Diyor's Test Assignment

Redesigning the Novakid Parent Dashboard as a Value Hub

Readme

  1. 1 Reading the brief
    The assignment, the ten questions it raises, and what I cut from scope.
  2. 2 The audit
    A hands-on walk through the current Novakid parent account — 23 screenshots, six audit lenses. Four direct competitors, 86 screenshots, one trial lesson completed at Cambly Kids.
  3. 3 Voice of the parent
    128 verified parent quotes from 15 platforms with a four-tier authenticity audit. Nine theoretical frameworks. Eight persona interviews. A seven-market cultural analysis.
  4. 4 Thesis
    The reframe: progress is a translation problem, not a measurement one. Four dimensions of progress. A rough business-math sketch.
  5. 5 Hypothesis Atlas
    35 hypotheses grouped by what they bridge. Each one backed by the quote corpus, theory, a persona, and a reference pattern from a known product.
  6. 6 Design Surfaces
    Eight product areas that absorb the 35 hypotheses — each one a screen or channel the parent actually sees.
  7. 7 IA
    Current-state sitemap vs proposed Value Hub architecture. Customer journey moved to §08.
  8. 8 CJM
    Six-stage parent journey. Coverage matrix (8 surfaces × 6 stages) then per-stage deep dive: today's pain, the Value Hub intervention, research anchor per stage.
  9. 9 Hi-Fidelity
    Three surfaces (Home, Post-Lesson, Plateau) — Primary plus concept variants (different approaches, not style dressing). Design system + decisions log live on adjacent Figma pages.
  10. 10 Live prototype
    A working, keyboard-navigable web prototype. Five surfaces × five markets × eight weeks of replay. Hold ? to see the research quote behind every decision. novakid-prototype.pages.dev →

01 Reading the brief

Objective

Redesign the Parent Dashboard by integrating Progress Reporting into the core experience. Move beyond a functional utility and design a "Value Hub" that demonstrates a child's growth at a glance.

The task

  1. Explore: log in to Novakid, audit the parent experience, find where the value is hidden and where friction lives.
  2. Reimagine: build a dashboard that is both a management tool and a living progress report.
  3. Solve: how should teacher feedback, curriculum milestones, and quick actions (booking, billing) coexist to prove the product pays for itself?
Original brief (PDF)

Questions the brief raises

Each question links to the section where it is addressed.

Q1
Where is the value of lessons hidden and where does friction exist?
§02 audit synthesis
Q2
How do 2–3 competitor platforms solve parent progress visibility?
§02 competitors
Q3
What information architecture puts progress, teacher feedback, and quick actions on the same screen?
§07 future IA §06 surfaces
Q4
What does "progress" mean to a parent paying €9–20 per lesson — and does Novakid currently show it?
§03 patterns §04 four dimensions
Q5
What design decisions support parent engagement and why?
§06 surfaces §05 atlas
Q6
Can audit findings and design proposals be tied to measurable business outcomes (churn, retention, LTV, NPS)?
§04 business math §05 atlas
Q7
What triggers a parent to open the dashboard: push notification, email, billing anxiety, or the child?
§06 weekly digest
Q8
How should the dashboard represent a learning plateau — and communicate that stalled progress is a normal part of language acquisition?
§06 plateau protocol
Q9
How does the dashboard evolve from 0 lessons to 40 lessons — does it grow in value or stay static?
§08 CJM
Q10
In what format should teacher feedback reach the parent?
§05 Bridge 2 — Teacher

02 The audit

Novakid Platform Audit

heuristic-evaluation.mdformal-ux-audits.mdtrial-lesson-notes.md23 screenshots

I registered a trial account on Novakid, completed a real trial lesson, and screenshotted every parent-facing screen (23 screenshots). Then I ran four formal audits: a 6-lens heuristic evaluation6-lens heuristic evaluationA custom rubric scoring the dashboard on six lenses: value communication, information hierarchy, progress visibility, emotional tone, actionability, and data density., Nielsen's 10 heuristicsNielsen's 10 Usability HeuristicsJakob Nielsen's classic ten checks — visibility, match with the real world, user control, consistency, error prevention, recognition over recall, flexibility, minimalist design, error recovery, and help., Don Norman's 7 principlesDon Norman's 7 Design PrinciplesFrom The Design of Everyday Things — discoverability, feedback, conceptual model, affordances, signifiers, mappings, constraints., and a Cognitive WalkthroughCognitive WalkthroughA task-based method that walks a first-time user through each step, asking: will they try the right action, will they notice it, will they understand it, will they see progress?.

