Twenty-eight questions,
answered without hedging.

It is a strategy-engineering and market-research environment for evaluating complete trading frameworks rather than isolated rules. The argument it makes is that an entry signal is one of seventeen components that determine how a system behaves, and that the other sixteen — stop geometry, partial policy, branch architecture, concurrency, correlation, friction, regime permission, capital authority and the evidence standard — are usually left as defaults nobody examined. Today it is a working preview of parts of that environment plus a substantial body of educational and research material. It is not a broker, a signal service, a course, or a product that can currently be purchased.

No, and the distinction matters. The engine will tell you that a configuration is internally inconsistent, that friction consumes the modelled edge, that concurrency is concentrated by correlation, that the sample is too small to support a conclusion, or that the drawdown envelope makes recovery unrealistic. Those are statements about coherence, not recommendations. It will never tell you what to trade, when to trade it, how much to risk on a real account, or that a configuration is a good idea. Nothing on this site constitutes investment advice, and no future version is intended to.

No. The Foundry does not provide investment advice, trade signals, brokerage services, account management or performance guarantees, and has no plans to. There are no alerts, no watchlists, no setups of the day and no entry notifications at any tier. The reasoning is set out on the roadmap: the moment a tool issues a signal, the user stops evaluating frameworks and starts following instructions, and every diagnostic on the site becomes decoration around an obedience relationship.

Every record in the archive sits in exactly one state, and every state has a stated exit criterion — a record advances because a condition was met, not because it looked promising. Two of the states are terminal and both are used. Simulated means the result exists but has not been through the validation protocol. In Calibration means it is being tuned and its numbers are expected to move. Validated means it survived all eight validation conditions — a description of process, not of profit. Failed is a destination: the record keeps its abstract, method and reasoning, because a discarded negative result gets rediscovered later at greater expense. Archived means it is complete and no longer being worked on. Nothing is marked Validated on a single history, however favourable it looks.

From publicly discussed trading doctrines — interviews, published writing, widely reported approaches — abstracted into decision frameworks. Each profile states the structural commitment it represents: how position size responds to conviction, how losses are cut, whether the approach pays for its tail or harvests its middle, what market conditions it presumes. What the archive does not contain is anybody's actual rules, entry criteria, parameters or track record, because those are not public and reconstructing them from interviews would be an invention presented as research. The profiles are educational abstractions and the archive says so on every one of them.

Because the archive is a study of decision frameworks, not of people. Each profile is an educational abstraction of publicly discussed doctrines. Claiming to reconstruct a private trading method from public interviews would be dishonest — and attaching a name to such a reconstruction would additionally imply an endorsement that does not exist.

No. The Foundry reads no market data in this phase. The Parameter Laboratory is a deterministic model of directional relationships between parameters — it shows you what moves when you change something, not what would have happened historically. Production backtesting is described on the roadmap as future functionality, and it is the largest single item of unstarted work on the list.

They are real outputs of a real model, and the model is documented. They are not measurements of any market, account or strategy. Every interactive output carries a simulated-data disclosure at the point of display rather than in a footnote at the bottom of the page.

It models: a strategy archetype with a baseline win probability, a tail multiple, a preferred volatility band and a design ATR coefficient; a working stop derived in pips from volatility and coefficient; a friction charge expressed as a share of that stop; an outcome ladder covering losses, partial outcomes, runners and shock losses; a trade frequency; concurrency reduced to effective independent bets by a correlation term; a seeded path simulator that groups trades into concurrency blocks; and an optional governance layer reviewed at block boundaries that compresses risk as decline from the equity peak deepens.

It does not model: prices, bars, or any instrument; order books, queue position or partial fills; the behaviour of a stop triggered in a thin market; news, sessions or liquidity as anything other than parameters you set; correlation that changes during a shock; or the specific rules of any strategy, yours or anybody else's. Slippage is an average charged symmetrically, which is precisely wrong in the conditions where slippage matters most, and the shock-loss term has the smallest real sample behind it of anything in the model while having a large influence on survival estimates.

