Sequence

What to stop, keep, and build, in order — with the pro platform narrowed to non-clinical coaches, and the three life-stage contexts reinstated.

Condensed 8 September 2026, contexts reinstated 9 September · DiGA and certification remain out of scope by instruction

One rule instead of six

No feature may render a verdict on the person. The six non-negotiables are instances of this one rule; stating it once, positively, is more defensible in public and stops the product refusing things for reasons it no longer holds.

What to stop

Leading with “AI” anywhere a buyer is deciding — store metadata only. See what to call it.
Treating free-text logging as a differentiator. It is table stakes; the reference database is the differentiator.
The DiGA and certification track for the professional bridge — out of scope by instruction, not assessed and not rejected on merit.

What to keep, untouched

The three permission-negations: no judgement, no weighing, no searching. See softness.
The honest-tracker positioning and the database×AI architecture underneath it. See the tech.
The credit-based (not time-based) trial — a sourced answer to subscription fatigue, not just a mechanic.

The build order

Phase	What	Why first
1	Wire the daily tip at runtime; ship the EU AI Act Article 50 disclosure (already legally due).	Cheapest, already specified, one is a legal obligation in force since 2 August 2026.
2	Build and test the read-back — hand five users their own week and ask what they learned.	Everything downstream (coverage, the pro platform, the diary framing) assumes this works. If it doesn’t, nothing else matters yet.
3	The shareable weekly summary — user-initiated, no professional account.	Tests the pro-platform hypothesis at near-zero cost. See the pro platform.
4	If summaries get shared: the pro platform’s three screens (diary view, pattern view, settings view), gated behind an explicit client accept/decline step.	Only worth building once phase 3 shows real demand.
5	Android `shortcuts` / `share_target`.	Cheap, unused, no native-app conversation needed first.

Metrics that would actually tell you something

Read-back: do users report learning something from their own week, unprompted?
Sharing: what fraction of active users generate and share a summary at least once?
Coach interest: of shared summaries, how many recipients ask “can I see this regularly”?

When this would be wrong

If the read-back test fails, the honest-tracker positioning is a promise the product cannot yet keep, and the priority becomes fixing that rather than building anything on this page.