Kinedu app redesign

• Today becomes finishable: one 3-activity daily plan (~10 min) with a featured-next mechanic — replacing an endless feed where the plan, articles, classes and trackers all competed.
• The trial becomes visible and designed: day chip, accumulating value, and a day-5 recap timed before Apple's reminder — built for an opt-out trial (no fake "subscribe" ask; "see next week's plan" + annual upsell + honest charge-date line).
• Personalization starts at minute one: the baseline assessment moves into onboarding, so profile and pace exist on day 0 — instead of "Update milestones" begging and zero-line charts later.
• Progress leads with wins, not deficits: a pace band with a healthy zone (words first, multipliers one layer deeper), wins-first buckets, micro-milestone deltas — and percentile charts that never plot missing data as zero.
• Every completed activity feeds the data loop: the Done payoff asks one micro-milestone question — assessment as a by-product of play.
• The tracker talks back: age norms on every card, weekly pattern insights, caregiver sync promoted from a buried menu item to a visible card.
• Live classes get merchandised, not scheduled: age-filtered cards on Today, LIVE NOW banners, expert bios — instead of a month grid of anonymous dots.
• 4 tabs instead of 5: Today / Progress / Tracker / Explore, with Explore as one pill-driven catalog — Activities / Library / Masterclasses / Live — personalized by default inside each section, over 2,000+ activities, 1,000+ guides, 8 masterclasses and live classes.
• A principled free tier: logging and check-ins always free, the plan and the meaning premium — celebrations never locked, and every lock names the real value waiting behind it.

• The Today feed is a flat mall. Live class, plan, age explorer, tracker grid, articles, expert content, live classes again — equal weight, no finishable primary action.
• The progress system depends on manual data entry nobody does — and renders the absence of data as a terrifying flat-zero line under the population percentile curves.
• The tracker logs but never talks back. 8–10 daily logging moments, zero intelligence shown. Free competitors already do silent logging.
• The UI does not know a trial is happening. No day marker, no accumulating value story, no designed moment before Apple's day-5 charge reminder.

Job: answer "what do I do with my baby right now?" in one glance — and be finishable in ~10 minutes, so the day ends in completion, not scroll fatigue.

• One hero element: Today's plan (3 activities, ~10 min) with segment progress and a single featured activity. Featured = the next one up: completing it promotes the next activity to featured (in the prototype: Peek-a-boo done → "We can stand up!" featured → "Mirroring the sounds" promoted after). Other plan items stay accessible below.
• Trial chip ("Free trial — day 5 of 7") makes the week legible; tapping it opens the value recap. Premium replaces it with the week-rhythm card ("Week 4 with Kinedu · recap arrives Sunday"); free replaces it with the upgrade pill.
• "Why these" line under the plan ties activities to the pace check — the product thesis, restated daily.
• Pace snapshot, Tracker strip (2 live rows with intelligence microcopy), one live class for her age, one expert content card. Each module has a "View all" that routes to its full surface (Progress, Tracker, Explore) — modules are previews, not destinations.
• Removed from Today: age-explorer circles, baby-tracker icon grid, article carousels, duplicate live class section. All live in Explore now.

Job: get the parent from intent to playing in seconds — then turn the completed activity into a progress data point while the moment is still warm.

• The developmental benefit is the first line, not buried italic text ("Builds leg strength and balance — the next step in her gross motor pace").
• Meta chips up front: duration · materials · age band. "No materials" is a friction-killer and gets called out.
• Steps are 3 short numbered lines. Materials tab merged into chips + steps.
• A "works toward" card names the exact micro-milestone this activity practices.
• One big Done button. Save / Swap / Share demoted to text links ("Swap" replaces the ambiguous "Change").
• The payoff sheet is the key addition: celebration + plan progress + one question matched to the activity just played ("Did Emma pull up to stand today?"). Yes logs a micro-milestone instantly. This is assessment without the questionnaire: it keeps pace data fresh as a by-product of play, which solves the root problem of the old "Update milestones" begging.

Job: answer "is she okay — and is what we're doing working?" in a way that's emotionally survivable for an anxious parent, with every gap attached to an action.

• Pace of development is the headline, shown as a healthy-zone band (shaded green range + dot), not a 0–2x gauge. On a gauge, 1.0x renders half-filled — the best outcome looks like a C grade. On a band, "on track" sits visibly inside the healthy zone.
• Words first, multipliers in the detail layer. Calibrated scale: focus area / on track / ahead, plus early read when a domain has too few assessed milestones to deserve a verdict. Numbers under 1.0 read as a grade ("my child is at 60%"); superlatives above 1.0 ("remarkably advanced") set up regression-to-mean disappointment at the next check-in.
• Wins first: summary buckets ordered outstanding → on track → to practice ("to practice" replaces "Reinforce," which collides with the Today section name and leads with deficit).
• Micro-milestone skill rows (from the June 2026 Milestones release — the best new component): progress bar, real micro counts ("7 of 12"), "+2 this week" deltas. Skills with zero data show "not started · try the first activity," never counted as behind.
• "New milestones this week" celebration card; monthly PDF report promoted from an afterthought link to a card with a concrete job ("for your pediatrician visit").
• Check-in reframed: "Update check-in · takes 2 min." Science credibility footer: 830 milestones across 74 skills · Stanford · Harvard.

