Bastion.
Own the upside. Rent out the noise. Put a floor under the rest.

Owning shares gives you exactly one way to win: the price goes up. While you wait, the shares pay a dividend of one or two percent, if that. The standard alternatives for anyone who needs cash flow from a portfolio are not much better. Bonds lock up capital for safety and pay little for it. The four percent rule is a polite way of saying sell a slice of yourself every year and hope the market cooperates.

Bastion started from a different question: not which fund should I buy, but what am I actually trying to engineer. The answer was three things at once. Real participation if the market keeps doing what it has done for a century. Real income landing every week regardless of what the market did that day. And a maximum loss chosen in advance, rather than discovered in a crash.

Nobody sells that as a single product. So it was built.

The construction, which contracts, at which levels, in which proportions, stays in the kitchen. What can be shown is what the machine delivers, because every property below is measured on real runs against a live brokerage account.

This is not a backtest. Since February 1, 2026 the engine has run 42 positions across a live brokerage account, $2.14M of capital deployed against a fixed $4.16M worst case, through a year that has been anything but kind to the highest-beta names it holds. Every figure below is realized cash from the trade log, net of every losing trade.

And the honest half of the ledger, because it is what separates a track record from a brochure. The same eight months put 20 of the book’s 30 fortresses underwater on the screen, a live drawdown of about $1.09M concentrated in the highest-beta sleeves. After all income and hedging, the book’s net live position marks at about −$442K. That is the trade in plain sight: $585K of cash in the bank against a paper mark that has not been realized, on a book engineered so that the mark cannot fall below a number that was fixed on day one. The income does not stop while the screen is red. Section 4 shows why that matters more than the mark.

Bastion is not bought on margin. The capital is committed upfront, not borrowed on top of the structure, so nothing forces a sale because an account ran out of buying power. The only thing that turns a markdown into a real loss is choosing to sell into it. Give the position real runway, a horizon like December 2028 rather than a headline about this quarter, and the pattern history keeps repeating is that the market has always eventually recovered.

Meanwhile the income keeps landing. Across a century of S&P calendar years, roughly three in four end at a level where the index build beats holding the shares outright. The markdown years come about once a decade (2022-class, healed in about two years) and once a generation (2008, healed in about four). Sitting through the round trip is what Bastion is paid for: the meter runs the entire time the screen is showing you a markdown, while the shareholder waits for free.

Modeled outcomes on the index book, per $100K of worst-case budget, equal max-loss sizing. Red cells are marks at the year boundary, not money spent; they realize only on unwind.

The engine has four settings, so the risk dial matches the person holding it rather than a generic box on a questionnaire. Every build below is sized to the same $100K of maximum loss on the index, from a live run, so the row is the only variable.

Leverage is exposure per dollar committed, not a promised return. Marks are vol-adjusted screen numbers, not bills. Point the same machine at a high-volatility single name and everything roughly doubles: twice the income density, twice the endurance required.

Take the engine off a single stock and run it as a portfolio. The model book below is rebuilt by rule every night from the live book’s realized run-rates, not hand-picked: no correlated cluster above 30%, crypto capped at 8%, built for capital preservation, all-weather income, and diversification. It is what Bastion looks like as a complete answer to the cash-flow problem rather than a single trade.

Every night’s predictions for this model are frozen to a ledger and graded a month later, constituent by constituent, including the exclusions. The model is held to its own forecasts.

Income is the first thing the engine does. It is not the only thing. The same architecture that caps a worst case and pays a weekly premium can be pointed at the opposite problem: capturing a very large, very uncertain upside without betting the account on it.

The working example is the crypto sleeve. A conventional margined book of miners, exchanges, and Bitcoin exposure is worth about $26M if the names reach their end-2027 targets. Getting there the conventional way means carrying roughly $1.19M of worst case plus a forced-liquidation cliff around a 20% drop, and paying a steady bleed of crash insurance that decays exactly when the recovery arrives.

The Bastion version of the same bet reproduces the same $26M target payoff with $137K of net cash and a fixed $633K worst case, funded in full to that reserve so nothing can force-liquidate it. It participates later, deep out of the money until the move is well under way, and that is the point: the shape of the path is traded for the removal of the cliff. Rolled forward in mid-2027 as deeper tenors become liquid, it is a moonshot with a known price tag.

Live marks as of August 22, 2026, modeled to end-2027. Targets are the book’s own scenario assumptions, not forecasts.

A strategy that demands a decision most days cannot live in a spreadsheet. So it lives in software: a private, end-to-end platform that plugs into a live brokerage, pulls market data, rebuilds itself every night, and keeps its screens current in real time through the trading session. It is strictly read-only. It models, ranks, and recommends. It never places a trade.

Replacement cost at Bay Area engineering rates: an estimated $2.7–4.4M and 4.5–6.5 person-years of senior work. Built by one person, part-time, while running against a real account.

If the risks were trivial the premium would not exist. You are being paid precisely because you carry three things.

Weigh that against the other side of the ledger: a five-figure monthly income stream measured in realized cash, a multiple of the upside in the outcomes that happen most years, a worst case that is a chosen number instead of everything, and a floor below which the crash is free. The risks are the price tag. The item is underpriced.

Equal max loss is not equal risk. It is a reallocation. Bastion concentrates its markdowns into the probable band, where history says they heal, and collects a 30–50× income stream as rent for sitting through them. One of these is a machine built on purpose. The other is a hope with a ticker symbol.