Nielsen's 10 Usability Heuristics (2.4/5) Visibility of system status: 1/5 — no learning data on the main dashboard +
Heuristic Score Root problem
H1. Visibility of system status 1/5H1. Visibility of System Status — 1/5Dashboard shows zero learning data after a completed lesson. 16 skill badges appear visually identical with no mastery distinction. Program tab has no "you are here" indicator. No learning progress data on main dashboard
H2. Match between system and real world 2/5H2. Match Between System and Real World — 2/5Skills labeled in pedagogical jargon ("speak about objects using 'it's'") instead of parent language. Curriculum codes like "Lv0.Un1.Ls5" are unparseable. Help center defaults to Turkish despite English UI. Curriculum codes and billing language instead of parent language
H3. User control and freedom 3/5H3. User Control and Freedom — 3/5Clicking "Achievements" switches to a different nav structure with no breadcrumb or visual transition. Subscription page has no save-draft or back capability for 4 simultaneous decisions. Context switch between parent dashboard and Achievements section
H4. Consistency and standards 2/5H4. Consistency and Standards — 2/5Parent nav and Achievements nav are structurally different (items appear/disappear). Three empty-state patterns across pages. Four distinct visual languages: purple cards, cream badges, plain text lists, cartoon Game World. Two navigation systems, three empty-state treatments, four visual languages
H5. Error prevention 3/5H5. Error Prevention — 3/5Subscription page shows prices with "$" symbol but in Turkish Lira range, risking purchase confusion. Promo code field has no format hints or validation. Limited action surface naturally reduces errors elsewhere. Subscription page requires 4 simultaneous decisions, currency symbol confusion
H6. Recognition rather than recall 2/5H6. Recognition Rather Than Recall — 2/5Dashboard shows zero context about learning state — a returning parent must recall from memory what the child studied. 16 skill badges have no visual distinction. Library has 25+ categories with no level-based personalization. Dashboard shows zero context about child's learning state
H7. Flexibility and efficiency 2/5H7. Flexibility and Efficiency of Use — 2/5No customizable dashboard, no keyboard shortcuts, no quick rebook. A 12-month user sees the same interface as a day-one trial user. Library has no search or level filter. No customization, no accelerators, no progressive disclosure
H8. Aesthetic and minimalist design 3/5H8. Aesthetic and Minimalist Design — 3/5Dashboard has room for progress data but fills it with promotional notifications. Schedule page: one lesson entry uses 10% of viewport, rest is whitespace. Empty states are blank pages with one line of text. Content voids presented as minimalism, not intentional simplicity
H9. Error recovery 3/5H9. Error Recovery — 3/5Help center main page loads in Turkish despite English platform setting. Promo code field provides no error feedback for invalid codes. Balance history shows three identical "Trial lesson" entries with no explanation. Help center locale mismatch, no inline validation feedback
H10. Help and documentation 3/5H10. Help and Documentation — 3/5No contextual tooltips on Achievements page or subscription purchase page. No onboarding walkthrough for the dashboard. Help center locale mismatch undermines accessibility. Help is reactive, not contextual