Job: show exactly where one skill stands and the next concrete thing to practice — with the percentile chart as opt-in depth for data-hungry parents, never the default read.

• Micro-milestone checklist kept (achieved-at dates, bulb→practice link per item) — flip the button prominence: Practice activities is primary, percentile view is a link. The current release makes the anxiety graph the solid blue button and the action the outline one.
• Percentile chart fix: the child's line starts at a labeled "first check-in" marker. The no-data region has no line. The current chart plots months of missing data as a flat zero under the population curves — a parent reads it as "catastrophically behind," then as "broken." Both kill trust during the exact week that decides conversion.

Job: let a parent find anything in a 3,000+-item catalog fast — personalized by default inside each section, but never competing with Today for the "curated for you" role.

• Architecture for a large catalog: search hero ("Search 3,000+ activities, guides and classes") + four content-type pills as the only selector — Activities / Library / Masterclasses / Live — with honest counts (Activities 2,000+, Library 1,000+, Masterclasses 8, Live schedule + recordings).
• "For [baby]" is not a pill — it's the default state inside every section. Activities opens filtered to her age band with real counts; Library opens age-targeted (285 of 849 guides match a 7-month-old); Live opens with her age band's classes first. Mixing a personalization lens with content types in one selector is a category error — and Today already is the cross-content personalized surface, so a curated segment in Explore would duplicate it. Segmented controls belong among peers; they do not belong on Today, where they'd force a mode choice before value.
• Library topic chips are age-relative. What an article library means changes with the baby: at 7 months the real catalog is dominated by feeding, sleep, health/growth and early reading; at 30 months it's behaviors, potty, language. The chip set should be derived from what actually exists for the child's age — not a fixed global taxonomy. (The API has a real cross-content domain taxonomy — sleep, feeding, reading, safety… — but it's null on ~90% of articles; see the taxonomy map in §17.)
• Masterclasses earn their own pill as a distinct content type, presented as series cards with real lessons and durations (Baby Sleep · Dr. Spencer) rather than anonymous tiles. In the free tier, the first lesson of each series plays as a preview — a natural freemium surface.
• Live classes get real merchandising one level down: named expert with credentials, "you'll learn" bullets, tie-back to the baby's focus area, recording promise for parents who can't attend live. The Live section opens filtered to the baby's age band.

Job: capture the baby's day in two taps and give meaning back — the daily habit anchor for 0–12m parents, and the bridge from logging to development insight.

• Logging mechanics kept as-is (the feed timer modal is solid). Added the intelligence layer: contextual norms on every card ("usually ~3h," "window ~2h"), pattern insight banner ("longest night stretch grew 40 min this week"), and a "why this matters at 7 months" explanation one level down.
• Caregiver sharing promoted from a profile dropdown to a visible card ("logs sync instantly" → "Pau logged 2 feeds today"). Shared data is shared lock-in.
• Empty states never render broken math (the audit found "Wake window: 504:17 hrs"). Every empty state proposes the action that fills it.
• Insights are rule-based v1 — wake-window math, feed-gap norms, week-over-week deltas. No ML required to ship.

Job: prove the app knows this specific baby within the first two minutes — then convert that proof into the trial start while perceived value is at its peak.

• The baseline check-in moves into onboarding so the app is personalized from minute one — this is the single highest-leverage build; everything downstream is empty without it.
• Results screen: celebration → pace band → per-domain rows (with "early read" guards on small samples) → handoff to action ("Language is where Emma gains the most — her first week's plan starts there", followed by the science fact for her next milestone) → "See her plan" = trial start. Classic quiz-to-paywall: perceived value peaks at the results moment; the current beta celebrates and stops.
• Credibility footer at the decision moment: Kinedu Skills framework · 830 milestones across 74 skills · Stanford & Harvard research. Longitudinal promise planted: "her pace re-checks June 19."

Job: make the accumulated week undeniable before Apple's reminder triggers the cancel decision — deepen commitment and offer the annual switch, never ask for a subscribe action that doesn't exist.

