Platforms and trader tools – who owns the chart, the alert and the execution of an order
A trading platform, a broker and a mobile app can look like a single system, yet they have different responsibilities. This section explains where chart prices come from, where alerts run and what confirms an order has been executed. It also covers criteria for choosing tools, testing their behaviour and combining them into a consistent workflow. Ten articles take you from these basic responsibilities to automation and a documented process of your own.

- Four layers of responsibility instead of one product name. The same property – an available order type, the source of quotes, the way an alert is evaluated, the position mode – may be a feature of the terminal software, a condition of your broker account, or a limitation of the app version, including a subscription-plan limit, which belongs to the app layer and not to a separate fifth layer. The fourth layer is your trading plan: your own rules, which no vendor sells. Before you ask which platform is better, work out which layer the property you actually care about belongs to.
- The mechanism outlives the state of the product. The statements "an alert is not an order", "a chart is not the execution price" and "a historical test is not the future" hold regardless of version. Feature names, subscription-plan limits, supported languages and price lists change between releases, so in this section they appear as a state on a given date with a pointer to the vendor's documentation, never as a permanent truth.
- Automation moves operational risk, it does not remove it. Code, hosting and a language model take over part of the manual work and add failure modes of their own: a lost connection, a duplicated order, a result nobody can reproduce, a model answer that sounds credible and is false. That is why the section opens with the question of where a mechanism physically runs, and closes with the procedure that ties everything together.
1. Four layers of responsibility
The question “which platform is best” needs a more precise starting point: which feature do you want to compare? Four responsibilities help organise the answer: platform software, broker conditions, the app version and your own working rules. This is an educational framework, rather than four physically separate systems: a terminal is an application, and one product can cover several functions. The distinction still tells you where to find the specification and whom to ask.
- The terminal – the platform software. What you have installed or open in a browser: charts, watchlists, the order ticket, the scripting environment, the tester. The terminal sends an instruction and shows the answer. It is not a party to the transaction.
- The broker – the environment your account sits in. Instrument specification, quoting hours, the source of quotes and its provider, server time, execution policy, account type and the list of available symbols. This is where it is decided whether an order is accepted and at what price.
- The app – the access channel and its version. The desktop version, the browser version and the phone app carry the same brand, but need not have the same set of tools, the same order types or the same ability to run programs. The same name is no proof of the same function. Inside this layer, a separate sub-owner is often the subscription plan, which switches parts of the functionality on and off independently of the terminal and the broker – that is still the app layer, not a fifth one.
- The trading plan – your own rules. When a position may be opened, which conditions must be met first, what happens if a tool fails and where the result is recorded. An application can provide a template or a ready-made rule set, but you still have to assess its suitability, define limits and take responsibility for its use. Installing a tool does not complete that work for you.
The test is simple and works on any sentence you want to write or read about your tool: check whether it can be finished without naming an owner. "This platform has a tighter spread" fails the test, because pricing conditions are a property of the account, not of a program window. "This platform gives me the order type I am looking for" passes conditionally – provided you check that the type is also supported by the server your account sits on.
After "what does it do" and "whose is it" comes a third question: where is it evaluated. The answer decides what happens when you shut the laptop, lose signal or quit the app. Three examples from vendor documentation – each with a check date, because this is the reference layer and it can change together with the product version:
- An alert condition is sometimes evaluated server-side. TradingView's documentation describes alerts as created and triggered on the platform's servers and states that closing the page does not stop them from running[1]. That does not mean an alert places an order – it means only that the condition is evaluated without your computer taking part; the whole alert mechanism is taken apart in article 11.2. [STATE AS OF: 2026-08-22]
- An algorithm does not always need a running terminal. The cTrader Algo documentation states that the cTrader application does not have to stay open or connected to the internet for a bot launched in the cloud to keep working[2]. So neither "every algorithm requires a switched-on computer" nor "every algorithm runs without one" should be taken as a universal rule. Runtime environments are compared in article 11.1, while where a program actually runs and what happens after a lost connection is settled by article 11.5. [STATE AS OF: 2026-08-22]
- The programming layer is sometimes a storey above the terminal, not a link to the broker. The official documentation of the MetaTrader5 package for Python describes it as a tool that fetches data through inter-process communication directly from the MetaTrader 5 terminal[3]. In that arrangement the script is another floor on top of the terminal, not a replacement for the broker's interface; the whole thread is carried to its conclusion by article 11.8. [STATE AS OF: 2026-08-22]
Each of those three facts is true as of the check date and each can change with a version. What lasts is the conclusion, not the detail: before you build a procedure on any feature, establish where that feature is executed and what happens to it when your device drops off the network. The same function under the same name can have two different execution locations in two product versions.
