

Então para de colecionar palavras. Começa a roubar chunks. Aqui vão 20 que valem o roubo hoje, agrupados pelo momento em que você vai usar de verdade.

Para decidir

How about…? — a abertura universal para sugerir qualquer coisa. («que tal…?») "How about we ship the simple version first?"
make up your mind — se decidir, depois de pensar. («decide logo») "Take your time — but I need you to make up your mind by Friday."
give it a try — tentar, sem prometer que vai dar certo. («dar uma chance / tentar») "I've never used that framework, but I'll give it a try this weekend."
fair enough — «seu ponto é justo, aceito». O jeito elegante de encerrar uma discussão. "You don't want to deploy on a Friday? Fair enough."
It depends on… — a resposta honesta para metade das perguntas. («depende de…») "It depends on how much traffic you're expecting."

Para investigar e resolver pepino

figure out — resolver pensando. («descobrir como / sacar») "Give me twenty minutes and I'll figure out why the build is failing."
come up with — produzir uma ideia ou uma saída. («bolar alguma coisa») "She came up with a workaround before the meeting even ended."
sort out — desenrolar uma situação bagunçada. («resolver / desenrolar») "Let's sort out the access permissions before onboarding anyone else."
set up — deixar algo pronto para usar. («configurar / montar») "I set up the staging environment so we can test safely."
end up — acabar num lugar que você não planejou. («acabar fazendo…») "We skipped the tests and ended up debugging all night."

Redes de segurança do dia a dia

just in case — por precaução. É o seu «por via das dúvidas», palavra por palavra. "Take an umbrella, just in case."
by accident — sem querer; o oposto de on purpose. («sem querer / por acidente») "I deleted the file by accident — good thing we had a backup."
make sure — conferir, não supor. («confere se… / garante que…») "Make sure the database is backed up before you run the migration."
How come…? — um why mais leve e conversado. («como assim…? / por que é que…?») "How come the deploy took so long yesterday?"
step by step — uma etapa de cada vez, na ordem. («passo a passo») "Don't panic. Walk me through it step by step."

Trabalho e manter contato

get back to you — te responder depois, quando souber. É A frase da daily e do Slack de trabalho. "I don't have that number yet — I'll get back to you by noon."
keep in touch — manter contato quando algo muda. («vamos manter contato») "Good luck with the new role. Let's keep in touch."
keep track of — acompanhar algo ao longo do tempo para nada se perder. («acompanhar / ficar de olho em…») "I keep track of every invoice in one spreadsheet."
in advance — antes do evento, não na última hora. («com antecedência») "Book the train tickets in advance; they sell out on holidays."
You're on mute — a frase mais falada do home office. No seu time, é «você tá no mudo»; na call com o gringo, é assim: "We can see you talking, but you're on mute."

Como aprender de verdade

Não leia esta lista dez vezes. Escolha cinco chunks, fale cada um em voz alta dentro de uma frase sobre a sua vida — seu projeto, seu time, seu fim de semana — e use de verdade nas próximas 24 horas: na daily, numa mensagem, num comentário de PR. Um chunk que você falou uma vez numa conversa real vale mais que cinquenta grifados. Cinco por dia, durante quatro dias, e a lista inteira é sua: não decorada, disponível.

Um hábito a mais, que se paga sozinho: quando você pescar um chunk «na natureza» — um podcast, uma série, a mensagem de um colega gringo — anote a frase inteira da pessoa, não só o chunk. É o contexto que depois te diz quando fair enough é concordância amigável e quando é o jeito educado de encerrar a discussão. Seu caderno tem que ser uma coleção de frases reais dos outros, esperando você se colocar dentro delas.

Estes 20 são só uma prateleira da despensa. Tenho uma amostra grátis de outro conjunto — mais 20 chunks, com frases de exemplo e cartões para imprimir — aqui (PDF), e a coleção completa, com 100 chunks do dia a dia e cartões Anki para revisão espaçada, está na Tobywater Digital. Roube os grátis primeiro. Esse é o método inteiro.

