Risk matrices compare options against each other; this one anchors a single proposal in how comparable workThe 'outside view' (Kahneman & Lovallo): instead of estimating from your plan's specifics, ask how often comparable past work has landed. The unknowns you can't enumerate are already priced in. has historically landed'Landing' isn't one universal bar — each reference class names its own (ships without breaking, still in use six months later, old system retired…). The criterion line under the blurb tightens it per class. — a base rate you can adjust for what you know about your own situation.
Comparable work — new feature in a known product — lands about 70% of the time, anchored to DORA tiers; software-estimation surveys. My situation shifts that by ±0pp to about 70% — leans toward landing. The 90% credible band is 51%–84%. The band widens as your situation drifts from the calibrated middle.
Building inside a product you know — a screen, an endpoint, a flow. Most ship; the misses tend to be scope, not catastrophe.
Starting from 70% — new feature in a known product
| Lever | Choice | Effect | Δpp |
|---|---|---|---|
| Team experience | Some experience | ±0pp | |
| Novelty of approach | Adapted pattern | ±0pp | |
| External dependencies | A few | ±0pp | |
| Rollout strategy | Phased | ±0pp |