What we will prove with you. On what timeline. With what counts as failure.

The first ninety days

Six milestones, in order, each with a named falsifier.

• Week 0: onboarding from your AI history.

  You drop your ChatGPT, Claude, and Gemini exports. Mandaire bootstraps from years of self-articulation you have already paid for. By the end of the week, the system knows your work, your network shape, your decision style. Without you having explained any of it. Then we add email, calendar, and one messaging path (typically iMessage or WhatsApp).

• Week 1: first daily brief.

  Morning, your time. The brief surfaces what changed overnight in your relationships, your inbox, your calendar; what is approaching that you have not noticed; what is decaying that you said matters. You read it and correct what is wrong. The corrections become rules. Most users find the first brief 60% useful; by Day 14 it is 85% useful, because the corrections have compounded.

• Day 14: first "about to drop" flag.

  Mandaire surfaces a near-miss: a thread, a relationship, or a commitment you would otherwise have missed. You either act on it or tell the system the flag was wrong. Either signal trains the model.

• Day 30: first refusal you are grateful for.

  The system holds a send, pauses a calendar invite, or queues an approval where you would otherwise have moved fast. You read the refusal note, agree with it (or override), and the refusal compounds in the refused-paths log. If you have not seen a refusal you are grateful for by Day 30, the system is being too compliant.

• Day 60: corrections changing autonomous behavior.

  Behavior the system would have taken in the first month is now visibly different because of corrections in the ledger. This is the compounding claim under live test. We point to the specific instance.

• Day 90: taste memory, and the predictability test.

  Mandaire produces your Day-90 taste memory. You read it. You decide whether the bullets describe decisions you would actually have made yourself, or whether they describe a stylized version of you that flatters but does not predict. The bet pays off if the answer is the former and fails if it is the latter.

What counts as failure

Five failure modes, named. If any fire by Day 30, we talk before Day 60.

• The daily brief is generic and you stop reading it after week 2.
• The system holds the wrong sends and lets the risky ones through.
• The surfacing is noisy. Too many "you might want to follow up with X" flags that you do not actually care about.
• The artifact trail is impressive but you never reach for it.
• You find yourself wanting a single-purpose tool (Superhuman, Reflect, Notion, Slack) for the specific thing and Mandaire only adds friction.

If any of the failure modes are firing at Day 30, we talk about it before Day 60. If Mandaire is failing the bet for your specific case, you should hear that from us, and you should be able to walk away with your data, your decision history, and your taste memory intact. That is what trust-and-openness over lock-in means in practice.