The decision schema I wish I'd built six months earlier

The pain it solves

I had three Struvo agents making decisions: roadmap-decider, pricing-checker, lead-router. Each one returned a different format. Each one tagged decisions differently. There was no audit log.

Result: I'd ask "did we approve that pricing change?" and have to grep three different agent logs to find the answer. Half the time the answer didn't exist because the agent had decided in chat and never written anything.

The fix

A shared schema. Every decision agent in Struvo now reads the same canonical contract at business-ops/products/struvo/01-canonical/DECISIONS-SCHEMA.md.

Every decision output has:

• verdict — one of GO, REVISE, DEFER, CUT
• reasoning — three to five sentences
• evidence — file paths or wiki entry IDs supporting the verdict
• risks — what could go wrong
• revisit_at — date when this should be re-evaluated (null for evergreen)

Every decision lands in DECISIONS-YYYY-MM.md — a monthly append-only log.

Why those four verdicts

GO: ship it. REVISE: not ready, here's what to change. DEFER: not now, here's when. CUT: never, here's why not.

Three was too few. Five was too many. Four felt right after about two weeks of use.

The verdict that surprised me with its value was DEFER. Half my decisions are "yes but later." Without DEFER, those landed as GO with a footnote (and got built early) or CUT (and got rebuilt later). DEFER captures "yes, when X is true."

What you can steal

If you have multiple agents making decisions:

1. Pick a verdict vocabulary. Three to five values. Write them down.
2. Pick an output format. Three to seven fields. Write them down.
3. Pick a log location. One file, append-only, dated.
4. Make every agent read the schema before responding.

Nothing in the schema is technically clever. The clever part is having ONE schema across agents instead of each agent inventing its own.

I lost months not having this. The fix took an afternoon.