Because the engine recomputes everything from the full parameter set on every input event. Nothing is cached, interpolated or animated toward a stored value — the entire evaluation runs again, and the outputs are whatever it produced this time. Move the same slider back and the numbers return exactly, because the model is deterministic and every simulated path is seeded.

The reason so many outputs move at once is deliberate. A panel that lets you widen the stop without also changing frequency, friction share, tail participation and drawdown depth is modelling something other than a market. Widening the stop reduces the friction charge as a share of risk, reduces the number of qualifying trades, increases the tail multiple and deepens the currency loss on the trades that fail — all four are consequences of one change, and the engine is built to show them together rather than to let you improve one figure in isolation.

Because a single flattering number should not be able to carry a configuration. Negative net expectancy, excessive per-trade risk, drawdown beyond a survivable depth, friction consuming most of the edge, and high correlation across concurrent positions each impose a ceiling regardless of what the growth figure says. The cap is not a bug to be worked around; it is the part of the score that stops it behaving like every other composite metric that can be gamed by one dimension.

Not in this phase. There is no export, no download, no saved study, no shareable link that carries your parameters and no local storage of any kind — close the tab and the configuration is gone. Saved studies and report generation are on the roadmap at Level 3, and both depend on the same missing persistence layer that blocks thirteen of the twenty-five planned modules. In the meantime, the practical workaround is a screenshot and a note of the parameter values, which is unsatisfying and honest.

MARS — the Montex AlphaRail System — is a spreadsheet-based governance and analytics framework that exists as a separate product. The AlphaRail Foundry Laboratory is the research environment being built around those ideas. The MARS Intelligence section explains the system; the workbook itself is sold separately.

The workbook is a finished spreadsheet system with its own workbooks, calibration tables, reporting cycle and documentation. It runs in a spreadsheet application you already have, it keeps your data on your own machine, and it does not depend on this website in any way. It is bought separately, on its own terms.

The Foundry is a web research environment, most of which does not exist yet. It studies the ideas the workbook implements — capital authority above signal quality, gates derived from decline against the equity peak, a tier ladder capped by the prevailing gate — and models them illustratively so they can be understood and argued with by somebody who does not own the workbook.

Neither includes the other. Buying the workbook does not enrol you in the application. Using the Foundry does not give you the workbook. No planned tier bundles them, and the site does not treat one as an upsell of the other.

Because the calibration is the product. The gate drawdown boundaries, the pool percentages, the tier tables and the workbook's own Monte Carlo completion figures are the result of the work that makes MARS a system rather than a diagram, and publishing them would be giving away a separate product on a page describing a different one. What is published is the structure and the principle — which is enough to apply the ideas to a framework you built yourself. Every governance figure on this site therefore uses an illustrative ladder rather than the workbook's calibration, and each of those figures says so where it appears.

None. There is no analytics package, no tracking pixel, no cookie, no session identifier, no fingerprinting, no advertising network, no embedded video, no map, no comment system and no third-party script of any kind. The site makes no external requests at all — the typefaces are inlined, the stylesheets and scripts are served from the same origin, and there is no font, CDN or API call to anywhere else.

The registration form validates in your browser and shows a confirmation. It sends nothing, writes nothing to local storage, and clears itself afterwards. The motion preference is the single piece of state the site holds, and it is held in memory for the current page only.

This is a description of what the code does rather than a statement of intent. The privacy page records the same position formally.

Not in this phase. The Data Laboratory shows the intended mapping workflow with a demonstration file, clearly labelled as a preview. Nothing is uploaded, processed or transmitted. The workflow you can operate — ten steps, nine validation states, a quality score and an excluded-record accounting — is an accurate representation of the intended module, and it is reading a file that ships with the site rather than one from your machine.

Largely, yes — although it is not packaged as an offline application. Because there are no external requests and no server calls, a page that has finished loading keeps working with the network disconnected: the parameter rails recompute, the Monte Carlo demonstration runs, the charts redraw and the filters work. What will not happen is navigation to a page you have not already loaded, since each page is a separate document fetched normally. There is no service worker and nothing is cached deliberately, so behaviour beyond that depends entirely on your browser.