Don Norman's 7 Design Principles (2.4/5) Conceptual model failure: parent expects learning dashboard, platform delivers booking tool +
Principle Score Key finding
DN1. Discoverability 2/5DN1. Discoverability — 2/5No cross-link or preview on the dashboard pointing to child's progress. "Achievements" label is ambiguous. Game World and Album features are buried deep in sub-navigation with no surface-level visibility. Progress hidden behind a context switch; Game World content not linked from dashboard
DN2. Feedback 2/5DN2. Feedback — 2/5UI interaction feedback (tab states, price updates) works fine. But no post-lesson summary, no per-skill mastery notification, no celebration on achievement. Learning feedback is completely absent. Adequate UI feedback; zero learning feedback after lessons
DN3. Conceptual model 1/5DN3. Conceptual Model — 1/5Parent expects "learning companion" but gets a booking-and-billing tool. 3 of 6 nav items serve business goals, not parent goals. Parent area and Achievements are architecturally separate applications with different layouts. Booking-and-billing tool vs. parent's expectation of a learning companion
DN5. Signifiers 2/5DN5. Signifiers — 2/5All 16 skill badges use identical visual treatment — no checkmark for mastered, no lock for not-started. 15 of 16 curriculum topics have identical gray markers with no completion signifier. No visual state on skill badges; no progress signifiers on curriculum topics
DN4 Affordances 3/5, DN6 Mappings 3/5, DN7 Constraints 4/5 — adequate; findings concentrated on learning, not UI mechanics.
Cognitive Walkthrough Booking a lesson: 90% success. Checking what child learned: 15%. Understanding weekly progress: 5%. +
Task Success Failure point
Check child's learning progress after a lesson 15%Task 1: Check progress — 15%Fails at Step 1: dashboard has no post-lesson content. Completed lessons have no detail view. Achievements exist but with no per-lesson attribution — parent cannot answer "what did my child learn today?" No post-lesson summary; achievement badges show no individual skill state
Book the next lesson 90%Task 2: Book lesson — 90%Succeeds because booking is the platform's primary conceptual model. Prominent CTA, clear teacher marketplace, green time slots. Only gap: no learning-outcome data on teachers to inform the choice. Minor: no learning-outcome data on teachers
Understand what the child learned this week 5%Task 3: Weekly summary — 5%Effectively impossible. No "this week" filter exists in any view. Schedule shows attendance only. Achievements are cumulative with no timestamps. Parent gives up after checking 4 sections. No temporal dimension in any progress data
Find and contact a different teacher 60%Task 4: Find teacher — 60%Booking succeeds but informed comparison fails. Teacher profiles show personality tags only — no student outcomes, no specialization data, no "send message" option before committing to a booking. No learning-outcome data on teachers; no pre-booking contact
Understand subscription options and pricing 45%Task 5: Understand pricing — 45%Parent can see options but faces 4 simultaneous decisions with no guidance. "$" symbol with TL-range values is confusing. "Speaking practice" add-on appears as a small note. Must leave purchase flow to find help articles. 4 simultaneous decisions; no contextual help; currency confusion
6-lens heuristic evaluation (1.5/5) Scores per lens — Value Communication 1/5, Information Hierarchy 1.5/5, Progress Visibility 1/5, Data Density 1/5, Actionability 2/5, Emotional Tone 2.5/5 +

