Describe a decision in five clicks. The tool returns the protocol it actually deserves: who decides, how fast, what you write down, and what you owe the people downstream. Reversible decisions should get less process, not more meetings.
An opinionated heuristic, not a policy. It's here to make the argument specific: you'll usually end up arguing about one chip, which is the point.
Decisions aren't the sum of clean independent factors. Reversibility on its own is irrelevant if nobody feels the decision, and reach on its own isn't a reason to slow down if the call is cheap to undo. So the model checks a handful of categorical gates first, and only does arithmetic second. Nothing here is a weighted average.
Blast radius, precedent, coupling, and information, plus reversal cost times blast radius. A sum plus a product, never divided by anything, so no axis can average another one down. Range 0–25.
Blast radius is the only axis that appears twice, because it's what makes every other axis matter. Reversal cost counts only as a multiplier of blast radius: an irreversible decision nobody feels is free, and a trivially reversible decision that reaches everyone still deserves a heads-up, not a committee. Addition would treat those as the same increment.
Because the four additive axes max out at 13, the One-Way Door tier (17+) is unreachable without meaningful reach riding on top of meaningful reversal cost. That comes from the shape of the formula, not from a tuned threshold.
On top of the score, five gates set floors the score can exceed but never undercut. The loud-customer exception stays Deliberate even when its arithmetic says Decide and Tell. Gates only raise; no gate can lower a tier the score earned.
| Gate | Condition | Effect |
|---|---|---|
| G1 | Blast radius ≥ 1 | Floor: Decide and Tell (someone outside your team is downstream) |
| G2 | Precedent = 3 | Floor: Deliberate (customers will read this as a statement about how you treat them) |
| G3 | Coupling = 3 | Floor: Deliberate (people will build on this for a year) |
| G4 | Information = 3 and Reversal cost ≥ 2 | Floor: Deliberate (a cheap test exists and unwinding is expensive) |
| G5 | Reversal cost = 3 and Blast radius ≥ 2 | Floor: One-Way Door (irreversible and customer-facing) |
| Tier | Score range | Decider |
|---|---|---|
| Just Decide | 0–4 | Whoever is doing the work |
| Decide and Tell | 5–10 | Still one owner, no committee |
| Deliberate | 11–16 | One named decider, roles named out loud |
| One-Way Door | 17–25 | The accountable executive, named in writing |
An average lets a low axis pull the result down: 'it's urgent' and 'it's easy to roll back' would cancel a trust problem. Here they can only lower the score, never a floor. B appears in both terms because reach is what makes every other axis matter; R appears only as a multiplier of B because reversal cost in isolation isn't risk (an irreversible decision nobody feels is free). That asymmetry is the whole Type 1/Type 2 door insight.
max(B+T+K+L) = 13 < 17, so the top tier is unreachable unless R×B ≥ 4. With R = 0 the score caps at 13 and no One-Way gate can fire. Verified by enumerating all 1,280 answer combinations: zero exceptions. A genuinely free-to-reverse decision can never land at One-Way Door by any path, which is the point.
It costs nothing to reverse and touches one customer. It still isn't a two-way door, because you just told everyone what it takes to get an exception. The gate on trust fires regardless of the score. A pure weighted model gets this exactly wrong, and that preset exists to prove the gates are load-bearing.
Urgency is folded into the information question ('the window closes first') where it can lower the score. A standalone urgency axis is the dilution attack this structure exists to prevent: urgency can keep rigor cheap, but it can never open a gate: 51 of the 1,280 combinations are urgent and still land at One-Way Door.
A decision reversible in a sprint but maximal on every other axis scores exactly 17, One-Way Door. In practice R=1 with K=3 is close to self-contradictory (if people build on it for a year, unwinding isn't a sprint), and when the model errs, it errs toward more rigor. That is the intended failure direction.
98 of 1,280 combinations, and every one of them has blast radius zero. If it only touches your team, stop asking. If anyone else is downstream, the floor is Decide and Tell: you still decide alone, you just don't decide silently.