The R&D Funnel

Intelligence is not built in a straight line. It is explored along an arc — and along that curve, most directions are noticed, weighed, and let go long before anyone writes a line of code. The R&D Funnel is how Arc reads that curve: a maturity model that tracks a single project from a raw signal to something in the world, one evidence-gated stage at a time.

It answers one question, and only one: how far has this project travelled from idea toward a usable system? Not what it is, not whether anyone is working on it today, not how much it matters to Arc's story — those are separate questions, kept deliberately separate. The funnel is about readiness alone.

The filter is the asset

Arc generates far more directions than it builds. A frontier house lives on contact with hard, uncertain problems, and most of them resolve, on inspection, into not yet, not here, or not worth it. That filtering is not waste. The directions validated and set down are Arc's record of judgement — the evidence that this is research and not packaging. A house is known as much by what it declines as by what it ships.

So the funnel is wide at the mouth and narrow at the neck. Many signals enter; few become missions; fewer still become flagships. The discipline is to let a direction prove itself at each gate rather than to carry every interesting idea forward on enthusiasm.

Many signals, few missions, fewer flagships.

The eight stages

Each stage is a gate, not a label. A project does not move forward because someone renames it; it moves when it has earned the next gate's evidence.

• Signal — an opportunity, a pain point, a technical trend, a research result, a customer problem has been noticed. It is worth recording, not yet worth committing to.
• Concept — the signal has been named and given shape: a one-line thesis, a rough problem, a visible user or technical route. Clear enough to discuss; not yet proven.
• Validated — a critical assumption has held. The feasibility, the real user need, the gap in the field, or the commercial case has survived first contact with evidence.
• Prototype — a core mechanism runs. A scaffold or pipeline exists that proves the thing can be built. It may be rough, partial, and far from user-ready.
• Alpha — the system demonstrates end to end. The core path works, but still needs explanation, setup, or supervision — a thing you can show, not yet a thing others can use alone.
• Research Preview — a small number of real users, collaborators, or early customers can try it, with support and monitoring. It carries genuine value while remaining experimental.
• Productisation — the value is proven, and the work turns to making it a product: stability, deployment, documentation, onboarding, access, support — the unglamorous distance between it works and it holds.
• Shipped — it has entered the real world. Released, delivered, deployed, in continuous use.

Three distinctions that keep it honest

A maturity model is only as good as its gates, and three of them carry most of the weight.

Prototype proves it can be built; Alpha proves it can be shown. A full-stack scaffold and a core pipeline are a Prototype. What comes back running end to end — demonstrable, but not deployable — is an Alpha. Much of what is loosely called an "MVP" is, precisely, an Alpha.

Alpha is for experts; Research Preview is for real users. The gate between them is the first genuine outside use, however small. That is where a system stops being a demonstration of feasibility and starts being a test of value.

Product is not a stage — Productisation is. Once a thing genuinely is a product, it is ready to ship; there is no holding pattern called "Product" in between. So the funnel names the work of becoming one — Productisation — and lets Shipped carry arrival. Arc retired the term MVP from the funnel for the same reason: it means four different things to a student, a founder, an engineer, and a customer, and a gate cannot afford an ambiguous word.

Why a funnel, and not a status

The funnel is one lens among several. A project early in maturity can be strategically central; a one-off delivery, complete and closed, can still be a reserve asset worth reviving. Those truths only stay legible if maturity is read on its own axis, never folded into a single status that also tries to carry priority, progress, and importance at once.

Read this way, the funnel is more than tracking. It is how Arc matures a deep technical option — carrying a direction far enough to know what it is worth, then holding it at the right stage until the conditions to exercise it arrive. Most signals never reach the neck. The few that do are the ones Arc chose, deliberately, to build.