Skip to content

Stage-Gate Exploration

Staged exploration process — instantiates Opportunity-Gated Adaptive Diversification

Runs exploration as a sequence of funded stages separated by decision gates, releasing more budget only to the lines that clear each gate's evidence bar.

Fanning many lines into a new space is only affordable if each line is cheap to start and cheap to stop. Stage-Gate Exploration is the process that makes it so: it splits exploration into a fixed sequence of stages, each ending in a gate where accumulated evidence decides whether a line advances, waits, or dies — and it releases budget in tranches, so early bets are small and only proven lines earn the expensive later stages. Its defining move is conditional escalation: commitment is a function of demonstrated progress, never granted up front, which turns one big irreversible bet into a chain of small reversible ones.

Example

A pharmaceutical program is exploring several candidate compounds against a newly druggable target. It doesn't fund them all to completion — it runs them through staged gates in the spirit of the familiar Phase I → II → III progression. The first stage is cheap and broad: basic safety and signal across all candidates. At the gate, most drop; only those clearing an explicit, pre-set bar receive the far larger next tranche for a focused efficacy trial. Another gate, another cut. By the final, most-expensive stage, the budget is concentrated on the one or two compounds the evidence has earned.

The power of the pattern is the asymmetry: the program spends little to learn a lot early, and spends a lot only where learning has already de-risked the bet. A compound that fails does so having consumed a small stake, not the whole program — and the gate that killed it did so on evidence agreed before the stage began.

How it works

Stage-gating differs from simply "reviewing progress" by pre-committing the structure of escalation:

  • Lay out the phase plan. Define a fixed sequence of stages, each with an explicit gate criterion set in advance of the evidence.
  • Meter the budget. Release the exploration budget tranche by tranche — small early, large late — so cost tracks demonstrated progress rather than initial enthusiasm.
  • Run the gate loop. At each gate, feed accumulated evidence into a selection decision — advance, hold, or kill — and let only survivors draw the next tranche.

Tuning parameters

  • Gate spacing — many closely-spaced gates or a few wide stages. More gates learn and cut faster but add review overhead and can starve slow-maturing lines.
  • Gate bar height — strict bars kill more (fewer false positives, but they cut good-but-slow lines); lenient bars let more through and risk funding sprawl.
  • Tranche sizing — how steeply budget escalates across stages; steeper escalation punishes late failure harder but concentrates learning where it's cheapest.
  • Fail default — whether missing a gate kills a line or merely holds it pending more evidence.

When it helps, and when it misleads

Its strength is that it caps downside on every individual bet and directly counters escalation: because commitment is staged, a losing line is stopped having spent a fraction of what a full bet would have cost, and the sunk cost never grows large enough to trap the decision.

Its failure modes are the classic ones for any gated process. Gates become theater — rubber-stamps where nothing is ever killed, so the staging is nominal and budget flows regardless. Relatedly, the review is run backwards: the gate is convened to justify continued funding of a favoured line rather than to test it, which is exactly the sunk-cost escalation the mechanism exists to prevent.[1] And bars tuned for fast signal quietly kill lines whose payoff is real but slow to appear. The discipline that keeps it honest is to fix each gate's criterion before the stage runs and to actually honor kill decisions — a gate that cannot say no is not a gate.

How it implements the components

Stage-Gate Exploration fills the archetype's staged-commitment components — the ones that make wide exploration affordable and stoppable:

  • phase_plan — the fixed sequence of stages and the gate criteria that separate them.
  • exploration_budget — the resource released tranche by tranche, escalating only for lines that clear a gate.
  • feedback_and_selection_loop — the gate itself: evidence in, an advance / hold / kill decision out, repeated each stage.

It does not shelter a line between gates (that's Protected Pilot Lane), map the niches being explored (that's Opportunity Landscape Mapping), or make the final keep-and-merge call once the space fills (that's Preserve–Prune–Recombine Review).

References

[1] The sunk-cost fallacy — continuing to invest in a course of action because of what has already been spent rather than expected future value. Staged gates are a standard structural defense against it: by keeping each committed stake small and re-decidable, they stop any one line's sunk cost from growing large enough to capture the decision. ↩