Decision gates in project management

A decision gate is a governed checkpoint with three parts: the entry criteria (the deliverables and evidence required to be reviewed), the decision itself, and the consequence that decision has on everything downstream. The review answers one question — given what we now know, should the program proceed, and in what form?

The four canonical outcomes cover the realistic space of answers. They are not interchangeable, and a good gate process forces an explicit choice rather than letting a program drift past the checkpoint unexamined.

R&D is gate-shaped because its work is uncertain and its phases are expensive. You do not want to fund a Phase 2 trial, a pilot fabrication run, or a flight test until the evidence from the prior phase justifies it. Gates are where that capital discipline lives.

In biotech, decision milestones gate the path to and beyond IND: a candidate-selection gate, a development-candidate gate, the IND-enabling package, and clinical phase transitions — each a Go/No-Go on real data, often with hundreds of thousands of dollars per day of delay riding on the timing. In deep tech, Technology Readiness Level (TRL) gates mark the jump from breadboard to prototype to qualified system; today those calls are too often informal hallway decisions rather than governed checkpoints with probabilistic backing. In federally funded research, program gates and decision points are frequently contractual — DARPA and ARPA-style programs structure funding around go/no-go decision milestones that determine whether the next option period is exercised.

A milestone is a marker — it says "this happened" or "this is due." A gate is a fork: the same point in the plan branches into different futures depending on the decision. Treating the two as the same is the central modeling error in most PM tools, and it has real cost.

When a gate is only a milestone, a No-Go does not remove the downstream tasks — someone has to do it by hand, and the board deck still shows work that is no longer happening. A Pivot does not rewire dependencies — the analyst rebuilds the network manually. A Defer slips one diamond but leaves every successor on its old date until a PMO spends a day re-planning. The decision is captured; the plan it should have changed is not. For a program where a single gate governs millions in spend, that gap between "decision recorded" and "plan updated" is exactly where schedules quietly go wrong.

Modeling a gate as a real schedule object — not a milestone — means the decision and its consequence are linked. A No-Go or Pivot decision triggers a retroactive reschedule that recomputes and propagates the consequence through the dependent plan, instead of leaving someone to redraw the network by hand. The decision and its schedule impact are recorded together, and the audit trail captures who decided what, when, and on what evidence.

This is also where gates connect to the rest of the method stack. The evidence at a gate is often probabilistic — a P50/P80 finish date from Monte Carlo, a Cost-of-Delay figure, a buffer-consumption reading from a fever chart. The decision feeds back into the schedule, which is re-simulated, which updates the next gate's outlook. Gates, risk simulation, and buffers stop being separate reports and become one closed loop.

In CritPath AI, decision gates are first-class schedule objects. A gate carries a real Go/No-Go/Pivot/Defer decision. Deciding No-Go or Pivot triggers a retroactive reschedule — the canonical rescheduling engine re-cascades the consequence through the dependency graph, recomputing CPM, the critical chain, buffers, and Monte Carlo finish dates in one step. A Go decision proceeds without forcing a reschedule. No incumbent PMO tool ships gates as live schedule objects with this propagation.

Because the AI copilot is grounded in the same dependency graph, you can ask what a given gate decision does to your committed date before you make it — and the decision, its rationale, and its schedule impact land in an append-only audit log that supports AACE 132R-23 Level 4 governance. All of this is part of the $10/user/month platform, with AI usage billed separately by metered usage.