An architecture guild's charter - a monthly 60-minute meeting with a rotating presenter and the stated purpose of sharing knowledge and aligning on best practice - tracks attendance as its success metric and produces no decisions and no artefacts. Twenty months in nothing about how services are built has changed. Which single change to the charter is most likely to change the estate?

What is actually required

A guild exists to raise the quality of other people's architectural judgement. Judgement develops by committing to a call on a real case and then finding out, which is a feedback loop; a presentation is an information transfer and information is not the scarce input here. Senior engineers already know the patterns. What they lack is repetitions of deciding with something at stake.

That diagnosis explains the twenty months. Nothing in the charter creates a stake, so nothing in the estate could have changed, and the attendance metric was measuring the input to a process with no output.

The one change that matters

A real in-flight design brings three things a presentation cannot: a deadline, a team that will act on the outcome, and a consequence that becomes visible within a quarter. Attaching the guild to one each month makes every attendee's opinion cost something.

The published outcome is the compounding part. Twelve written outcomes a year is a precedent set, and a precedent set is how a standard forms without anyone having to mandate it — the thirteenth team reads the four decisions closest to theirs instead of arguing from scratch. That is also the only route to influence a guild has, since it holds no authority over anyone.

What to remove and what to leave

• Remove the attendance metric. It is why a forum producing nothing survived twenty months: it was succeeding against its own measure. Replace it with designs reviewed with a published outcome, and the share of those where the outcome changed the design. Watch the second number: near zero means a rubber stamp, near 100% means designs are arriving too late to be shaped cheaply.
• Remove "share best practice" as a purpose. It cannot be failed, so it cannot direct anything.
• Leave the monthly cadence. In a mid-size estate roughly one design a month is worth this much attention; a weekly slot would starve the format and fill with filler.
• Leave the rotation, but rotate who leads the review rather than who presents. Leading a review in front of peers is the repetition that builds the skill.

Why the other options fail

• Weekly. More instances of a format that produces nothing. Frequency multiplies cost, not output, and attendance will fall, which under the existing metric reads as failure and gets the guild cancelled.
• Mandatory attendance. Converts a volunteering problem into a compliance problem. People attend and disengage, and the guild spends the only currency it has — the credibility that makes people bring real problems to it voluntarily.
• Publish minutes. The cheapest-looking option and the emptiest: minutes of a meeting that decided nothing faithfully record that nothing was decided. It is the change that lets everyone feel the problem was addressed.

When this is the wrong answer

In a fifteen-engineer company, delete the forum and have the conversation. And where the estate's problem is a missing platform rather than uneven judgement — four teams writing the same retry library — the guild is the wrong instrument: build the thing once. If an architecture review board already exists, do not create a second queue reviewing the same designs; give the guild the teaching role instead and pair less experienced engineers onto the board's real cases.