launch-retrolisted

# Launch Retro Lessons with receipts. Every lesson names the event it came from, and decisions get judged on what was knowable — not on how the dice landed. ## Verification gates (defined first; output is blocked until all pass) - **G1 — Observed-event citation:** every lesson cites the specific event(s) or number(s) it derives from. "We should communicate better" with no event fails; "39 of 61 tickets were nav confusion → the tooltip cut cost us" passes. - **G2 — Decision ≠ outcome:** each logged decision is graded twice — decision quality (given what was knowable then) and outcome quality (what happened) — and the grades are independent. A good decision with a bad outcome stays a good decision; a good outcome never retroactively blesses the decision that got lucky. Judging one by the other fails the gate. - **G3 — Open items surface:** undiagnosed near-misses and unexplained numbers appear as OPEN RISKS with owners/actions — a retro that only processes closed items is incomplete. No invented events, sentiments, or dynamics. ## Steps 1. **Build the event ledger:** every event, number-vs-assumption, ticket cluster, decision, and near-miss from the input, each with its figure. This ledger is the citation universe for every lesson. 2. **Grade each logged decision twice.** Decision: was it right given the information available at the time? Outcome: what actually happened? Write both grades with their basis. The four quadrants all occur — good/good (staged rollout catches th