That is the intention, and the Future Application page describes the five planned stages honestly, including which parts are furthest from existing. No launch date, guaranteed invitation or founder discount is promised. Of twenty-five planned modules, six are working software today, ten are interactive previews and nine are descriptions with no implementation behind them.

Intended launch prices are now published on the plans and pricing record: a permanently free tier, Foundry Plus at $39.99 a month, Foundry Pro at $59.99, Foundry Quant at $79.99, and institutional access scoped per engagement. Annual billing takes 10, 15 and 20 percent off the three paid tiers respectively. This answer previously said no number appeared anywhere on the site, which stopped being true when that page was published; it is corrected here rather than quietly left standing. They remain intended prices — nothing is on sale, no card is being collected, and the figures may change before launch, because the infrastructure costs they have to cover (storage, and the compute for server-side simulation) cannot be estimated until the infrastructure exists.

No. It does not reserve a place, join a queue with a position, fix a price, lock in a discount, commit to a date, or promise that the application is built at all. What it is for is telling us which single module you would need first and which tier you assumed it was in — the input that shapes the build order within the fixed dependency constraints. The form is also a demonstration: it validates in your browser and stores nothing.

Four things, permanently and without an account: the educational material in the learning centre, the research archive including its failed records, the public framework library of twelve worked framework records, and the parameter preview — which is a real engine performing real calculations rather than a locked screenshot. A paid tier could add depth, persistence and scale. It will not add the right to read.

Any current version of Chrome, Edge, Firefox or Safari, on desktop or mobile. The site uses no framework and no build step, so there is very little that can break in an unusual environment. The heaviest pages are the ones with canvas scenes and the Monte Carlo demonstration; on an older phone those will run more slowly, and turning motion off reduces the load considerably. Layouts are tested down to a 360-pixel viewport with no horizontal scrolling at any width. JavaScript is required for the interactive laboratories; the written material, tables and navigation remain readable without it.

It is built to be. Semantic landmarks and a single level-one heading per page; a heading order with no skipped levels; visible focus styling on every interactive element; touch targets of at least 44 pixels; keyboard operation for tabs, accordions and every parameter rail; status conveyed in text as well as colour; dense tables placed in scrolling containers rather than shrunk; and a plain-language text summary attached to every chart so that a screen reader receives the finding rather than a list of coordinates. The accessibility page records the approach in full, including what has not been verified. If something does not work with your assistive technology, that is a defect and it is worth reporting.

It switches off scroll-triggered reveals, parallax, counters, animated canvas scenes and transition effects across the entire site, leaving the layout and every piece of content exactly as it was. Nothing is hidden by turning motion off — the canvas scenes are decorative and everything they depict is also stated in the prose next to them. The toggle sits in the site header and is keyboard-operable. If your operating system is already set to reduce motion, the site starts in the off state without being asked.

Through the contact page, which explains what correspondence is possible in this phase. The reports that are most useful, in order: anything that reads as a promise or a guarantee; a figure or claim that appears to be a measurement of a real market; an interactive output missing its simulated-data marker; a broken link, a layout that overflows, or a control that does not respond to the keyboard; and an argument that one of the engine's modelled relationships is simply wrong.

That last category is the most valuable and the least often submitted. The engine encodes claims — that friction scales against the stop, that partials trade tail participation for capture efficiency, that correlation converts concurrency into concentration — and each of those is arguable rather than settled.

Neither, until it is fixed — a contradiction is a defect on both pages. Where a conflict concerns what is built and what is not, the module register on the roadmap is the authority, because the level pills and the distribution figures on that page are generated from it rather than typed by hand. Where a conflict concerns a modelled relationship, the laboratory methodology section describes the engine's actual behaviour. Report either kind through the contact page.

Yes. It records what is operational, what is a preview and what is not present, in the same three-level vocabulary used everywhere else on the site. Because the site is static and makes no external requests, there is no service to be down in the usual sense — the status page is a statement about capability rather than about uptime.