A chart is not proof of the execution price. A candle may be built from the bid, the ask or the last traded price, depending on the instrument and the data provider, and two charts of the same symbol from two sources need not look identical. The sentence "price never touched the level, so the order could not have been activated" only becomes a claim once you establish which price series decides the event. Before that, it is a guess. Where the price on the chart comes from and why two data sources can show the same pair differently – article 11.2.
An alert is not an order. Between a met condition and a position on the account stands a chain: condition → alert → event delivery → intermediary application → broker interface → validation → order → server response. Every transition can be delayed or broken, and the whole is no more reliable than its weakest link. A notification on your phone says the condition was met – it does not say the trade is executable right now at the level that triggered that condition. Configuring the alert itself is covered by article 11.2, and the bridge between an event and an order by article 11.5.
2. What this section does not do
This is a tooling section: technical and procedural. It does not teach you how to make money and it is not a ranking – not out of politeness, but because those topics have their own owners in this course, and developing them here would mean two different versions of the same knowledge.
- It does not pick a broker and does not rate execution models. Who the counterparty is, how to read an offer and where the conflict of interest arises – the section on choosing a broker.
- It does not build strategies and promises no edge. Configuring a tool is not an edge; where an edge comes from and how it is tested – the section on strategies.
- It does not replace risk methodology. We show where to enter a limit and how to check a calculator's inputs, but position size, risk budget and portfolio limits belong to the risk management section.
- It is not a psychology course. The journal and the checklist appear here as a tool and a procedure; working with your own behaviour, discipline and the compulsion to check the account – the trader psychology section.
- It does not classify recommendations and does not discuss legislation. What meets the criteria of an investment recommendation and how client funds are protected – the section on regulation and safety.
- It does not fix price lists, subscription-plan limits or version numbers. Wherever such information is needed for a calculation or a decision, it appears with a check date and a pointer to the vendor's documentation – never as a constant.
Nor will you find here a sentence saying that some tool is the best. You will find a criterion: what exactly you are comparing, under which assumption, and whose the property you are comparing is. A comparison without those three elements is not analysis, it is advertising material with a different cover.

3. Section map – ten articles, one logic
The articles run from the layers of responsibility through tools and automation to validation and synthesis. You can read them in any order, but the procedure in article 11.10 fuses elements you have to meet separately first.
11.1. MT4, MT5 and cTrader – industry standard and alternatives
How three terminals differ, and what cannot be compared without naming a broker and a version. The article separates the terminal from the broker, describes MQL4 and MQL5 as two separate environments, the current state of cTrader Algo[4], the availability of desktop, browser and phone versions, the tester at overview level, migrating indicators and robots as a compatibility problem, and server time and the data source. Instead of a verdict you get a comparison table with no "better" and "worse" ratings, a second table assigning an owner of the decision to every feature, and three user profiles – manual, running an existing MQL4 robot, and building new automation – each with its own selection criterion.
11.2. TradingView – charts, alerts and analysis workflow
Where the quotes come from, why the chart can differ from your broker's, and what an alert really does. The article covers the symbol and the data source, intervals, indicators, Pine Script at the level a user needs, the difference between an indicator and a strategy, the strategy tester as a simulation, price alerts and strategy alerts, Bar Replay, broker integrations as a broker-dependent feature, and repainting understood as a divergence between historical and real-time behaviour[5]. Two mandatory elements: a breakdown of a case in which the signal on history and the real-time alert behaved differently, and a comparison of the same pair from two data sources – without treating either of them as the only true one.
11.3. Economic calendars, scanners and calculators
Three tools that reduce manual work, none of which is a standalone trading signal. For the calendar: the current reading, consensus, previous and revised figures, publication time and time zone, country and currency, the importance classification treated as a provider label, delays, revisions and the primary source of the data. For the scanner: the filter, the condition, the instrument universe, the refresh frequency, the difference between a ranking and a signal, and the trap of screening hundreds of conditions in order to pick the one that looked best in the past. For the calculator, a box titled "Before you trust a calculator, check seven fields": symbol, contract size, account currency, conversion rate, required margin, commission and swap, and the unit of the result.
11.4. Mobile trading – managing positions from your phone
What can safely be done from a phone and what must not be assumed. The article describes the differences from the desktop and browser versions, notifications, loss of connectivity, a forced app shutdown, the split between what runs on the broker's side and what requires a running app, system updates, screen lock and multi-factor authentication, apps impersonating platforms, and the risk of a shared device. At its core are three failure states: the phone loses internet after an order has been sent; the app shows an order as pending and it is unclear whether the server accepted it; the phone is out of reach and a position is open. For each state the article gives a verification procedure instead of guesswork.
