Most decisions fail not because the choice was wrong, but because the reasoning was thin, the assumptions were invisible, and nothing was written down. Decision Memory is a structured process for thinking well — and for remembering how you thought.
Don't organize your thoughts before you start. Paste in everything — messy, incomplete, contradictory. The app reads your brain dump and structures it into a decision draft: options, criteria, assumptions, stakes, reversibility, and what your gut is saying.
Most decision tools ask you to fill out a form before they help you. That's backwards. Decision Memory structures your thinking for you, so the cognitive help starts immediately.
The app classifies your decision — is it routine, complex, or urgent? — then picks the single best framework for it. You can run second opinions from 17 others. Each framework gives you a recommendation, a confidence score, and the one assumption your answer most depends on.
Frameworks include: Weighted criteria, Pre-mortem, Expected value, Regret minimisation, Steelman + red team, Outside view, Bayesian update, Second-order thinking, and 10 more. The system never applies the same lens to every decision.
Before you commit, open the debate panel and push back. The AI has read your full analysis and its job is to challenge it — find the weak assumption, ask the uncomfortable question, argue the other side. It changes its position when you make a good point.
This is the step most tools skip. A recommendation that hasn't been challenged isn't a recommendation — it's a first draft.
When you run analysis, the app retrieves decisions you've made before that are semantically similar to this one. Not keyword matches — genuine conceptual similarity, using vector embeddings. A career decision from three years ago surfaces when you're facing a new one, even if the words are different.
Only decisions you've locked appear in memory retrieval. Drafts are excluded. You control which decisions can influence future analyses — and you can opt any decision out.
When you're done, you lock the decision. Not save — lock. Locking captures the complete reasoning record: your rationale, the options you considered and rejected, every framework you ran, every assumption you were making, and the evidence available at the time.
Two years from now, when you second-guess this choice, you'll be able to open this record and see exactly what you were working with. The second-guessing usually dissolves. The decision looks different when you remember you were reasoning under genuine uncertainty.
Review dates are auto-scheduled when you lock a decision — 14 days for routine choices, 30 and 90 days for high-stakes or irreversible ones. The review separates decision quality from outcome quality. These are not the same thing.
A decision can be made carefully and still produce a bad outcome. Conflating the two teaches you the wrong lessons. The review scores both independently, and asks what was genuinely unknowable at the time — the question that dissolves hindsight regret.
After enough decisions and reviews, the Patterns page becomes a mirror. It shows which frameworks you never use, where you reach high confidence before adequate challenge, and which reasoning steps you consistently skip.
The AI also generates personal heuristics from your completed reviews: patterns specific to you, with the decisions that produced them as evidence. "You consistently underestimate timelines on technical projects." "You revisit locked decisions when anxious, not when evidence changes."