60–80% of the worth is in the part you didn't model, and your discount rate decides if you build for it.
Picture a junior analyst at an investment bank at 1 a.m., building a discounted-cash-flow model to value a company. The spreadsheet is a thing of obsessive beauty: ten years of explicit projections, revenue lines and margin assumptions and tax rates, every quarter forecast and footnoted, hundreds of cells of defensible, hard-won detail. It is the most precise document she will produce all year. And then, at the very bottom, in a single cell, sits a line called terminal value: the company's worth in all the years after the forecast ends, the part she did not model at all. She computes it from a one-line formula and two guessed numbers.
That one cell is usually 60% to 80% of the entire answer.
Read that again, because it is one of the strangest and most under-internalized facts in all of valuation. The part the analyst modeled with surgical care, the ten years she can actually see, is the minority of what the company is worth. The majority lives in the part she explicitly didn't model: the open-ended future beyond the horizon, collapsed into a single number governed by two hand-wavy assumptions. She spent 95% of her effort on 20% of the value. And here is why this should matter to anyone who builds software: you do the exact same thing, every time you justify a project by what it'll do this quarter, and the lever that decides whether you can even see the other 80% is a number most engineering orgs have never said out loud.
A discounted-cash-flow valuation has two parts. The first is the explicit forecast: typically five to ten years of projected cash flows, each one discounted back to today. This is the part that looks like rigor. The second part is the terminal value: a single figure capturing every cash flow after the explicit window, stretching to infinity, because a healthy company doesn't politely stop existing in year ten.
The standard way to compute it is the Gordon Growth Model, named for economist Myron Gordon, and the whole of it is this: Terminal Value = FCF × (1 + g) / (WACC − g). Just three inputs. FCF is the final year's free cash flow. g is the perpetual growth rate: how fast you assume the company grows forever, usually pinned to long-run GDP, somewhere around 2% to 3%. And WACC is the discount rate: the rate at which future money is converted to present money.
Two things about that formula should make you nervous. First, the entire infinite tail of the company's existence is governed by two assumptions, g and the discount rate, both of them educated guesses. Second, the formula is violently sensitive to them: a swing of one percentage point in the growth rate can move the valuation by 20% to 30%. The most consequential number in the whole model is also the least precisely knowable. The precise part is small; the imprecise part is most of the value. That isn't a flaw in DCF; it's a true description of where worth lives. Most of what something is worth is in its long tail, which you cannot forecast and are therefore tempted to ignore.
Now hold a software project up against that structure, because it has the same two parts.
The explicit forecast of an engineering project is its visible, measurable, near-term impact: the feature ships, the conversion metric ticks up, the latency drops, the demo lands in the all-hands. This is the part you can put in a planning doc and defend in a review. It's the part that feels like rigor.
The terminal value of that project is everything after the horizon, and it's exactly the part you can't put a clean number on. It's the platform's optionality: the dozen features it makes cheap to build that nobody's even proposed yet. It's the refactor's decade of compounding maintainability: every future change that's now an hour instead of a week. It's the abstraction's reuse: the multiplier that pays out every time someone builds on it for years. It's the reliability work that quietly prevents a hundred future incidents that, because they never happen, never appear in any metric you can point to.
And just like the analyst's spreadsheet, the terminal value is usually the majority of the real worth, and the part you systematically ignore, precisely because you can't forecast it. When a team says "we can't justify the platform investment, the near-term ROI doesn't pencil out," they are doing something mathematically identical to valuing a company on its explicit forecast alone and writing zero in the terminal-value cell. They are confidently valuing the 20% and declaring the 60–80% to be worth nothing, not because they decided it's worthless, but because it didn't fit in the spreadsheet. The most valuable engineering work is mostly terminal value. Which means the question of whether you ever build it comes down to one number.
That number is the discount rate, and it is the most quietly powerful parameter in finance, in climate policy, and in your roadmap. Its job is to answer a single question: how much is a dollar in the future worth today? The mechanism is the present-value formula: a future amount divided by (1 + r) raised to the number of years away. The higher the rate r, the more savagely distant money gets crushed toward zero.
Make it concrete. Imagine a piece of work that pays off $1 million in year 15: a platform that becomes load-bearing, a reliability foundation that holds for a decade and a half. What is that worth today?
The identical future payoff is worth nearly five times more to the patient valuer than the impatient one. Stretch the horizon and it gets brutal: $1 million in year 25, discounted at 20%, is worth around $10,000, about one percent of its face value. At a high enough discount rate, the far future isn't merely discounted. It's deleted.