The six lenses converge on the same root problem already surfaced by Nielsen and Don Norman: the AI layer is invisible to the parent. Value Communication and Progress Visibility both score 1/5 because the dashboard never translates the 1,500 micro-skills Novakid already tracks. Emotional Tone is the lone outlier — Game World is warm for the child but the parent dashboard has no trace of it. Full per-lens analysis in heuristic-evaluation.md.

Competitor landscape: what parents see elsewhere

Cambly KidsGoStudentLingokidsPreplybreadth-scan.mdwalkthrough.md

I registered in 4 direct competitors, completed a real trial lesson at Cambly Kids, and captured 86 primary-source screenshots. Here is what I found:

Novakid Cambly Kids GoStudent* Lingokids Preply
Model 1-on-1 live (25 min) 1-on-1 live (30 min) 1-on-1 live (50 min) Self-paced app Adult marketplaceKids-aware onboarding, adult platform11-step onboarding wizard branches for kids (school type, child's level, schedule). 40K+ "for kids" tutors. But post-login experience is identical to the adult platform: 50-min lessons, no child-safe UI, no COPPA.
Progress view Buried in sub-tabsProgress exists but is buriedCan-Do skill badges and curriculum exist in the Achievements tab, but this tab opens a different navigation context ("two apps in one shell"). Main dashboard shows zero learning data. Cognitive walkthrough: 15% success rate. None (parent)Cambly Adult HAS progressAdult Cambly has a Progress page (WPM, speaking %, achievements), a 3-axis Speaking Level Test, and a Learn page with 7 self-study tools including CamblyAI. None exists in Kids — proving infrastructure is there but wasn't extended. AI summaries*Sales-gated, unverified firsthandGoStudent reportedly has AI lesson summaries, recordings, and Magic Quizzes. Their marketing claims parents revisit summaries 2.3x on average. We could not access the dashboard (sales-gated after 15-step registration). The wizard includes a grade-improvement chart before the frequency question — a textbook anchoring sequence. App-only (Plus)Play metrics, not learning outcomesProgress Reports in iOS app: Activities Completed, Top Practiced Skills, Most Played Subjects. 2-week rolling window. Plus tier only. Not on web. Lesson InsightsMarketing claim, not verified in-productPreply markets AI Lesson Insights (vocab, grammar, speaking feedback). Our authenticated walkthrough found zero progress visibility — "My lessons" is a booking list only. Feature may be tutor-side only or not yet launched in all markets.
Baseline Auto by age Auto by age Advisor-assessed None Self-reported
Can-do skills Badge + text, 80+ est. Per-course, 6 each Not public Skill categories None
Free trial Free Paid trial Free, advisor-scheduled Freemium Paid (refundable)
Lesson data 1,500 micro-skillsClaimed, not independently verifiedSource: Leta Capital case study (Novakid's investor). Reproduced by TEFL Academy and Teast.co. No independent audit. The number is plausible but citation chain is thin. Stars, notes, video, slidesRich data, none surfaced to parentPer-slide star ratings, teacher vocabulary notes in chat, whiteboard co-creation, full video recording, lesson slides. All available for review but NOT structured or pushed to parents. AI summaries, quizzes* Activities, skills, words AI Lesson Insights*Public claim, not verifiedPreply markets AI-generated lesson summaries from audio transcription. Not visible in our authenticated walkthrough — possibly tutor-side only or not launched in all markets.

* GoStudent: dashboard gated behind sales call. Data from registration flow + public sources only.

Audit synthesis: root problems identified

Cross-referencing the platform audit, four formal UX evaluations, and the competitor analysis gives the following root problems:

Problem Source
Critical — existential for Value Hub redesign
01 Progress data is invisible to the parentData exists internally, none of it surfacesNovakid tracks 1,500 micro-skills, Can-Do badges, and multi-level curriculum — parent UI surfaces none of it. 6-lens 1/5, Nielsen H1 1/5
02 Wrong conceptual modelBooking tool vs. learning companionPlatform architecture serves business operations (book, pay, configure); parents expect it to serve learning. Don Norman DN3 1/5
03 Two disconnected apps in one shellParent dashboard ≠ AchievementsDifferent navigation, different visual language, no cross-linking between the two sections. Nielsen H4 2/5
04 Post-trial dashboard is only a booking reminderWasted conversion momentCritical conversion screen — nothing about what the child learned or why the parent should continue. 6-lens, Cognitive Walkthrough
05 No baseline assessment at onboarding — no before/after comparison possible Don Norman DN1 2/5
Significant — high priority
06 Teacher feedback is structurally generic2-minute window, no depthTeachers have ~2 minutes between lessons. No post-lesson summary, no "what was covered" structure. Don Norman DN2 2/5
07 No learning-oriented actions for parentsOnly operational, no pedagogicalParents can book, pay, configure. No messaging, goal-setting, progress sharing, or feedback loop. 6-lens Actionability 2/5
08 System language, not parent languagePedagogical jargon vs. everyday words"Speak about objects using 'my'" instead of "Can talk about toys." Curriculum codes instead of human descriptions. Nielsen H2 2/5
09 Can-Do skill badges are visually identicalNo mastery signalNo distinction between mastered, in-progress, and not-started skills — all badges look the same. Don Norman DN5 2/5
10 Library is 25+ categories dumped flatNo filtering, no personalization25+ categories with no filtering by student level, no personalization, no recommendations. Nielsen H7 2/5
11 Information hierarchy is operations-firstLearning pushed to sub-tabsScheduling and billing dominate the dashboard; learning is secondary. 6-lens Info Hierarchy 1.5/5
12 Emotional tone mismatchWarm for child, cold for parentChild's Game World is warm and engaging; parent dashboard is functional and cold. 6-lens Emotional Tone 2.5/5
13 No parent-native progress dashboard in kids EdTechMarket gap — Novakid can own thisGoStudent gates the dashboard behind a sales call. No competitor currently serves parents directly with a progress dashboard. Competitor analysis

03 Voice of the parent

I collected 128 real parent quotes, coded them bottom-up by theme, grounded the findings in nine established frameworks, and simulated eight parent interviews.