Nota de transparência: este artigo é uma adaptação localizada do meu texto original em inglês, preparada com assistência de IA e revisão editorial humana. Os chunks e os exemplos são o conteúdo de ensino; a adaptação (cenários, equivalências e tom) é o trabalho editorial.

Deja de memorizar palabras. Roba estos 20 chunks de inglés

Toby Water — Fri, 23 Oct 2026 01:56:49 +0000

Deja de memorizar palabras. Roba estos 20 chunks de inglés

Ya sabes miles de palabras en inglés. Lees artículos, sigues la documentación y apruebas exámenes. Y entonces alguien te hace una pregunta simple en una call, y tu cerebro empieza a armar la frase como quien arma un mueble con tornillos sueltos: una palabra a la vez, buscando el verbo correcto, la preposición correcta, el orden correcto. Cuando la frase por fin existe, la conversación ya siguió sin ti.

El problema no es tu vocabulario. Es el tamaño de tu unidad.

Los nativos no construyen frases palabra por palabra. Sacan chunks de la estantería: unidades de varias palabras listas para usar, como it depends on…, get back to you o just in case. Los lingüistas lo llaman lenguaje formulaico, y según la mayoría de las estimaciones, una parte enorme del habla cotidiana está hecha de eso. Tu cerebro guarda un chunk igual que guarda una palabra suelta: como un solo ítem. Un solo espacio de memoria, una sola búsqueda, y sale una frase completa y gramaticalmente correcta, con las preposiciones ya soldadas.

Por eso los chunks le ganan a las listas de palabras en tres frentes:

Velocidad. Una búsqueda en vez de cinco: dejas de traducir a mitad de frase.
Gramática gratis. Nadie que practica el chunk dice "depends from" ni "make sure to locking the door": la estructura correcta viene adentro.
Naturalidad. Los chunks son la forma en que los nativos realmente dicen las cosas. I'll get back to you suena a persona. I will respond to you later suena a contrato de términos y condiciones.

Y hay un dividendo extra en comprensión: cuando un chunk vive en tu cabeza como una unidad, empiezas a escucharlo como una unidad. El inglés rápido deja de sonar como una mancha de palabras sueltas y empieza a separarse en bloques que reconoces. Por eso quienes aprenden con chunks suelen notar que primero mejora su oído y después su habla: ya no estás decodificando, estás reconociendo patrones, igual que en tu idioma.

Así que deja de coleccionar palabras. Empieza a robar chunks. Aquí van 20 que vale la pena llevarte hoy, agrupados por el momento en que de verdad los vas a usar.

Para decidir

How about…? — la apertura universal para proponer algo. («¿Qué tal si…?») "How about we ship the simple version first?"
make up your mind — decidirte, después de pensarlo. («decídete») "Take your time — but I need you to make up your mind by Friday."
give it a try — intentarlo sin prometer que va a salir. («dale una oportunidad / inténtalo») "I've never used that framework, but I'll give it a try this weekend."
fair enough — «tu punto es válido, acepto». La forma elegante de cerrar una discusión. "You don't want to deploy on a Friday? Fair enough."
It depends on… — la respuesta honesta a la mitad de las preguntas. («depende de…») "It depends on how much traffic you're expecting."

Para resolver y arreglar problemas

figure out — resolver algo pensándolo. («descubrir cómo / resolver») "Give me twenty minutes and I'll figure out why the build is failing."
come up with — producir una idea o una salida. («ocurrírsete algo») "She came up with a workaround before the meeting even ended."
sort out — desenredar una situación desordenada. («ordenar / resolver el lío») "Let's sort out the access permissions before onboarding anyone else."
set up — dejar algo listo para usar. («configurar / montar») "I set up the staging environment so we can test safely."
end up — terminar en un lugar que no planeaste. («terminar haciendo…») "We skipped the tests and ended up debugging all night."

Redes de seguridad de todos los días

just in case — por precaución, ante un «por si acaso» (sí, es literalmente tu «por si acaso»). "Take an umbrella, just in case."
by accident — sin querer; lo contrario de on purpose. («sin querer / de casualidad») "I deleted the file by accident — good thing we had a backup."
make sure — verificar, no suponer. («asegúrate de…») "Make sure the database is backed up before you run the migration."
How come…? — un why más suave y conversacional. («¿cómo es que…?») "How come the deploy took so long yesterday?"
step by step — una etapa a la vez, en orden. («paso a paso») "Don't panic. Walk me through it step by step."

