Record AR-STD-010 · Risk disclosure
Everything on this site is a model.
Your account is not.
This is the most important page in the facility. It sets out what trading risk actually is, what the preview engine does and — at greater length — what it does not do, why simulated results flatter themselves in ways live trading never will, and the specific hazards of the leveraged instruments most readers of this site are using. Read it before you take anything else here seriously.
01The nature of trading riskAR-STD-010·01
Trading involves a substantial
risk of loss.
That sentence appears on every trading site, is read by nobody, and is nonetheless true. It is worth stating what it actually means rather than treating it as a formality.
Losses are the normal condition of the activity, not a malfunction of it. A framework with a genuine positive expectancy still loses on the majority of individual trades in several of the families described on this site, and still spends long stretches below its previous equity peak. A trader who has not internalised that will interpret ordinary variance as failure and abandon a working system at its worst moment — which is, statistically, the most common way that a positive expectancy is converted into a negative outcome.
Losses can exceed expectations, and in leveraged products they can exceed deposits. A stop is an instruction, not a guarantee. Markets gap. Liquidity withdraws precisely when everybody wants it. The distribution of outcomes has a left tail that is fatter than most models, including the one on this site, allow for.
Most participants lose. Where retail loss statistics are published under regulatory requirement, the proportion of retail accounts losing money on leveraged products is consistently reported at a large majority. Nothing on this site changes that base rate, and no analytical framework should be read as evidence that you will be an exception.
You should not trade with money you cannot afford to lose entirely, and you should not treat capital allocated to trading as though it were savings with a different label.
The single most important sentence on this site
No content here is a recommendation to trade, a prediction of any market, or evidence that any approach will be profitable. Every figure, chart, score and number is produced by a documented model from stated assumptions, and models are not markets.
Four categories of risk this site does not remove
02No advice, no signals, no brokerageAR-STD-010·02
What this facility
does not provide.
Each of the following is a categorical statement, not a limitation of the current phase. None of these services is offered now and none is planned.
If you are looking for advice
Seek it from a professional who is authorised in your jurisdiction to give it, who is obliged to consider your circumstances, and who carries the regulatory responsibility that goes with the role. This facility is none of those things and cannot substitute for one.
03The status of every number hereAR-SIM-047
No number on this site
was measured from a market.
Every figure, score, equity curve, distribution, drawdown envelope, expectancy reading and probability shown anywhere in this facility is produced by one deterministic model — referred to throughout as the preview engine — from parameters that you or the page have set. It is not a record of trading. It is not a backtest. It is not derived from price history, and there is no price history anywhere in the system.
The engine takes 159 named controls and a strategy archetype and derives a working stop, a friction charge expressed against that stop, an outcome ladder, a trade frequency, an effective concurrency after correlation and a drawdown envelope. Those feed a seeded path simulator that groups trades into concurrency blocks and, where the governance layer is enabled, reviews capital authority at block boundaries.
What the engine encodes are relationships that are argued for in the research records: that friction scales against the working stop rather than against the trade, that partial-taking trades tail participation for capture efficiency, that correlation converts concurrency into concentration, that sequence alone changes maximum drawdown. Those relationships may be right. They are still assertions about structure, not observations of markets, and the magnitudes attached to them are illustrative.
Consequently, a configuration that grades well in this laboratory has demonstrated nothing whatsoever about its behaviour in live trading. It has demonstrated that it is internally coherent under a set of assumptions someone wrote down. That is a genuinely useful thing to know and it is a much smaller thing than it can appear to be on a well-rendered chart.
Standard disclosure, verbatim
“Simulated demonstration data. This visualization illustrates proposed product behaviour and does not represent historical, live, or guaranteed trading performance.”
Monte Carlo outputs additionally carry: “Demonstration simulation based on selected assumptions. Not a forecast.” Parameter-driven outputs carry: “Interactive demonstration — not a historical backtest.”
These markers appear inside the header of the figure they describe, not in a page footnote. A disclosure that requires scrolling is a disclosure designed not to be read.
Determinism is not accuracy
The engine returns the same output for the same input on every device, forever. That property makes claims checkable and arguments possible. It says nothing at all about whether the model resembles a market — a wrong model can be perfectly reproducible.
04Model boundaries · limitationsAR-SIM-047·L
Nine things the engine
does not contain.