What parents write about

40 Trustpilot Global22 Regional (6 markets)12 Russian platforms19 Alt sources35 EdSource + RU platforms
Method note. I scanned ~18,500 reviews across 15+ platforms in eight languages, kept 128 verbatim quotes, and tagged each to one of six recurring topics. Percentages are how often each topic appears in the 128-quote corpus. One quote can cover several topics, so they don't add to 100%.
Topic % of reviews Sentiment
Teacher quality ~65% Overwhelmingly positive
Child engagement ~55% Overwhelmingly positive
Progress / improvement ~40% Mostly positive
Scheduling / flexibility ~30% Mixed
Billing / cancellation ~15% Mostly negative
Dashboard / parent account ~2% Rarely mentioned

The dashboard is mentioned in ~2% of reviews. Parents talk about "progress" constantly — but in qualitative, real-world terms ("can order food," "best in class," "sings in English"). They almost never reference the dashboard itself. It is functionally invisible in the parent experience.

Persona interviews on progress mental models

interview guide8 persona interviews
Method note. Eight parent personas (six JTBD jobs + two edge cases) walked through 18 questions about how they think about progress when no product is in the room. Novakid kept out of the prompt. Design probes, not real user research.
All 8 parents named a real-world moment as their strongest proof +

Every persona spontaneously named one specific real-world moment as the clearest memory of their child's progress: a Greek restaurant menu, a café in Marmaris, Lake Como, a London Minecraft weekend, a beach in Brittany. None named a score, a level, or a dashboard stat. Design takeaway: the hub should connect to moments like these, not replace them.

All 8 hit a plateau; the platform stayed silent +

Every persona described a plateau — a stretch of weeks between month two and month eight when the child seemed to stop progressing. Four of the eight (🇵🇱 Warsaw, 🇮🇹 Milan, 🇩🇪 Munich, 🇹🇷 Istanbul personas) said they almost cancelled during it. In every case they were reassured by someone outside the product: a relative with a language background, an outside teacher, a school meeting. The platform said nothing. The plateau is the highest-stakes moment for retention, and the product is currently silent through it.

7 of 8 built their own tracking systems +

Phone notes, monthly video recordings, spreadsheets, paper notebooks, project logs, illustrated diaries, shared Notes documents — seven of the eight personas described their own tracking system. The 🌍 MENA persona is the outlier: refuses to track at all, which is itself a pattern. All seven trackers said they would stop if the platform captured progress moments well. This is the clearest product-demand signal in the whole research.

None chose pure story or pure data — all 8 chose a sentence with evidence underneath +

Asked to choose between a story-paragraph dashboard, a data-heavy dashboard, and a layered one (short sentence at the top, evidence underneath), all eight chose the layered one. 🇮🇹 Milan persona captured the reasoning: "if there's no 'based on what,' I don't trust the sentence." Neither pure prose nor pure data wins. The working pattern is a short, specific claim with receipts below it.

Parents picture progress in three distinct ways — the hub must serve all three +

Asked to picture their child's progress as an image, the eight metaphors fell into three groups. Stepped — a staircase, a construction project: these parents want a clear roadmap. Organic — an opening flower, two parallel rivers: they want moments and relationships. Directional — a bridge being built, an investment portfolio, a running motor: they want proof of forward motion. A single visual tone won't serve all three; the hub's default view has to read clearly for each.

Two edge-case personas surfaced design categories the reviews miss +

The 🇩🇪 Munich persona has basic-school English and cannot read an English weekly digest. EdTech products routinely assume both parents are fluent — that assumption fails for a large share of the real audience. The 🇫🇷 Lyon persona has two children at different levels; parents with two children are not "parents × 2," and the common child-switcher dropdown actively hurts them. Neither case is strongly visible in the review corpus; both are real product categories the hub must accommodate.