11.5. Trading automation – EAs, scripts and VPS
Where a program actually runs and what happens when it stops. The article separates the robot, the indicator and the script, MQL4 from MQL5, and covers compilation, launch, permissions for automated trading, logs, identifying your own orders and positions, local execution versus hosting and the cloud, restart and state recovery, a lost connection, a duplicated order, the kill switch, spread, volume and trading-hour limits, and a trial on a demo account before the live one. Separately: the difference between an ordinary server and hosting integrated with the platform, a list of typical production failures, and an annual infrastructure cost calculation worked on an explicit example, with prices marked with a check date.
11.6. Trading journal and performance analysis
The journal as a database of the process, not an album of results. The article covers importing history versus recording by hand, the minimum set of fields, tagging setups and errors, planned entry and exit versus actual, costs, compliance with the plan and process error, and from the metrics it defines win rate, average win and average loss, expectancy, profit factor, drawdown and the R multiple, with MAE and MFE as an extension. A mandatory element shows a single strategy with a positive result in which part of the trades broke the plan and a large share of the profit came from one event. Separately: the difference between an investor password and a master password, the recommendation to grant read-only access, and the privacy consequences of a public profile.
11.7. Demo accounts, backtesting and forward testing
How to check an idea before committing real capital, and how to tell validation apart from a curve fitted to history. The article separates the demo account, paper trading, the historical test, optimisation, the forward test and the live account, and covers data and the tick-generation model, spread, commission, swap, slippage and latency, gaps, look-ahead and data leakage, overfitting, the number of trades, parameter stability, the split into fitting and control samples – in-sample and out-of-sample – walk-forward analysis and the Monte Carlo test as an extension. At its core sit a documented validation sequence, a test reproducibility protocol[6], and a cost test showing, on an explicitly labelled example, that changing the cost by a fraction of a pip can flip the result of a strategy with a low average profit per trade. It is rounded off by an example of a result that looks superb and is unrealistic in four places at once: it enters at a price it could not have known before the decision, pays neither spread nor commission, uses the single best parameter set out of thousands tested, and is judged over the very period it was fitted to.
11.8. Python in trading – data analysis and automation
What Python is good for in a trader's hands and how to get from a notebook to a tool you can entrust with repetitive work. The article covers the virtual environment and dependencies, notebook versus script, the structure of candle and tick data, time and time zones, fetching and cleaning data, calculations, export, integration with the terminal, an overview of REST and WebSocket interfaces, error handling, logging, keeping secrets outside the code, and tests. The minimal example is deliberately non-trading: read a file, parse the time in UTC, compute the change, detect gaps, save the result. A box titled "A research notebook ≠ an execution program" lists what the prototype lacks – configuration, secrets, logs, tests, retries, timeouts, state control, monitoring, a kill switch and deployment.
11.9. AI and LLMs in trading – applications, limits and risks
Where a language model has real value, where it is risky, and how to build an arrangement in which it makes no uncontrolled financial decisions. The article covers the difference from classical machine learning, the applications that hold up – analysing documentation, extracting data from announcements, tidying the journal, generating tests – and also hallucinations, knowledge cut-offs, non-deterministic answers, prompt injection, data protection, versioning of the model and the prompt, testing model quality on a fixed set of examples – one you do not tune the prompt on and on which you check for regressions after a model change – restricting permissions, and the audit log. The mandatory architecture runs from data sources through a normalisation layer and the model to a structured proposal, a rule validator, risk limits, optional human approval and only then the execution layer, the broker and the audit log – with a clear statement of which links must stay deterministic. Separately: the anatomy of an "AI bot" offer as a repeatable fraud pattern[7], and an explanation of why the article publishes no signal-generating prompts.
11.10. Building your trading workflow – a complete system
How to join everything into a single track, without chaos and without two parallel sources of truth. The article describes one source of truth for the plan and the journal, the watchlist, the calendar, alerts, pre-session preparation, the trade plan, the execution checklist, monitoring, recording, review, archiving material and versioning changes to the process, plus two modes of working: a light manual one and an elaborate systematic one. It closes with a five-point emergency procedure – how to check the account state through another channel, how to stop automation, how to verify open positions and orders, how to record the incident, and how to resume work only after the state has been confirmed – and with a test for spotting a tool that adds nothing.
