An architecture review board with a six-week queue is changed overnight: its approval becomes advisory, and teams may proceed without it. What happens over the following six months?

Weeks one to four

Throughput of the board collapses and nobody notices a quality change. Most of what was queued did not need a board: it was routine work that had been swept into scope by a trigger defined as "significant", which in practice meant "anything a nervous manager escalated".

Lead time drops sharply for those changes. The board's calendar empties. This is the period where leadership concludes the board was pure overhead, and it is the most dangerous reading available, because nothing that would have failed has had time to fail yet.

Months two to four

Two divergent behaviours appear.

• Teams that valued the review keep coming, because they were getting something - a second opinion, a cross-team dependency they had not seen, air cover for a decision they wanted to make anyway. Attendance becomes a signal: an advisory board's calendar is an honest measure of whether it was useful.
• Teams that were being constrained stop coming, and their decisions stop being visible. The board loses not its authority but its information. This is the real loss, and it is invisible for a while.

The second effect is the one to design against. A board that saw everything could spot the third team building the same queue. An advisory board sees a self-selected sample.

Months four to six

The first decision that should have been challenged reaches production - usually an integration coupling two domains that should not have been coupled, or a datastore chosen for a workload it does not fit. The cost lands on a different team and a later quarter, which is why the feedback loop back to the governance change is weak enough that most organisations never connect them.

The organisation now chooses between reinstating a gate (and rebuilding the queue) or replacing it with something structurally different.

What stops the bad ending

• Keep a mandatory trigger, but make it narrow and objective. Not "significant": crosses a domain boundary · introduces a new datastore or language to the estate · touches regulated data · is hard to reverse. Four rules a team can self-assess in a minute, covering perhaps 5% of changes.
• Replace approval with a published decision record. The board's information comes from reading ADRs rather than from holding a gate, which costs teams almost nothing and preserves the cross-cutting view.
• Measure the board on lead time and on decisions changed, not on volume reviewed. A board that changed no decisions in a quarter is not a light-touch board; it is an inactive one.

When this is simply the right answer

Below roughly 150 engineers, the cross-cutting view exists in people's heads and a gate adds latency to information that already flows. The gate earns its cost when no single person can name every service, which is also the point at which "advisory plus a narrow mandatory trigger" beats both extremes.