Judge panel
The same question goes to several drafters who each see it differently. One judge compares the drafts topic by topic and writes down every choice with its reason.
Use it When a design has many decisions and I do not trust one draft to see all of them.
The problem
One architect writes one design. It is consistent, and it is blind where its author is blind. A design written with only the platform in mind forgets what a stolen token can do. A design written with only security in mind forgets that I want to save a link from my phone.
How it works
- One brief goes to three architects. Each writes a full draft from one point of view: platform, security and product.
- The three drafts go to a lead. The lead does not pick a winning draft. It lists the topics where the drafts disagree.
- For each topic the lead picks one option and writes down why.
- The result is one spec. The table of disagreements stays at the top of it, so anyone can see later what was chosen and what was turned down.
What it costs
Three drafts instead of one, and a fourth pass to merge them. Most of the three drafts say the same thing, so most of that work is spent three times. The value is in the part where they differ.
Where it fails
The judge can only choose between what the drafters wrote. If all three miss something, the panel misses it too. That is why the spec went through a second pattern afterwards, adversarial verify.
The lead can also pick by majority instead of by reason. Two drafts agreeing is a hint, not a proof. The rule that every choice needs a written reason is what keeps that honest.
Where I used it
For the spec of the AI lab. The three drafts disagreed on 26 topics. One example: the security draft wanted a fifth Worker for the scheduled jobs. The lead chose four, because a fifth would need its own binding to every private store, and that makes the private surface wider, not smaller.
Used inAI lab