
The reasons lost even when a decision is written down
Writing down a decision does not necessarily preserve its reasons. What context does someone inheriting it need in order to maintain, review or replace it on defensible grounds?
Minutes can state precisely what was approved while leaving the difficulty of understanding it almost untouched. The date, the people responsible and the instruction survive; the problem it was meant to solve may remain only in conversations nobody thought would need to be preserved. Whoever arrives later finds a decision still in force, but not always the means to distinguish what remains a constraint from what has become a habit.
Reconstructing that context once the outcome is known presents a further difficulty. In Baruch Fischhoff’s experiments on hindsight judgment (opens in a new tab), published in 1975, presenting an outcome of past events as true led participants to consider it more foreseeable and changed the relevance they assigned to prior information. That finding does not prove that writing down reasons eliminates the bias; it does require us to distrust the ease with which we can explain them afterward. A record constructed only at the end risks preserving what now seems reasonable for us to have thought.
What a decision took for granted
Consider a hypothetical situation: a company centralizes purchasing approvals while changing its management system. During the transition, it needs to control errors it does not yet know how to detect in another way. A year later, the system is stable, but approvals still follow the same path. The record says who approved centralization; it does not explain that it responded to a temporary vulnerability. To a new employee, the process may look like a permanent control requirement. To someone else, unjustified bureaucracy. Neither yet has enough context to interpret what they inherited.
Recovering the original reason allows a different question: does the condition that made this control necessary still exist? The answer might justify removing it, but does not settle the matter on its own. During that year, the process may have begun to serve other functions, or new obligations may have emerged. A review requires examining those current dependencies. Understanding why a practice began helps us discuss whether it should continue without assuming that exactly the same reasons sustain it today.
NASA’s decision analysis guidance (opens in a new tab) makes that relationship between a choice and its basis explicit: the information to be retained includes context, alternatives considered, assumptions, uncertainties and the rationale for the final decision. It is a systems engineering reference whose formality reflects that setting. What matters for this management discussion is something narrower: whoever receives a choice should be able to reconstruct what tipped the comparison and what change could make it less defensible.
What we are willing to accept can also change, even if the facts remain the same. Maintaining a second supplier may once have seemed an unjustifiable expense; today we may accept that cost to depend less on a single one, even without changes in its service conditions. Recognizing this avoids pretending that the earlier choice was necessarily a mistake or that the new one merely corrects a fact. Organizational memory needs to keep both the constraints and the criteria used to decide among them visible. Some differences between two decisions express a change in priorities, not a loss of consistency.
Preserving the possibility of change as well
If reasons are recorded as a conclusive defense, the record can become another obstacle to review. Michael Nygard’s proposal for documenting software architecture decisions (opens in a new tab), published in 2011, offers a simple solution to part of that problem: when a decision is replaced, the earlier record is retained but marked as superseded and linked to the new one. The past remains available without being confused with the current instruction. Its usefulness here does not depend on adopting the format: an organization needs to be able to remember what it stopped doing without continuing to order that it be done.

That care does not require recording every conversation or retaining every version of a disagreement. What matters is keeping anything whose absence could change the interpretation: an alternative rejected because of a temporary condition, an uncertainty left open or a relevant objection that approval did not resolve. A final decision does not retrospectively turn participants’ doubts into settled matters. Nor does preserving them mean leaving the decision indefinitely suspended.
When that context has already been lost, we should not fill it in with a plausible explanation presented as fact. We can acknowledge that the original reason is unverified and assess whether the practice is justified by current information. That allows a decision without manufacturing continuity. The reason we maintain a rule today may differ from the reason it was created; recording that difference is more honest than attributing foresight to its authors when we do not know whether they had it.
When learning from a successful project, we try to recognize what deserves to stay with us. A further requirement emerges here: that understanding must remain open to discussion when the people who can explain it are no longer there. The record needs to be accessible to whoever will review the decision, and the organization must allow a change in conditions to reach someone with the authority to respond. There is little point in preserving a premise if noticing that it no longer holds cannot initiate any review.
Continuity does not require every new person to decide everything again. It requires them to recognize what they are sustaining when they maintain a choice and what they would have to justify to change it. Preserving the reasons should give them that grounded freedom. If all we can pass on is that “this has already been decided,” memory ends exactly where obedience begins.