Simulated personas produce hypotheses; real-quote patterns and theory below confirm or push back. Decisions resting on a persona finding carry a note to validate with real parents — see the validation roadmap.

Theoretical grounding

theoretical-frameworks.md
Method. Nine frameworks cross-check the §03 patterns; four (JTBD, PEAS, SDT, British Council plateau) do load-bearing work in P1–P7 and the thesis. The remaining five support edge cases — tap any framework for the short version.
Framework Core contribution
Self-Determination TheorySDT — Deci & RyanParent-side needs: autonomy, competence, relatedness. Hub must support each — clear controls (autonomy), understandable progress (competence), two-way teacher channel (relatedness). Block one = churn; all three = minimum bar. Parent needs = autonomy + competence + relatedness. All three must be supported.
Expectancy-Value TheoryEVT — Eccles modelRenewal = expectancy × (attainment + utility + intrinsic − cost). Each term needs its own evidence on the hub. Opaque cost or lost expectancy breaks the chain. A renewal-defence artefact covers all four in ≤30 seconds. Renewal decomposes into 4 value terms; all must be evidenced.
Jobs-To-Be-DoneJTBD — Christensen school"Progress is the currency students spend." 7 functional + 3 emotional + 2 social jobs parents hire the hub for. The 6 jobs below come from this analysis. Every screen testable against "which job does this serve?" Source of the 6 jobs below. Every surface serves a job — or it is decoration.
Parental Educational Anxiety ScalePEAS — Guo et al. 2024, 2025Anxiety is measurable, 3-dimensional (learning / health / resources). Transmitted 49% via overprotection, 15% via rejection; reduced 36% by emotional warmth. Novakid partly functions as anxiety relief. Uniform-positive feedback paradoxically amplifies anxiety, not reduces it (see P2). Uniform-positive feedback paradoxically amplifies parent anxiety. Anti-pattern for P2.
British Council plateau researchSLA intermediate plateauDocumented phase of second-language acquisition. Understanding develops before speaking — a 2–8 month window where visible output stalls though internal learning continues. Parents don't know. The plateau protocol (P5) names it proactively, turning a churn trigger into a credibility moment. Plateau is a documented SLA phase (months 2–8). Source of plateau protocol in P5.
Hofstede cultural dimensionsHofstede — UA, IDV, PDIHigh uncertainty avoidance (UA) predicts preference for the structured CEFR ladder (A1–C2) — especially PL, DE, RU. Individualism vs collectivism (IDV) predicts peer-ranking uptake (opt-in in RU/TR; default-off in DE/FR). Power distance (PDI) shapes teacher-authority framing. Grounds the cultural variation table below.
Baumrind parenting typologyBaumrind — authoritative / authoritarian / permissive / uninvolvedNot a job axis (JTBD handles that), but a design dimension. Authoritative: structure + warmth. Authoritarian: clear metrics. Permissive: hurt by pressure-framed copy. Uninvolved: anomaly-only alerts. Design dimension, not job axis. Informs default tone and pressure calibration.
Behavioural economicsEndowed progress, loss aversion, sunk-cost (flipped)Endowed progress: visible ongoing progress motivates continuation. Loss aversion: "lose the progress built" beats "save the money". Sunk-cost flipped: in subscriptions, parents cancel invisible-value investments rather than over-committing. Cancellation framing: "lose progress" outperforms "save money." Invisible-value drives churn.
Dashboard UX researchF-pattern, info overload, non-expert constraints46.7% of dashboard users hit info overload as primary failure mode. F-pattern eye-tracking — top-left gets maximum attention. Rare use + non-expert (the parent profile) = interpretation needed, not raw data. Top-left hero must be an interpretation, not a metric.