The list of what a model omits is more informative than the list of what it includes, because the omissions are where the money is lost. Each of the following is absent from every number on this site, and each of them is a real cost in live trading.
| Absent factor | What the model assumes instead | Practical consequence |
|---|---|---|
| Real market data | No price series exists in the system. Outcomes are generated from a modelled distribution shaped by the control state. | Nothing here has been tested against what any instrument actually did, in any period. |
| Real fills | Entry and exit are assumed to occur at the modelled level, subject only to an aggregate friction charge. | Partial fills, requotes, rejected orders and the difference between an intended and an achieved price are all absent. |
| Liquidity | Unlimited depth at the modelled price. Position size never affects the outcome. | Size-dependent slippage, thin sessions and the withdrawal of the book under stress do not appear anywhere. |
| Gaps beyond a modelled shock term | A shock-loss term costs materially more than one unit of risk at a stated frequency. | Weekend gaps, event repricing and limit moves are compressed into a single parameter with a chosen magnitude. Real gap distributions have no such ceiling. |
| Correlation regime shifts | A single correlation coefficient applies across the run, converting concurrency into effective independent bets. | The empirical behaviour — correlations converging toward one in a shock, exactly when diversification is being relied upon — is not simulated as a dynamic process. |
| Counterparty risk | The counterparty always performs. Deposits are always available. | Broker insolvency, segregation failure, withdrawal restriction and negative-balance events are outside the model entirely. |
| Execution failure | Every intended action occurs. Every stop is placed and honoured as instructed. | Platform outages, connectivity loss, mis-keyed sizes, an order not placed because attention was elsewhere — none are represented. |
| Psychological failure | The operator follows the framework exactly, indefinitely, including through its worst modelled drawdown. | This is the least realistic assumption in the model, and the deviation it ignores is the most common cause of real losses. |
| Regulatory and tax effects | No jurisdiction, no leverage cap, no tax, no financing beyond a swap term where enabled. | Net outcomes in reality differ from modelled ones by an amount that varies with where you are and how you are taxed. |
The direction of the error
Every omission above makes modelled results better than reality would be, not worse. There is no absent factor that would flatter a live account relative to this model.
Why model at all, then
Because relative comparisons survive omissions that absolute ones do not. “This costs more than that” is more robust than “this returns X”, and the instruments here are built for the first kind of question.
What must not be inferred
Any expected return, any probability of profit, any drawdown you should expect, any position size you should use, or any conclusion that a framework is fit for live capital.
05Past and simulated performanceAR-STD-010·05
Simulated results have
a specific, known bias.
Past performance is not indicative of future results. The familiar warning is about historical records, and it applies to any record of any strategy anywhere, including ones audited to a high standard. A profitable history is consistent with a genuine edge, and it is equally consistent with a favourable sample of a system that has none.
Simulated and hypothetical performance carries an additional and different problem, and it is worth stating precisely because it is the standard point that everybody quotes and few people unpack. Simulated results are prepared with the benefit of hindsight. The person constructing the simulation already knows how the period they are modelling behaved. They know which volatility regime prevailed, which parameters would have been comfortable, which instruments were worth including. Every one of those decisions is made in possession of information that was not available at the time, and none of them feels like cheating while it is being made.
Hypothetical performance also does not involve financial risk, and no simulated record can account for the effect of risk on decision-making. A drawdown in a chart is a shape. The same drawdown in an account is a series of decisions taken by someone who is not sleeping well, has an opinion about why it is happening, and is under pressure from people who are asking. There are frequent sharp differences between simulated results and actual results subsequently achieved by any particular trading programme, and they run in one direction.
This site does not present a historical simulation at all — which removes the hindsight problem by removing the history, and replaces it with a different problem: the relationships in the model were also chosen by someone, and they were chosen because they seemed right.
Five ways a good record misleads
No performance is claimed anywhere
This facility publishes no track record, no returns, no account statements and no claim that any framework described here has been profitable for anyone. If you encounter content that appears to claim otherwise, it is either not from this site or it is a defect that should be reported.
06Leverage and marginAR-STD-010·06
Leverage does not increase edge.
It increases everything else.
The instruments most readers of this site use are leveraged, and leverage interacts with every analytical concept documented here in ways that are not intuitive.
What leverage actually does
It multiplies exposure relative to deposited capital. A small adverse move against a large notional position produces a loss that is large relative to the account, which is the entire point and the entire hazard. A framework with a positive expectancy per unit of risk does not become more profitable when risk per trade is increased; it becomes more volatile, and past a threshold the volatility drag reduces the compounded outcome even though the arithmetic expectancy is unchanged.
That threshold is lower than most traders assume, and it moves down as correlation between concurrent positions rises. This is one of the few places where the analysis on this site produces an unambiguous practical conclusion: the growth-optimal risk fraction is not the practical one, and the practical one is well below it.
Margin close-out
Margin is a performance bond, not a cost. If account equity falls below the required level, positions may be closed by the provider automatically, at a time and price you do not choose, potentially crystallising a loss that a wider stop would have survived. Close-out is a function of account equity and provider rules, not of whether your analysis was correct.