4. How to read this section
The sequence 11.1–11.2 establishes the foundation: what belongs to the terminal and what belongs to the data provider. Articles 11.3–11.4 cover supporting tools and access from a phone. Articles 11.5–11.6 cover delegating work to software and measuring your own process. Article 11.7 validates an idea before capital is committed. Articles 11.8–11.9 cover Python and language models. Article 11.10 brings the elements together into one procedure.
If you already have experience and are looking for the answer to a specific question:
- Considering a change of terminal? → 11.1: comparison criteria and a table assigning an owner of the decision to every feature.
- An alert behaved differently from what the history promised? → 11.2: eight things to check in order, from the open candle to the platform's current limitations.
- The result of a historical test diverges from the live account? → 11.7: data, costs, the split into fitting and control samples, execution realism.
- Want to move repetitive work into code? → 11.5 on the terminal side, 11.8 on the Python side.
- Have the tools but no process? → 11.10 – and go back to the elements that process fuses together.
- Want to separate the real uses of language models from the marketing? → 11.9, including the anatomy of an "AI bot" offer.
The articles assume familiarity with concepts from earlier sections: expectancy, the reward-to-risk ratio, position size, the protective order, drawdown, spread, slippage and swap points. If you know any of them by name only, start with the risk management section; written procedures and work on your own behaviour are described in the trader psychology section.
5. What comes next
After working through this section you should be able to:
- say, for any property of a tool, whether it belongs to the terminal, to the broker, to the app version – including the subscription plan – or to your own trading plan;
- say where the price shown on your chart comes from and why the same pair may look different to somebody else;
- configure an alert and know what its firing actually means and what it does not;
- use a calendar, a scanner and a calculator without treating them as a signal generator;
- determine where your program is physically executed and what happens to it after a lost connection;
- judge when you need a separate runtime environment and what it does not solve;
- keep a journal that lets you analyse the process, not only the result;
- tell a demo account, paper trading, a historical test, a forward test and a live account apart, and recognise a test built on unrealistic assumptions;
- use a language model as a supporting tool, with deterministic control before every step that touches money, and recognise an "AI bot" offer that is a fraud;
- build a repeatable working track: data → analysis → plan → execution → journal → review.
The platforms and tools section closes the trader's operational infrastructure. The next sections – Choosing a broker and Regulations and safety – answer the question of who you hold your account with and under what legal framework.
We start with the first layer – with what belongs to the terminal itself, and what is a property of the account that terminal works on.
FAQ – platforms and tools
Which platform should I start with?
Can a charting tool replace a trading terminal?
Do I need a separate environment for my program to run non-stop?
The historical test looks good, the live account looks different. Where should I start looking for the cause?
Can a language model run my account for me?
Sources and bibliography
Product facts were verified on 22 August 2026 in vendor documentation. The reference layer – feature names, subscription-plan limits, supported languages, integration availability and price lists – changes with versions and is not fixed on this page; the durable layer, the description of the mechanism, stays valid for longer. Before deciding, check the current state in the primary sources listed below.
- TradingView Help Center, "About TradingView alerts" – tradingview.com/support/solutions/43000548327. Vendor documentation: alerts are created and triggered on the platform's servers, and closing the page does not stop them from running. Checked 2026-08-22.
- Spotware Systems, "cTrader Algo – Cloud features" – help.ctrader.com/ctrader-algo/documentation/cloud-features. Source of the statement that the cTrader application does not have to stay open or connected to the internet for a bot launched in the cloud to keep working. Checked 2026-08-22.
- MetaQuotes Software Corp., "MetaTrader module for integration with Python" – mql5.com/en/docs/python_metatrader5. The documentation describes the package as a tool that fetches data through inter-process communication directly from the MetaTrader 5 terminal. Checked 2026-08-22.
- Spotware Systems, "cTrader Algo" – help.ctrader.com/ctrader-algo. Checkpoint for the current scope of the cTrader algorithmic environment, including supported languages. Checked 2026-08-22.
- TradingView, "Repainting" – Pine Script User Manual: tradingview.com/pine-script-docs/concepts/repainting. Definition of repainting as a divergence between a script's behaviour on historical data and in real time. Checked 2026-08-22.
- MetaQuotes Software Corp., "Strategy Tester" – MetaTrader 5 help: metatrader5.com/en/terminal/help/algotrading/testing. Checkpoint for the testing procedure described in article 11.7: the current data modes, execution settings and forward-test period should be checked in this documentation before being relied on. Checked 2026-08-22.
- Polish Financial Supervision Authority (KNF), "KNF public warning list" – knf.gov.pl/dla_konsumenta/ostrzezenia_publiczne. The register referred to by article 11.9 when describing "AI bot" offers; before relying on a specific announcement, check its current content. Checked 2026-08-22.