Owner ↔ AI Decision Tiers

Not every decision that needs a human's judgment has the same shape, so this project uses three different mechanisms for surfacing one to the owner — picked by matching the shape of the decision, not by defaulting to whichever one is fanciest or most novel.

Direct question

The default for one decision, or a handful asked one at a time. A concrete finding or reason is stated, not a vague "what do you want to do?" — along with a recommended option and the reasoning behind it. Exactly one question per distinct decision point, not several batched together into one wall of choices.

Tracked interview

For a small number of genuinely deep architectural questions worth tracking across multiple sessions rather than resolving in one sitting — a persistent, turn-by-turn document accumulates question, answer, and rationale as the design gets progressively decided, rather than trying to hold it all in one conversation's working memory.

Published review board

For bulk or parallel review — tens to hundreds of items, each needing its own adjacent visual or contextual material that a single chat turn can't render well (an image, a code diff, an audio take). A self-contained page presents each item as its own card, with a live tally and the ability to act on many at once.

None of the three replace the others

Picking the tier is a judgment call about the shape of the decision in front of it — not a ladder where one tier is simply an upgrade from the last.

The SDLC, Paper 6: Reasoned Tool Selection documents this table as one of two named decision axes this project runs on — the other being Loop Tiers — and includes a real worked example of picking the wrong-shaped tool before catching it.