Trabajo y mantener el contacto

get back to you — responderte después, cuando sepa. Es LA frase de la daily y del Slack laboral. "I don't have that number yet — I'll get back to you by noon."
keep in touch — seguir en contacto cuando algo cambia. («sigamos en contacto») "Good luck with the new role. Let's keep in touch."
keep track of — seguirle la pista a algo para que nada se pierda. («llevar el control de…») "I keep track of every invoice in one spreadsheet."
in advance — antes del evento, no a última hora. («con anticipación») "Book the train tickets in advance; they sell out on holidays."
You're on mute — la frase más usada del trabajo remoto. En tu equipo dicen «estás muteado»; en la call con el cliente se dice así: "We can see you talking, but you're on mute."

Cómo aprenderlos de verdad

No leas esta lista diez veces. Elige cinco chunks, di cada uno en voz alta dentro de una frase sobre tu vida —tu proyecto, tu equipo, tu fin de semana— y úsalos de verdad en las próximas 24 horas: en la daily, en un mensaje, en un comentario de un PR. Un chunk que dijiste una vez en una conversación real vale más que cincuenta subrayados. Cinco por día durante cuatro días y la lista entera es tuya: no memorizada, disponible.

Un hábito más que paga solo: cuando pesques un chunk «en estado salvaje» —un podcast, una serie, el mensaje de un colega— anota la frase completa de esa persona, no solo el chunk. El contexto es lo que después te dice cuándo fair enough es acuerdo amistoso y cuándo es la forma educada de cerrar una discusión. Tu cuaderno debería ser una colección de frases reales de otros, esperando a que te pongas a ti mismo dentro.

Estos 20 son una sola estantería de la alacena. Tengo una muestra gratuita de otro set —20 chunks más, con frases de ejemplo y tarjetas imprimibles— aquí (PDF), y la colección completa de 100 chunks de uso diario, con tarjetas Anki para repaso espaciado, está en Tobywater Digital. Roba primero los gratuitos. Ese es todo el método.

Nota de transparencia: este artículo es una adaptación localizada de mi texto original en inglés, preparada con asistencia de IA y revisión editorial humana. Los chunks y los ejemplos son el contenido de enseñanza; la adaptación (escenarios, equivalencias y tono) es el trabajo editorial.

The 2026 China Digital Setup Checklist Nobody Updated

Toby Water — Thu, 22 Oct 2026 19:19:16 +0000

The 2026 China Digital Setup Checklist Nobody Updated

Search "how to prepare for a trip to China" and you'll get confident, detailed, well-ranked guides — describing a country that no longer exists. The entry rules were rewritten in December 2024 and have kept moving since. The payment limits changed. The eSIM situation changed. Most top results still quote the old versions of all three.

I'm not going to pretend this post will age better than theirs. Policies move, so everything below is written the way I'd want to read it: dated, specific where the official number is known, and honest about where you should check the primary source yourself before booking anything.

1. Entry: there are now two visa-free doors, and most guides only know the old one

Door one: 30-day visa-free entry. Citizens of around 50 countries — most of Europe, plus Japan, South Korea, Australia, New Zealand, Brazil, Argentina, and a growing list of others — can now enter China with no visa at all for business, tourism, family visits, exchange, or transit, for up to 30 days. Beijing extended this unilateral policy through December 31, 2026, and the country list has grown more than once, most recently in early 2026. If your guide says the scheme "ends in 2025," it's quoting last year's notice.

Door two: the 240-hour transit. Since December 17, 2024, the old 72/144-hour transit-without-visa scheme has been replaced by a 240-hour (10-day) one. It launched covering 54 nationalities, 60 ports of entry, and permitted areas across 24 provinces — and the nationality list has expanded since (Indonesia joined in mid-2025, for example).