• New screen, timed before Apple's trial-ending reminder: value ledger (activities done, new milestones, tracker logs, classes), the week's biggest milestone with its science fact (including its safety guidance — expert voice at the decision moment), focus-area progress shown as logged micro-milestones ("Day 1: 2 of 9 → Today: 5 of 9" — observed behaviors, never a within-week "pace moved" claim; see §12), and "next week's plan is ready."
• Because the trial is opt-out, there is no subscribe action to ask for — the user is charged unless they actively cancel in Apple settings. A "keep your plan" button would imply a required action that doesn't exist. The recap's job is to make value undeniable so Apple's reminder doesn't convert into a cancellation.
• CTA stack: primary "See next week's plan" (commitment-deepening — engagement is the real cancellation preventer), secondary "Switch to annual · save 40%" (locks revenue before the charge date), plus an honest transparency line: "Your membership begins Thursday · $7/month · manage anytime." Honest pre-charge reminders reduce refunds, App Store complaints, and resentment churn.
• Micro-milestones make this screen honest: they genuinely move within 7 days, full milestones don't.

• H1 · Day-0 assessment in onboarding (quiz-to-paywall): results → plan handoff → trial start at peak perceived value; guarantees every trial begins with baseline data. Measure: results→trial-start rate; % of trials with baseline.
• H2 · Day-5 recap before Apple's reminder: honest value ledger so the reminder doesn't convert into a cancellation. The cleanest first A/B in the program. Measure: cancellations within 48 h of the reminder.
• H3 · Finishable Today (one 3-activity plan): a completable daily program reads as "something my baby is enrolled in," not a cancellable content app. Measure: D1–D3 plan completion and its correlation with conversion — likely the best leading indicator.

• H4 · Done→micro-milestone payoff loop: assessment as a by-product of play; micro-milestones genuinely move within 7 days, so the trial shows honest progress. Measure: % of Dones answered; micro-milestones per trial user vs. conversion.
• H5 · Tracker that talks back: 8–10 daily logging moments return intelligence instead of silence — the bridge from habit to what only Kinedu sells. Measure: tracker days/week vs. conversion.
• H6 · Anxiety-safe pace framing: wins-first, healthy-zone band, words before multipliers — the validation paper itself flags parental anxiety as a design risk. Measure: Progress repeat visits; check-in repetition.

• H7 · Caregiver invite in the week-1 journey: shared data is shared lock-in; cancellation becomes a two-person decision. Measure: 2-caregiver vs. 1-caregiver trial conversion.
• H8 · Science facts at moments of belief: 417 already-written trust injections at the payoff, assessment result and recap.
• H9 · One job per tab / Explore as a clean catalog: the enabler that makes H3 possible — Today can't shed the content mall until Explore can absorb it.
• H10 · Free tier as re-conversion engine: logging stays free and accumulating; every lock names value the user created.

• The live schema is snake_case (request_auth_token), while internal docs show camelCase (requestAuthToken) — doc drift in both directions.
• The gateway returns 503 without the expected User-Agent; with it, everything routes fine.
• Multiple root fields in one request fail with a SQLAlchemy "concurrent operations are not permitted" error — one query per request is the working rule (it bit the milestones integration too) — nested resolution would be far more usable than parallel root fields.
• The documented nested locales { ... } field on activities does not exist in the live schema — activities and locales require two pulls joined on activity_id.
• Tokens are single-session, ~15 minutes — fine for snapshot pulls, worth remembering for any live dashboard.
• The skills base query is broken: the resolver SELECTs the full model regardless of requested fields, and the DB lacks skills.illustration — so every skills query fails with "Unknown column" while skill_locales works fine. Until fixed, skill metadata (area, age ranges, official milestone counts) is unreachable; the prototype reconstructs it from the milestones themselves.

• Delivered: skill_locales (84 skill titles), milestones (830 active: 212 master + 618 micro across 74 skills — matching the new release's micro-milestone structure exactly), milestone_locales with science_fact on 417 milestones. Those science facts are a sleeping credibility asset — one line turns "trust us" into "here's the developmental science." In the prototype they now deploy at four moments of belief: the Done payoff (reframing the behavior just practiced — "dropping things is hypothesis testing"), the day-0 assessment results (under the focus area, before the trial ask), the day-5 recap (attached to the week's biggest milestone, including its safety guidance), and the skill detail. Principle: one fact, one line, always tied to the specific milestone in view — and visible in the free tier, where generosity earns trust.
• Blocked: the skills base query (see the resolver bug above) — area mapping, official age ranges and counts per skill wait on that fix.
• Still pending — the activity↔milestone link: it powers "works toward: …" on every activity detail and the Done-payoff question. activities_indicators_count already exists on ActivityType, so the join table exists; exposing activity_indicators (activity_id, indicator_id) is likely all it takes.
• Catalog only — no per-child data needed for design work.
• Article categories: the domain taxonomy below is the right topic system for the Library, but it's null on ~90% of articles (240 of the 285 age-matched ones). The named collections shown in the app ("Building a Healthy Brain", "Challenging Behaviors"…) live in the CMS, not the API. Request: backfill domain_id on articles (or expose the CMS collection mapping) — age-relative Library navigation depends on it.