Losses beyond deposit
In some jurisdictions and with some providers, losses on leveraged products can exceed the amount deposited, leaving a debt. Where negative-balance protection applies it limits this, but its availability depends on your jurisdiction, your provider and your client classification. It is your responsibility to know which applies to you; this site cannot tell you, because it does not know where you are.
Financing and holding cost
Leveraged positions typically incur an overnight financing charge that accrues for as long as the position is held. For short-horizon strategies this is negligible. For position-horizon frameworks — including several of the trend families described in the library — financing can consume a material share of expectancy over the holding period, and it is a cost that grows with time rather than with activity.
Where the model is silent
The preview engine models a swap term where enabled and charges friction against the working stop. It does not model margin requirements, close-out mechanics, leverage caps, or the possibility of a balance below zero. Those omissions matter most in exactly the scenarios where they would matter most.
07Instrument-specific riskAR-STD-010·07
Foreign exchange
and contracts for difference.
Much of the material on this site uses FX as its working example, because the framework families described are commonly deployed there. FX and CFDs carry risks that differ in kind, not only in degree, from those of cash instruments.
Availability and suitability vary by jurisdiction
CFDs are restricted or prohibited for retail clients in some jurisdictions, and leverage limits, negative-balance protection and marketing rules differ substantially between them. Whether these instruments are available or appropriate for you is a question for your jurisdiction's rules and a qualified professional, not for this site.
08Drawdown and the holding of itAR-STD-010·08
The number is not
the experience.
A maximum drawdown figure describes a depth. It says nothing about the two properties that determine whether a person survives it: how long it lasted, and what proportion of the total period was spent below the previous peak. A twenty per cent decline recovered in six weeks and a twenty per cent decline that grinds sideways for fourteen months are the same number and entirely different events.
The analytical treatment on this site decomposes drawdown into depth, duration, time under water and recovery efficiency for precisely this reason. What that decomposition cannot do is convey what it is like to be inside one. In a live account, an extended decline arrives with a running commentary: an explanation for why this time is different, evidence that the edge has decayed, the knowledge that a change of approach would stop the pain immediately, and the observation that waiting has not worked for some time now.
Systems are abandoned at the extreme, not at the mean. The failure mode is rarely a catastrophic loss; it is a rational-seeming decision, taken at the worst possible point in the distribution, to stop doing the thing that was working. Any modelled result that assumes indefinite adherence — and every result on this site does — is describing an operator who does not exist.
The practical implication is that a framework should be selected for the drawdown you can actually hold, not the one the model says is optimal. Those are different questions and only one of them has a mathematical answer.
The drawdown you will experience is deeper than the one you modelled
Not because the model is careless, but because the historical or modelled maximum is a single realisation. Resequencing the same outcomes produces materially worse paths with ordinary frequency, and the future contains sequences that have not occurred yet.
Four measures, one word
09Overfitting and sample adequacyAR-STD-010·09
The most likely explanation
for a good result is luck.
This is not cynicism. It is a consequence of how many configurations are examined and how few trades most conclusions rest on, and it applies to results produced in this laboratory as much as to results produced anywhere else.
Search inflates apparent quality
If you evaluate enough parameter combinations, some will look excellent by chance alone. That is not a flaw in the search; it is arithmetic. The more degrees of freedom a framework has — and the seventeen-component decomposition used here makes it clear how many that is — the more thoroughly a noisy sample can be fitted, and the more confident the resulting picture appears.
The parameter neighbourhood is the practical defence. A setting that performs well while its neighbours perform badly is an artefact; a setting that performs adequately across a broad plateau may be real. This is why the parameter laboratory shows the surrounding envelope rather than reporting an optimum, and why it does not have a button that finds one for you.
Interactive tools make it easier
A responsive interface that recomputes on every drag is an efficient overfitting machine. Moving controls until the score is satisfying is a search, and it produces the same inflated confidence as an automated one — with the additional problem that you will remember it as insight.
Sample size, honestly
The confidence interval around an expectancy estimate is wide, and it narrows with the square root of the trade count. In practical terms: a few hundred trades is enough to distinguish a large edge from nothing, and nowhere near enough to distinguish a small edge from nothing, or to rank two similar configurations against each other.
Most retail conclusions fail this condition alone, before any question of method arises. A trader comparing two variants over sixty trades each is reading noise, and the fact that the difference is visible on a chart does not make it evidence.
Out-of-sample separation is a minimum
Fixing parameters on one partition and evaluating on another is the least you can do, not the most. It is defeated by iterating: if the out-of-sample result feeds back into the parameter choice, the separation is gone and only the ritual remains. The eight validation conditions used here exist because a single defence is not enough.
Read the validation protocol →10The limits of Monte CarloAR-STD-010·10
A distribution of
the assumptions you supplied.
Monte Carlo analysis is the most powerful instrument on this site and the most frequently misread. It replaces a single equity curve with thousands of them, which is a genuine advance over reasoning from one realisation. What it does not do is tell you what will happen.