The transit fine print is where people get caught, so read it twice:

You must be travelling to a third country or region. Flying in from Tokyo and back to Tokyo is a round trip, not a transit, and doesn't qualify. Hong Kong, Macao, and Taiwan count as separate destinations for this rule.
You need a confirmed onward ticket with a seat, departing within the 240 hours. An open ticket or a vague plan is not enough.
The clock starts at 00:00 on the day after you enter, which quietly gives you most of an extra day.
Your passport should be valid for at least 3 months from entry.
You must stay inside the permitted areas attached to the scheme — it's not a roam-anywhere pass.
Tourism, business, and visiting friends and family are fine. Work, study, and news reporting are not — those still need a proper visa arranged in advance.

Neither door is a guarantee of entry; the officer at the desk decides. But walk in with the right door chosen and the paperwork matching it, and it's routine.

2. Money: your foreign card works — inside an app, with a fee line at ¥200

The single most useful change of the last two years: you no longer need a Chinese bank account to pay like a local. Both Alipay and WeChat Pay accept foreign-issued cards (Visa, Mastercard, and others depending on the app), and you can register with your own overseas phone number. No Chinese number, no local bank, no residency.

What the old guides get wrong, in order of how much it costs you:

Verify your passport in the app, before you fly. Verified visitors can spend up to the published ceilings for international cards — US$5,000 per transaction and US$50,000 per year. Unverified accounts run into much lower, sometimes unpredictable limits, usually at the worst moment (a hotel desk, a hospital registration). Do the scan at home.
Remember ¥200. Payments of ¥200 or less carry no platform fee on a foreign card. Above ¥200, a 3% fee on the full amount may apply. It's rarely worth splitting a bill over, but you should know why a ¥600 dinner costs more than ¥600.
Foreign cards can't do everything. Red packets and person-to-person transfers are off the table with an international card. If a landlord or a friend needs a transfer rather than a merchant payment, that's a different problem — solve it before it's urgent.
Set up both apps, then make one small test payment after you land. Alipay is usually the smoother first wallet for visitors; WeChat Pay matters because half the country's small merchants and hosts live inside WeChat. Belt and suspenders.
Physical cards and cash are the backup, not the plan. Big hotels and flagship stores may take your plastic directly; the noodle shop won't. Cash is still legal tender — carry a modest amount for the day everything else fails.

3. Connectivity: the routing matters more than the data

The question isn't "eSIM or SIM card." It's "where does this plan route my traffic?"

A travel eSIM routed outside the mainland is the default answer. Installed from a QR code before departure, activated on landing, your data exits via Hong Kong or Singapore — which means Google, WhatsApp, Instagram, and the rest of your normal stack just work, with no VPN and no registration queue at arrivals.
Read the fine print, because not every "China eSIM" routes that way. Some plans terminate locally and hand you a connection behind the firewall, gigabytes and all. If the product page won't say where traffic exits, assume the worse answer.
A local SIM is a deliberate choice, not a default. Buying one means in-person real-name registration with your passport (expect a face scan) at a carrier store, and the connection sits behind the firewall — Google and WhatsApp stay blocked unless you add a VPN. Its real value is the +86 number, which some local apps and verification flows still demand. Worth it for long stays; overhead for a 10-day trip.
"China doesn't do eSIM" is now outdated — and mostly irrelevant to you. In October 2025, China's three carriers received approval for domestic eSIM phone services. But those are provisioned under the same in-person real-name regime, built for resident subscribers. Interesting headline; doesn't change a visitor's playbook.
VPN rule, unchanged and still ignored: install it and test it before you leave home. Downloading one reliably from inside is the one step of this checklist that cannot be done late.

The pre-flight checklist, condensed

Confirm which visa-free door fits your passport and route — and re-check the current country list on the National Immigration Administration's site, because it changes.
If transiting: onward confirmed ticket to a third country/region, inside 240 hours.
Install Alipay and WeChat Pay; link your foreign card; complete passport verification.
Install a travel eSIM that routes outside the mainland; test it before departure.
Install and test your VPN at home, even if you expect the eSIM to make it redundant.
Carry one physical card and a small amount of cash as the failure-mode backup.

