Reverse discounting is a way to reason backward from a future outcome to the earlier capabilities that made it possible. Its practical purpose is to help decide where in a chain of technological development to intervene now.
The central claim is that an early enabling capability can influence many later developments at once. An intervention at a shared “trunk” of a technology tree may therefore have more reach than an intervention aimed at one later application.
The phrase “passing technology back through time” is metaphorical. No device, information, or signal travels into the past. A person starts with a future goal, traces its prerequisites backward, and creates something in the present—a standard, tool, dataset, API, benchmark, institution, or other enabler—that can support future work sooner.
1. Start with a future outcome
Imagine a future technology or outcome you want to make possible: a medical treatment, a new kind of energy system, a scientific instrument, or a widely used software capability.
Instead of asking only:
What should we build next?
reverse discounting asks:
What had to exist for this future outcome to be possible, and which of those prerequisites also supports other valuable developments?
That second question directs attention away from the final product and toward its dependencies. A future outcome may depend on specialized equipment, but that equipment may itself depend on shared data formats, measurement standards, manufacturing tools, trained people, regulations, or institutions.
The aim is to identify an earlier capability with multiple downstream uses.
2. Represent development as a dependency graph
The note models a field as a directed acyclic graph:\[ G=(N,E) \]
- A node is a capability or institution: a tool, standard, API, benchmark, regulation, dataset, or other prerequisite.
- A directed edge \(i\to j\) means that node \(i\) enables node \(j\).
- A node’s downstream nodes are the developments that depend on it, directly or through other nodes.
For example:\[ \text{shared data format} \to \text{compatible datasets} \to \text{better models} \to \text{new applications} \]
A data format can be easy to overlook because it is not itself the application anyone ultimately wants. But if many later projects rely on it, improving or establishing the format might affect all of them.
The graph is meant to capture the fact that technological development is usually not a single sequence. It is a network: developments branch, merge, and share prerequisites.
3. Estimate the value of intervening at a node
For each candidate node \(i\), the note proposes:\[ PV_i = L_i \cdot p_i \cdot a_i \cdot c_i – K_i \]
Here, \(PV_i\) is the estimated value of acting at node \(i\).
\(L_i\): leverage
Leverage is the value that could become reachable if the branch succeeds. It is intended to reflect:
- how many downstream developments the node supports;
- how much time the intervention could save; and
- whether the benefits can be reused across many projects.
A shared standard might have high leverage because it helps multiple organizations build compatible tools and products. A one-off improvement to a single application may have lower leverage because its effects are more limited.
The note gives a rough expression:\[ L_i \approx \left(\prod_{j\in \text{path}(i\to\text{leaves})} f_j\right)\Delta t_i \]
In this expression, \(f_j\) represents the number of downstream branches associated with node \(j\), and \(\Delta t_i\) is the time saved by intervening there.
This is intended as a picture of compounding reach: when each stage opens several later branches, the number of potentially affected developments can grow rapidly.
\(p_i\): credence
Credence is confidence that the proposed branch is real: that the future outcome is plausible and that the supposed prerequisites are genuinely relevant to it.
The note expects confidence to fall as one reasons farther back from a specific, observable outcome. A distant prerequisite may appear useful only because the analyst has a particular future in mind; the same prerequisite may turn out to be unnecessary or replaceable.
\(a_i\): actionability
Actionability is the fraction of the node’s potential leverage that a particular intervention can actually influence.
A node can be strategically important while remaining difficult to change. For example, a technology may be bottlenecked by a standard, but changing that standard may require agreement from many groups. In that case, the standard’s theoretical leverage may be large while the intervention’s actionability is small.
\(c_i\): capture
Capture is the share of the value created that the actor making the intervention can retain or appropriate.
A foundational tool may benefit many people without giving its creator much private return. A product serving one market may be easier for its maker to monetize. Capture matters for a private investor or company deciding whether the intervention is worthwhile.
\(K_i\): cost
Cost is the cost of intervening at node \(i\), including money, time, coordination, technical effort, and the risk of committing to the wrong path.
The equation combines these considerations: a node looks attractive when its reach, plausibility, actionability, and capturable value are high relative to the cost.
4. Why the note says value can grow backward
The note argues that tracing dependencies toward earlier nodes can reveal more leverage, for three reasons.
Fan-out
A shared prerequisite may support many later projects. If a standard is used by ten systems, and each of those systems supports ten applications, the standard may sit upstream of a large number of possible outcomes.
The note illustrates this with repeated fan-out. If a branch expands by a factor of ten over three stages, then:\[ 10^3=1000 \]
That conveys the intuition that early nodes can reach many later developments.