It cannot produce information that was not in its inputs. Resequencing a set of outcomes explores the effect of order; it does not explore outcomes that are not in the set. If your sample contains no loss larger than three units of risk, no resequencing will ever generate one, and the tail of the resulting distribution is therefore an artefact of the sample rather than a property of the strategy.
Independence is assumed and is usually false. Standard resequencing treats trades as exchangeable. Real trade outcomes cluster: losses arrive together because the regime that produces them persists, and a shuffle destroys exactly the structure that makes real drawdowns deep. A Monte Carlo drawdown distribution built on an independence assumption is systematically optimistic about the left tail.
The median path is not a forecast. It is the middle of a distribution of simulated outcomes under fixed assumptions. Half the simulated paths are worse. The percentile bands describe the model's dispersion, not the market's, and a completion probability is a statement about the model rather than about your account.
Stationarity is assumed throughout. Every path is generated from the same distribution. Markets change; edges decay; the volatility regime you calibrated in ends. Nothing in a Monte Carlo study of a fixed distribution can anticipate the distribution changing, and that is the failure mode that actually ends most strategies.
What Monte Carlo is genuinely good for
Demonstrating that sequence alone changes survival. Showing how wide the range of outcomes consistent with your own assumptions actually is. Making the point that a single equity curve is one draw and not a result. Those are real and they do not require the output to be predictive.
What it must not be used for
Sizing decisions taken from a completion probability. Confidence that a modelled worst case is a real worst case. Any statement of the form “there is an N per cent chance my account will…” — the model has no access to that quantity.
Standard marker
Monte Carlo outputs on this site carry: “Demonstration simulation based on selected assumptions. Not a forecast.” The phrasing is deliberate — the assumptions are selected, by you or by the page default, and the output inherits every one of them.
Open the Monte Carlo core →11What governance can and cannot doAR-STD-010·11
A governance layer reduces risk.
It does not remove it.
The MARS material on this site describes capital authority as a layer above signal quality. It is important to be exact about what such a layer achieves, because the temptation to treat a risk-management structure as a safety guarantee is strong and it is the more dangerous error.
What it can do
What it cannot do
The cost side, stated plainly
Governance is not free. Compressing risk during a decline reduces participation in the recovery that follows it, and in favourable conditions a governed account will underperform an ungoverned one running the same strategy. That trade — giving up upside in good states to survive bad ones — is the whole proposition, and anyone presenting a risk framework as costless is not describing it accurately.
12Regulatory status and jurisdictionAR-STD-010·12
Not regulated,
and not pretending to be.
The AlphaRail Foundry Laboratory is not authorised or regulated by any financial services authority in any jurisdiction. It holds no licence, no registration and no permission to carry on regulated activity, and it carries on none — no advice, no arranging, no dealing, no managing, no custody, no payment services.
It follows that none of the protections associated with dealing with a regulated firm apply to your use of this site. There is no compensation scheme, no ombudsman, no complaints procedure with statutory force, and no regulatory supervision of the content. Those protections exist for good reasons; their absence here is a reason to treat this material as what it says it is — educational and analytical — rather than as guidance you can rely on.
The published material is general and impersonal. It is not directed at any individual, does not take account of anyone's circumstances, and creates no advisory, fiduciary or client relationship of any kind between the Foundry and any reader.
Jurisdictional notice
13Independent professional adviceAR-STD-010·13
Take advice from someone
who is accountable for it.
Before trading any leveraged product, and before acting on any concept described in this facility, you should seek independent advice from a professional authorised in your jurisdiction — and you should do so on the basis of your own circumstances, which this site knows nothing about.
01
What to establish first
Whether these instruments are appropriate for you at all, given your objectives, your experience, your capacity to bear loss and the length of time you can leave capital at risk. That is a suitability question and it precedes every analytical question on this site.
02
What only you can decide
How much you can lose without changing your life, and how deep a drawdown you can hold without abandoning the framework. Both are personal limits and neither has a mathematical answer that a model can supply.
03
What to assume
That you will lose money at some point, that the losing period will last longer than you expect, and that no amount of preparation on this site or anywhere else converts trading into an activity with a guaranteed outcome.
In summary
Trading involves substantial risk of loss and is not suitable for everyone. All content in this facility is educational and general in nature, is produced by a documented model rather than measured from any market, and constitutes no advice, recommendation, signal or guarantee. Past performance and simulated performance are not indicative of future results. You are solely responsible for your own decisions and for any consequence of them.
Read alongside this.
This disclosure is one of four documents that together set out what the Foundry claims, what it collects, what it permits and where it falls short. None of them is boilerplate and each is specific to this site.
All datasets simulated · no advice given · no performance claimed · no regulated activity