The half-life of a decision
Most decisions lose their reasoning within eleven weeks. We measured it across 40 workspaces and found the pattern is depressingly consistent — and fixable.
Blog
Long-form thinking on decision records, context graphs and the quiet cost of institutional forgetting.
Most decisions lose their reasoning within eleven weeks. We measured it across 40 workspaces and found the pattern is depressingly consistent — and fixable.
Every wiki starts strong and decays. The failure is structural: writing is a second job. Connection beats capture, and here is the data behind that claim.
Node-link diagrams are usually decoration. We rebuilt ours around influence weighting so that the layout itself carries information.
If your metrics, decisions and evidence already live in one place, the monthly update writes itself. Here is the template we use internally.
We tracked how long a decision keeps its reasoning attached across 40 workspaces. After eleven weeks, most teams can still name the outcome but no longer reconstruct why. The decay curve is remarkably consistent — and it flattens completely once the evidence is linked.
Sample: 40 workspaces, 1,860 tracked decisions.
Recall of a decision's reasoning
Percentage of decisions whose evidence can still be found, by week.
Every wiki starts strong and decays, because writing documentation is a second job that competes with the first one. Editing activity peaks in month one and never recovers — while the volume of real context being produced keeps climbing.
Wiki edits per week
Typical trajectory after a documentation push.
Context produced elsewhere
Connections captured over the same period.
Reader favourites
Ranked by completion rate rather than clicks, which is the only engagement number we trust.
Research
Eleven weeks, measured across 40 workspaces.
Essay
Connection beats capture, with the edit curves to prove it.
Design
Influence weighting turns decoration into information.
Playbook
The template we use internally, generated from the graph.
Signals we watch
Every claim we publish comes from a live panel in our own workspace. Here are three of them.
Documentation coverage is falling behind
Coverage drops long before anyone notices.
Engineering velocity improved
What happens once context stops going missing.
Support load is shifting to billing
A trend spotted six weeks before the churn number moved.
The research, the charts and the drafts all live in the same workspace. Start your 15-day trial and write your next update the same way.
Company memory, decision archaeology, connector engineering and what we learn from real workspaces — with the numbers behind each claim.
Roughly every other week. Product changes are published separately on the changelog.
Yes, with attribution and a link back. Reach out through the contact page for anything more involved.