Here is the part that should reframe how you think about your own organization. Your team has a discount rate, even if no one has ever named it. It's encoded in how you talk: "what did this do for us this quarter," "we'll worry about that later," "ship it, we'll clean it up next sprint." That reflexive short-termism is a high discount rate, applied silently to every decision, and a high discount rate, by the math above, crushes the present value of anything whose payoff is distant. You are not neutrally evaluating the future. You have, without saying so, assigned it a price near zero.
This is the insight that turns a finance curiosity into a diagnosis. Consider which engineering investments have their value concentrated in the terminal tail: platforms, refactoring, reliability, abstractions, documentation, test infrastructure. Their defining feature is that the payoff is mostly distant and compounding: small or even negative this quarter, enormous over years. Their worth is overwhelmingly terminal value.
Now apply a high internal discount rate to them. You are discounting away exactly the part where their value lives. It's not a near miss; it's a precise decapitation. The platform's worth is in years 3 through 15, and a 20% rate has already declared years 3 through 15 to be worth pennies on the dollar. So the platform investment "doesn't pencil out," every single time, with mathematical reliability.
The chronic, industry-wide under-investment in foundational work is therefore not primarily a failure of taste or wisdom. Most leaders say they value reliability and clean architecture, and they mean it. The under-investment is a failure of arithmetic that no one performed out loud: an implicit, uniformly high discount rate that makes the mostly-future payoff of foundational work worth almost nothing in today's terms. You don't decide not to build the platform. Your unspoken discount rate decides it for you, and then hands you a story about ROI to make the decision feel reasoned.
If you think one parameter can't possibly carry that much weight, consider the single most expensive discount-rate argument in history. In 2006, the British economist Nicholas Stern published the Stern Review on the Economics of Climate Change, a 700-page report commissioned by the UK government. Its conclusion was stark: the world should spend roughly 1% of GDP now to avert far larger future damages, act aggressively, act immediately. The report landed like a thunderclap.
Then the economist William Nordhaus, who would later win the 2018 Nobel for his work on climate economics, published a critique, and his core point was devastating in its simplicity. Stern's dramatic conclusion, he showed, "depends decisively on the assumption of a near-zero social discount rate." Stern had used a discount rate of about 1.4% (built from a pure time-preference of just 0.1%, treating a future person's welfare as almost equal to a present person's). Nordhaus and most economists used 3% to 5%. That gap, a few percentage points on one parameter, was the entire disagreement. With Stern's patient rate, spending trillions today to protect the distant future is obviously worth it. With a conventional rate, the same distant benefits discount down to where aggressive immediate action no longer "pencils out." Same physics, same damages, same models, opposite policy, hinging on one number that encodes how much we think the future is worth.
That is the deepest truth here, and it's why this transfers across finance, climate, and your codebase without distortion. The discount rate is not a technical input. It's a values statement disguised as a number: a declaration of how much you care about the version of yourself, your company, or your planet that exists in fifteen years. Stern and Nordhaus weren't really arguing about math. They were arguing about the future's worth, in the only language a spreadsheet understands.
The practical move is not "always be patient"; sometimes a high discount rate is correct. A startup with eleven months of runway should discount the future steeply, because it might not have one; survival-critical work genuinely is worth more than a platform that pays off in year five they may never reach. The error is not having a discount rate. The error is having one you've never named, and applying it uniformly to everything.
So, three things you can actually do:
The analyst's spreadsheet has a lesson in it that took the climate economists a Nobel-grade fight to surface: most of the value is in the part you didn't model, and a single number you rarely say out loud decides whether you build for it. Say it out loud. The future is listening, and right now you might be telling it it's worth almost nothing.
Trust infrastructure is mostly terminal value. Don't write zero in the cell.
The trust an agent economy will run on, portable identity, a reputation that compounds with every verified outcome, and a tamper-evident record of what each agent actually did, is foundational, hard-to-reverse, compounding work. Its worth is overwhelmingly terminal value: it pays off in the years when agents transact at scale and "can I trust this one?" is the constant question. Price it at this-quarter ROI and it never pencils out; price it as the load-bearing foundation it is, and you build it before you need it. The Agent Trust Stack is that foundation, the part of the model worth more than the part you can see.
pip install agent-trust-stack · npm install chain-of-consciousness agent-rating-protocol