This is an illustrative simplification, not a general counting rule. Real technology graphs often share dependencies. Two branches may lead to the same downstream outcome, and a node may not independently multiply the value of every later node. In an actual analysis, downstream reach should be estimated from the graph rather than assumed to equal a product of fan-out numbers.
Non-rivalry
Many enabling ideas or tools can be used by multiple people at once. One group’s use of a shared standard does not normally prevent another group from using it.
This is the economic idea of non-rivalry. It helps explain why an upstream capability can have broad social value: the same investment can support many downstream projects.
Feedback loops
Some technologies improve the conditions for improving themselves. For example:\[ \text{more data} \to \text{better models} \to \text{better experiments} \to \text{more useful data} \]
The note calls the strength of such a reinforcing loop \(g\), and writes:\[ L = \frac{\prod f_j}{1-g} \]
As \(g\) approaches 1, this expression becomes very large. The intended intuition is that a strongly reinforcing system can produce much more growth once a useful loop is established.
In practice, feedback loops have costs, limits, delays, and diminishing returns. The expression is a conceptual model; it does not establish that leverage literally becomes unbounded when a feedback loop is strong.
5. The analogy with ordinary discounting
In ordinary financial discounting, a future payoff is converted into a value today:\[ PV = \frac{FV}{(1+r)^t} \]
A future payoff is worth less today when it is farther away or when the discount rate \(r\) is higher.
The note proposes a reverse counterpart. Define a reverse discount rate:\[ \rho_i=f_i-1 \]
Then, over a path of \(m\) stages:\[ L=(1+\rho)^m \]
The analogy is:
- ordinary discounting reduces a payoff as one looks forward across time;
- reverse discounting increases the estimated reach as one traces backward across stages of branching development.
The structural resemblance is that both use a repeated multiplier. But the quantities being multiplied have different meanings. Financial discounting applies a discount rate to time and money. The reverse formula applies branching to a dependency graph. Calling them “duals” is the note’s conceptual framing; it does not make them mathematical inverses in a strict financial or graph-theoretic sense.
6. Choosing where to “fork”
The note’s most important decision rule is that the earliest possible intervention is not necessarily best.
Moving earlier can increase leverage: one intervention may support more future work. But moving earlier can also reduce confidence: the proposed prerequisite may be uncertain, replaceable, or irrelevant.
The note expresses this tradeoff as:\[ \frac{L’}{L}=-\frac{p’}{p} \]
at the optimum. In words:
Move earlier while the percentage gain in leverage is larger than the percentage loss in confidence. Stop when those two rates are equal.
The equation is best read as a simplified balancing condition. It assumes the relevant functions can be measured and compared smoothly, and the displayed form does not include changes in actionability or capture. If \(a_i\) or \(c_i\) also vary with position, they need to be included in the optimization too.
The practical rule in the note is:
Choose the earliest shared prerequisite whose necessity is already clear enough to act on.
“Earliest” means farthest upstream. “Shared” means useful to several downstream paths. “Necessity is already clear enough” means there is credible evidence that the prerequisite matters, rather than merely a plausible story about how it might matter.
A fork placed too far downstream may miss the benefits of a shared enabler. A fork placed too far upstream may rest on speculation.
A special case: designated development points
Sometimes the target node has already been chosen for other reasons—for example, an organization has designated a technology or capability as a priority. The note treats the target’s confidence as supplied by that choice. Under that assumption, the practical question becomes how early the organization can act, given the prerequisites and the cost of accelerating them.
7. What “passing technology back” means operationally
The note uses a Git fork as an analogy.
- The “main branch” represents the originating future or baseline trajectory.
- An earlier commit represents an existing capability or point in the dependency graph.
- A fork represents a new development path that starts from that earlier point.
- The forked work is a present-day artifact, such as a document, standard, API, benchmark, tool, or institutional arrangement.
Under this analogy, the original path is not changed. The intervention creates another path that begins from an earlier enabler and proceeds forward.
So “passing technology back” means:
- Choose a desired future capability.
- Trace its dependencies backward.
- Find a shared prerequisite with meaningful downstream reach.
- Create or improve that prerequisite now.
- Let future projects build on it as they proceed forward.
The language of “fork pointers” emphasizes that the act changes which development path is available. It does not change the past.
8. How the note expects the future to arrive faster
The note identifies four mechanisms.
More developments advance together
An upstream intervention may remove a delay that affects several downstream programs. If it saves a year for a shared prerequisite, several projects might benefit from that year. The total portfolio impact could therefore exceed the time saved on any single project.
This depends on whether the projects actually share the prerequisite and whether the intervention is the limiting factor for each one.
