# Decision memory: keep the reasoning with the decision Build a decision record that connects the final call to its reasons, evidence and review trigger, so the next person can check how you got there. By the Zeke team - Updated 2026-09-16 - 5 min read Original guide: [Decision memory: keep the reasoning with the decision](https://zeke.so/guides/decision-memory) A task says 'remove the weekly CSV export.' The work is finished. A month later, someone asks for the export back, and the task tells you almost nothing about why it went away. Was it unreliable? Did a customer stop using it? Was another report supposed to replace it? The decision may have been sensible. The record is too thin to tell. Reopening the task now means finding whoever remembers the conversation, locating the evidence they used and checking whether that evidence still holds. Decision memory is the reasoning you can recover from that record. A document archive helps you find material. A useful decision record tells you which material supported the call, who made it and what would justify revisiting it. You can start with one paragraph and a few links in the place your team already works. ## Keep the chain from decision to evidence intact Use one record to hold the decision, its reason, the supporting evidence, the responsible person and the condition for review. These fields belong together. A reason without a source is hard to check; a source without a stated decision leaves the reader to interpret the whole conversation again. ### Decision record: retiring a weekly export For a decision to retire an export, fill the record like this. Replace source labels with links and the sample details with what your team actually agreed. **Decision and status** Retire the weekly CSV export after the dashboard replacement is accepted. Decided; removal is waiting on that acceptance. Keep the export available until then. **Reason and alternative** The export repeats information available in the dashboard, but someone still prepares and checks it each week. We considered keeping both. We chose retirement only if the dashboard covers the fields people still use. **Evidence** Field comparison: dashboard and CSV columns. Reader review: confirmation that the replacement covers the required fields. Decision note: the discussion accepting retirement subject to that review. The task completion alone does not establish these facts. **Owner and consequence** The reporting owner accepts the replacement. The operations owner removes the export job afterward and updates the report index so readers can find the dashboard. **Review trigger** Revisit if a required field is missing, the dashboard becomes inaccessible to a reader or a new reporting obligation needs an export. Add the new evidence to this record before changing the decision. Notice the condition in the first line. 'Retire the export' would erase an important distinction between agreeing on the direction and being ready to act. A decision can be settled while its implementation is still waiting on evidence. Preserve that distinction in the summary. ## Link the evidence to the claim it supports A link to a crowded project folder is a starting point, not evidence for a particular sentence. Point the reader to the comparison, comment or approved note that supports the reason. Give the source a useful label and enough context to locate the relevant passage. If access is restricted, say who can provide the record; do not copy private material into a broader audience's document. Separate what a source says from the conclusion the team drew. A field comparison can show that the dashboard contains the CSV's columns. It cannot prove that every reader has accepted the replacement. That second claim needs its own evidence. If nobody recorded acceptance, write 'acceptance not yet confirmed' rather than treating an unanswered message as consent. Conflicting records need visible handling. If a planning note says to keep the export but a later approved decision retires it, mark the earlier note as superseded and link to the new call. Do not delete the earlier reasoning simply because it is inconvenient. A future reader may need to understand what changed. ## Update the part that changed Write the first record while the discussion is still fresh. Put it beside the task or project plan people will return to, and name someone who can correct it. A separate decision repository is optional. Finding the accepted record from the normal workflow matters more than choosing a new home for it. When a condition changes, append the change, its date and its source. In the export example, acceptance of the dashboard changes the implementation status; it does not rewrite why retirement was chosen. If a new requirement reverses the decision, preserve the original call and link the replacement. This keeps a search result from making yesterday's answer look current. Then give the record to someone who missed the discussion. See whether they can recover the reason, evidence and next step without a private explanation. [Use the 10-minute memory test.](https://zeke.so/blog/company-memory-problem) ## Give an AI assistant something it can check An assistant can gather available notes and draft this record. Ask it to attach a source to each material claim and leave missing reasons unresolved. The person who made the decision reviews the interpretation. A fluent summary is not evidence that the team agreed to everything in it. Zeke can work with connected Google Drive documents after they have been scanned. That can help recover a decision and prepare a sourced brief. It does not mean the entire Slack archive has been captured or that Drive is continuously synced. Updated documents need a new scan. Zeke cannot recover a rationale nobody recorded; that gap needs a person, not a more confident answer. When the export request comes back, the next reader should be able to see the original reason and judge the new evidence. They may still reverse the decision. At least they will know what they are reversing. ## How much work goes into finding the reason again? Use your own estimates of context searches and interruptions in the coordination cost calculator. Treat the result as a way to choose a workflow to examine, not as savings Zeke has demonstrated. [Open the coordination cost calculator](https://zeke.so/tools/coordination-cost-calculator) ## Bring one decision whose reasoning is hard to recover Spend 15 minutes with the founders tracing the sources an answer would need and where a decision record could help. 15 minutes with the founders. Bring one recent example. [Book a demo with the founders](https://calendar.google.com/calendar/appointments/schedules/AcZssZ20_GeldbmSN2zDe3SUIdCOy56OFGvciCn691y6IYQCbTCUpFEP5oa5_nIj4gwNjBIMHdpPMN0R?gv=true) [Tell us about the workflow](mailto:founders@triadlabs.ai?subject=Zeke%20workflow%3A%20Decision%20memory%20guide%20%C2%B7%20Finding%20past%20decisions&body=I%20came%20from%20the%20decision%20memory%20guide.%0AWorkflow%3A%20Finding%20past%20decisions%0A%0AThe%20last%20time%20this%20happened%3A%0A%0AHow%20we%20handle%20it%20today%3A%0A%0AWhat%20gets%20delayed%20or%20harder%3A%0A) The email link opens a draft in your email app. ## Continue with the work - [How to write an AI meeting brief people can check](https://zeke.so/guides/ai-meeting-brief): Use the record to prepare the next decision. - [Follow-up automation that knows when to stop](https://zeke.so/guides/follow-up-automation): Turn an accepted next step into a bounded follow-up. [All working guides](https://zeke.so/guides)