This post is the short version. My China Arrival Pack covers the same ground in working depth — the app stack beyond payments, high-speed rail booking, ride-hailing with a foreign number, and the arrival-day sequence step by step. If you spot anything here that's gone stale, tell me: corrections are the whole point of this series. toby.water@outlook.com

Stop Memorizing Words. Steal These 20 Chunks Instead.

Toby Water — Thu, 22 Oct 2026 17:40:46 +0000

Stop Memorizing Words. Steal These 20 Chunks Instead.

You already know thousands of English words. You can read articles, follow documentation, and pass exams. Then someone asks you a simple question in a call, and your brain starts assembling a sentence the way you'd assemble furniture from loose screws — one word at a time, hunting for the right verb, the right preposition, the right order. By the time the sentence exists, the conversation has moved on.

The problem isn't your vocabulary. It's your unit size.

Native speakers don't build sentences word by word. They pull chunks off the shelf: ready-made multi-word units like it depends on…, get back to you, just in case. Linguists call this formulaic language, and by most estimates a huge share of everyday speech is made of it. Your brain stores these the same way it stores a single word — as one item. One slot of memory, one act of retrieval, and out comes a whole grammatically-correct phrase with the prepositions already welded in.

That's why chunks beat word lists three ways:

Speed. One retrieval instead of five means you stop translating mid-sentence.
Free grammar. Nobody says "depends from" or "make sure to locking the door" after drilling the chunk — the correct structure rides along inside it.
Naturalness. Chunks are how natives actually phrase things. I'll get back to you sounds like a person. I will respond to you later sounds like a terms-of-service document.

There's a listening dividend, too. Once a chunk lives in your head as a single unit, you start hearing it as one. Fast native speech stops sounding like a blur of separate words and begins to separate into blocks you recognize — which is why chunk learners often notice their listening improving before their speaking does. You're no longer decoding; you're pattern-matching, exactly what your brain does in your first language.

So stop collecting words. Start stealing chunks. Here are 20 worth taking today — grouped by the moments you'll actually use them.

Making decisions

How about…? — the universal suggestion opener. "How about we ship the simple version first?"
make up your mind — decide, after deliberation. "Take your time — but I need you to make up your mind by Friday."
give it a try — attempt something without committing to success. "I've never used that framework, but I'll give it a try this weekend."
fair enough — "your point is reasonable; accepted." "You don't want to deploy on a Friday? Fair enough."
It depends on… — the honest answer to half of all questions. "It depends on how much traffic you're expecting."

Figuring things out and fixing problems

figure out — solve by thinking it through. "Give me twenty minutes and I'll figure out why the build is failing."
come up with — produce an idea or solution. "She came up with a workaround before the meeting even ended."
sort out — untangle a messy situation. "Let's sort out the access permissions before onboarding anyone else."
set up — get something ready to use. "I set up the staging environment so we can test safely."
end up — arrive somewhere you didn't plan to. "We skipped the tests and ended up debugging all night."

Everyday safety nets

just in case — as a precaution against a maybe. "Take an umbrella, just in case."
by accident — unintentionally; the opposite of on purpose. "I deleted the file by accident — good thing we had a backup."
make sure — verify; don't assume. "Make sure the database is backed up before you run the migration."
How come…? — a softer, more conversational why. "How come the deploy took so long yesterday?"
step by step — one stage at a time, in order. "Don't panic. Walk me through it step by step."

Work and staying connected

get back to you — reply later, once you know. "I don't have that number yet — I'll get back to you by noon."
keep in touch — stay in contact after things change. "Good luck with the new role. Let's keep in touch."
keep track of — monitor something over time so nothing slips. "I keep track of every invoice in one spreadsheet."
in advance — before the event, not at the last minute. "Book the train tickets in advance; they sell out on holidays."
You're on mute — the most-used sentence in remote work. "We can see you talking, but you're on mute."

How to actually learn them

Don't read this list ten times. Pick five chunks, say each one out loud in a sentence about your life — your project, your team, your weekend — and use them for real within 24 hours: in a standup, a message, a comment. A chunk you've said once in a real conversation is worth more than fifty you've highlighted. Five a day for four days and this whole list is yours — not memorized, available.