Starting a productive process earlier
If improvement compounds over time, starting the process sooner can produce more cumulative progress than starting later with a somewhat higher rate. This is a general feature of compounding systems, though the size of the effect depends on the rate, duration, and practical limits.
Acquiring an option early
An early intervention can act like buying an option: a relatively small commitment may preserve the chance to expand later if the path becomes promising. If the path fails, the early work may be abandoned.
The “cheap” part is not guaranteed. Early work can still be costly, and uncertainty can make a project expensive to maintain. The option analogy is strongest when the intervention has limited downside, can be tested early, and leaves later choices open.
Attracting collaborators and resources
A visible early enabler can attract funding, talent, and partners. The original actor may contribute a useful starting point, while other people and organizations bear much of the cost of building on it.
That can multiply the original contribution. It also means the originator may capture only a small share of the resulting value.
9. The capture problem and public investment
The note identifies a tension:
- An upstream enabler can create large benefits for society.
- Because many actors can use it, the creator may be unable to charge for all those benefits.
In the formula, that means \(L_i\) can be high while \(c_i\) is low. A private company may therefore find a high-value foundational project unattractive, even though the project would benefit many organizations.
The note calls this a reason for underinvestment in trunk nodes: shared standards, benchmarks, data schemas, APIs, and regulatory precedents may lack a single actor with enough incentive to build them.
It suggests ways institutions can address that gap:
- Consortia can share costs and coordinate adoption.
- Mandates can make a common standard broadly usable.
- Prizes can reward useful foundational work.
- Public funding can pay for capabilities whose benefits are widely shared.
- Liability rules can create incentives to provide necessary safeguards or infrastructure.
These are policy possibilities, not automatic solutions. Each can also create coordination costs, lock in a poor design, or give decision-makers incentives to favor the wrong path.
10. How the framework can fail
The note lists several failure modes.
The wrong bottleneck
A node can look central on a diagram but have little practical effect if it is not the actual constraint. If progress is limited by another missing resource—such as manufacturing capacity, legal approval, or skilled labor—then improving the proposed enabler may save little time.
In the equation, this is a low \(a_i\): the intervention cannot influence much of the apparent leverage.
Too much uncertainty
The further back one reasons, the less certain the chain may become. If confidence falls faster than leverage rises, the earlier intervention has lower expected value.
Everyone sees the same opportunity
If many actors identify the same future and race to build the same enabling node, the early advantage may disappear. Competition can also create duplication or make it harder to capture the value.
Little private capture
An enabler can be socially valuable and privately unprofitable. That is the capture problem expressed as low \(c_i\).
Anchoring
Once an actor commits to a proposed fork, the commitment can become part of its identity. It may then keep defending the path even when evidence changes. The initial act of choosing a branch can distort later judgment.
Choosing the wrong target
If the organization has designated the wrong future development, accelerating its prerequisites can make the mistake more costly. The note calls this particularly dangerous because acceleration compounds the consequences of a mistaken choice.
11. The note’s claim about spacetime
The note explicitly says the framework requires no retrocausation—no causal influence from the future back to the past.
Its reasoning is:
- Past events and capabilities remain as they were.
- Someone in the present constructs an artifact based on a model of a possible future.
- That artifact affects what people can do in the future.
- The only “backward” movement is the direction of analysis, from a goal toward its prerequisites.
The “future” supplies the question; present action supplies the cause; later developments follow forward from that action.
The note also separates two claims:
- A particular future outcome may be contingent: it could have happened differently.
- The physical structure of spacetime may remain stable despite that contingency.
That is the note’s philosophical framing. Its practical model does not need a claim about the physical robustness of spacetime, because the proposed intervention is an ordinary present-day action.
12. The whole method as a sequence
In practical terms, the method is:
- Name a future outcome. Be specific enough to identify what success would mean.
- Map its prerequisites. Include technologies, standards, data, people, institutions, and regulations.
- Trace the graph backward. Look for nodes that support several downstream paths.
- Check whether the node is a real bottleneck. A central-looking dependency may not be the limiting one.
- Estimate the terms. Consider reach, confidence, actionability, capture, cost, and time saved.
- Compare candidate fork points. Earlier nodes may have more reach but also more uncertainty.
- Choose a tractable upstream intervention. Prefer a node whose relevance is legible and whose development can actually be influenced.
- Create a present-day artifact. Build the standard, tool, benchmark, data resource, API, or institution that lets downstream work begin sooner.
- Keep updating the model. Revisit the target, dependencies, costs, and evidence as the field develops.
13. The idea in one sentence
Reverse discounting means starting with a desired future, tracing backward to find an early prerequisite that supports several future paths, and creating that prerequisite now—while balancing its potential reach against uncertainty, influence, capture, and cost.