A ranked list without a reason is still a list. You will not trust it when the day gets tight.
Explainable recommendation research finds explanations help calibrate trust — but fail when excessive, ignored, or encouraging blind trust. WAH’s answer: don’t explain everything. Explain the decision, with just enough evidence for you to audit it.
Decision evidence, not AI explanation
We keep Why this? as the user-facing name. Internally it is DecisionEvidence — not an LLM sentence generated after the fact.
type DecisionEvidence = { leverage: { unlocks: Task[], deadline?: Date }, fit: { effortMinutes: number, availableMinutes: number }, ownership: { blockedBy?: string, controllable: boolean }, context: { calendar?: Event[], messages?: Message[] } }
AI says: “Do this.” WAH says: “Do this — here’s enough evidence for you to decide whether I’m right.” That gap is trust calibration, not trust maximization.
The three evidences
Every Do this next renders the same evidence triad. Example: Review customer interview notes — 35 min · Highest leverage → Why this? · Unblocks Product Brief · Fits your 90-minute window · Nothing is waiting on you.
Chain it unlocks: Interview notes → Product brief → Launch decision → Friday. Window it fits: 35 min inside 90. Who owns the unblock: you, and no dependency is blocking you. Miss one, the move changes — and you can audit why.
Orange = decision backed by evidence
Orange must be supported by sufficient evidence — default is Leverage · Fit · Ownership — not a rigid three-line rule forever. Sometimes Why this? is Unblocks launch + Fits your 25-min window. Sometimes Deadline in 2 hours + You’re the only owner. If the reasoning is weak, WAH stays gray. That restraint is the feature.
Core principle: If any line feels off, fix the graph — not the sort. Bad recommendation → inspect the underlying state (dependency, deadline, ownership, context, goal, availability). Bad recommendation → fix the dependency / mark Legal done / change deadline / update ownership → graph changes → recommendation changes. WAH is a decision-debugging system, not a manual override.
The task was never the decision
Same task: Review interview notes. 90 min available → Do this next. 20 min available → Don’t start. Product brief blocked on it → Highest leverage. Waiting for teammate → You can wait. Launch tomorrow → Critical; launch postponed → Tomorrow. The task didn’t change. The context changed. That is where WAH’s moat starts.
Lists remember. Graphs connect. Context changes the answer. Evidence explains the decision. WAH tells you what to do next — Evidence → recommendation → human judgment → action. Orange is a decision backed by evidence, not decoration.
See Why this change →