One more habit that pays off: when you catch a chunk in the wild — a podcast, a series, a colleague's message — write down their whole sentence, not just the chunk. Context is what later tells you when fair enough is friendly agreement and when it's a polite way to end an argument. Your notebook should be a collection of other people's real sentences, waiting for you to swap yourself into them.

These 20 are one shelf of the pantry. I keep a free sample of a different set — 20 more chunks with example sentences and printable cards — here (PDF), and the full collection of 100 everyday chunks, with Anki cards for spaced repetition, lives at Tobywater Digital. Steal the free ones first. That's the entire method.

You Don't Have a Vocabulary Problem. You Have a Chunk Problem.

Toby Water — Thu, 22 Oct 2026 13:44:23 +0000

Why "can read, can't speak" learners should stop collecting words and start collecting ready-made pieces of English.

For a long time, my English had a strange split personality. I could read articles, follow documentation, even write decent emails (slowly, with a dictionary tab open). But put me in an actual conversation and I froze — not because I didn't know the words, but because I was building every sentence from scratch, translating from Chinese in my head, one word at a time. By the time my sentence was ready, the conversation had moved on.

The fix, when it finally came, was almost insultingly simple: native speakers don't build most sentences from scratch either. They reuse chunks.

What is a chunk?

A chunk (linguists call them formulaic sequences or lexical chunks) is a multi-word unit that gets stored and retrieved as a single piece: I mean. By the way. The thing is… As soon as. A couple of. It looks like.

You already understand hundreds of these when you read. The problem is the gap between recognizing a chunk and producing it under time pressure. That gap doesn't close by reading more. It closes by deliberately practicing a finite list of chunks until they come out whole.

Why chunks beat word lists for speaking

Speed. A chunk is one retrieval instead of five. "The thing is" takes zero assembly; its word-by-word equivalent takes a full committee meeting in your head.
Naturalness. Chunks are how English actually sounds. Nobody says "I would like to supplement a point." Everybody says "By the way."
A finish line. "Learn more vocabulary" never ends. A list of 100 chunks ends in 20 days — and 100 well-chosen chunks cover a surprising share of everyday talk: opening, connecting, hedging, timing, quantity, impressions.

The method: 5 chunks a day, four steps each

Meaning + a scene. Read the explanation, then picture one moment in your own life where you'd say it.
Read the examples out loud, twice. I use 3 original example sentences per chunk. Second pass, cover the English and say it from the translation.
Steal one sentence. Change the names, places, details until it's true about you. Say your version out loud. This is the step everyone skips and the only one that really matters.
Review before new. Next day starts with yesterday's 5. On day 21, do a self-test: look at the meaning, produce the chunk in under 3 seconds, star the ones you miss, and only re-drill the stars.

Total: 15–20 minutes a day.

Start with these 20

Starting & connecting — I mean · by the way · to be honest · the thing is · for example
Quantity — a lot of · a bit of · a couple of · plenty of · too much
Time — on time · all the time · once in a while · sooner or later · as soon as
Impressions — it looks like · it sounds like · it feels like · kind of · more or less

If you can read all 20 but have never said half of them out loud, you have exactly the recognition/production gap this method is for.

Three traps

Collecting without speaking. A chunk you have only read is trivia. A chunk you have said 10 times is a tool.
Doing 20 a day to "finish faster." You'll retain four. Five a day, reviewed, beats twenty skimmed.
Never personalizing the examples. Generic sentences stay generic. The sentence about your boss, your rent, your flight is the one you'll actually reach for.

Get the free sample

I turned my own list into a study pack: 100 everyday chunks, each with a Chinese explanation, 3 original example sentences with translations, a usage tip, a 21-day plan, an Anki deck, and printable cards. The first 20 chunks — the ones listed above, fully worked out in an 11-page PDF — are a free sample.

You can get the sample and see the full version here: https://tobywater.pages.dev

If you try the sample, I want one piece of feedback: are 3 example sentences per chunk enough, or should v2 use short dialogues instead? That's a real decision I'm making, and learner answers beat my guessing.