Decisions, risks and next evidence
| Product decision | Reason | Tradeoff / next evidence |
|---|---|---|
| Return to a saved unfinished goal | Gives an invitation a concrete value and direct next step | Goal may no longer matter; ask learners whether it remains relevant |
| Require both consent and preferred channel | Permission does not mean every permitted channel feels welcome | Strict preference suppresses some otherwise permitted prompts; compare preference comprehension |
| Count attention across channels | Switching channels must not bypass frequency limits | Real deployments need unified consent and reliable cross-device state; none is integrated here |
| Preserve accepted snapshots and suppressed attempts | Avoid retroactively changing the meaning of a reviewed run | An exact attempt cannot be re-evaluated under a new rule; new times remain available |
| Keep strongest accepted run spacing/ceiling | Relaxing policy must not erase prior attention protections | Conservative rule can suppress more than the latest policy alone; test whether reviewers understand this |
| Offer reviewed reset with bounded Undo | Mistakes can be repaired within an explicit scope | Reset begins a fresh fictional scenario; not a production consent mechanism |
| Preserve corrupt/unreadable bytes | Recovery must not silently erase an unseen decision | Memory can be lost on refresh; warn and require a readable reviewed reset before replacement |
| Use fixed local offsets and a fixed clock | Reproducible timing and hand calculations | Does not model DST, timezone changes, actual delivery or calendar weeks |
| Bound histories at 30 runs and30 policies | Keep storage and validation finite | Extended exploration requires an explicit sample reset |
The principal rejected alternative is a broad generic dormant-user blast. Volume is easy to demonstrate but does not establish that returning serves the learner. This prototype includes no delivery infrastructure, tracking, personal identities, live AI, real campaigns or persuasion optimization.
Mo owns product and program direction; AI assisted implementation and verification. The fixture decisions are provisional. Proposed research must ask about re-entry usefulness, explanations of suppression, channel preference and fatigue. No participant sessions, customer endorsement, conversion lift or employer outcome is represented as evidence.