Parent patterns

parent-voices-analysis.md
Why patterns. I brought the four layers together — the 128-quote corpus, the theme-coding pass, the frameworks, and the 8 persona interviews — to show the patterns each layer alone could not prove. Seven patterns emerged that map directly to the Value Hub redesign. Two more (billing-trust failure, scheduling friction) are real in the corpus but scoped out as prerequisites to the hub, not design targets of it.
P1 Evidence of progress lives outside the dashboard +

Parents measure progress three ways: real-world moments (ordering food, travelling, talking to relatives), external validation (school grades, olympiads), and the child's voluntary use of English. Platform metrics almost never come up. The dashboard appears in ~2% of 128 quotes. The one praised dashboard feature is the recorded-lessons archive — parents use it as a do-it-yourself check. When neither the world nor the platform shows evidence, parents conclude the product isn't working. All eight persona interviews confirmed the pattern.

Hypothesis
If the Value Hub translates internal platform data into observable real-world behaviours ("Mia can now tell a story about last weekend") and captures parent-reported real-world moments, it becomes the bridge parents are already trying to build themselves

"My daughter has been taking this course for over three years and I must say the improvements are visible, even acknowledged by her school teachers."

🇮🇹 — Dorotea, Trustpilot Italy, 5★ — proof arrives from outside the platform

"Parents often rely on report cards to monitor their children's academic progress. If they're getting A's and B's, they don't look at anything else."

🇺🇸 — Araceli Simeón, Parent Organization Network, EdSource — the research context

"I use Novakid for my 11 years old son and both him and me love Novakid!! The feature I like the most is that I can re-watch his classes and follow his improvements :)"

🇪🇸 — Juan Ramon de la Torre, Trustpilot Spain, 5★ — the one dashboard feature praised is recorded lessons = a DIY verification tool

P2 Post-lesson feedback is structurally dishonest +

The clearest trust-break in the corpus is uniform positivity — notifications stay rose-coloured regardless of how the lesson went (realjaew, TR). Without bad news, parents can't tell a strong lesson from a weak one and stop trusting the channel (Parental Educational Anxiety Scale, Guo et al. 2024). It compounds with a missing learning plan (ALFAZANI TALHA) and parents forced into the tutor role (Margarita, IT). Reassurance-seekers find the bland tone untrustworthy; evidence-seekers find nothing to check. When teachers write specific, calibrated feedback, parents name it as the most valuable feature (Pınar, UK). The fix isn't more warmth — it's calibrated honesty.

Hypothesis
If post-lesson notes pair one genuine strength with one area to reinforce (plus a 1-sentence forward focus), the feedback becomes diagnostic — parents regain the ability to distinguish good lessons from bad, and the channel earns trust

"Notifications to parents are always positive. Even if the lesson went badly, everything is rosy pink... when 25 minutes are up, the teacher wraps up in a rush."

🇹🇷 — realjaew, Eksi Sozluk — the core trust-break: uniform positivity destroys the signal

"There is no method that is called LEAD-IN from the teacher. The child experiences great difficulty to understand WHY HE IS DOING this or that during the lesson."

🇮🇹 — Margarita, Trustpilot Italy, 2★ — when the teacher doesn't contextualise, parents are forced into the tutor role

"His teacher writes a review right after each lesson so that I know how to support him."

🇬🇧 — Pınar, Trustpilot UK, 5★ — counter-example: when feedback IS specific, parents cite it as the most valuable feature

P3 The curriculum is a black box (the Navigator blind spot) +

Four curriculum-visibility failures compound:

  1. No placement test — no baseline at onboarding (pelma pelma).
  2. Level doesn't auto-advance — Nejat paid 12,553 TL over 6 months, stuck at one level (confirmed by an independent teacher).
  3. Mismatch only fixable by escalation — Maria Angels Fiol (child too advanced), Giulia Trivero (too basic, 4 lessons on colours).
  4. Difficulty jumps without warning — Lyudmila: "the program became much harder — all these phrases were completely new."
The root: a curriculum invisible to the parent. 🇵🇱 Warsaw and 🇫🇷 Lyon personas both asked for "the full staircase." The Navigator job — the parent who wants to see the path ahead — is blocked.
Hypothesis
If the curriculum is surfaced as a first-class navigable roadmap — where my child is, what's coming next, when difficulty will step up — the Navigator job is served, and three of the four failures above self-correct through visibility

"The program was supposed to begin at Level 1 and progress according to the child's ability... I realized that without my intervention, advancement to levels 2 or 3 wouldn't have occurred."

🇹🇷 — Nejat, Sikayetvar.com — 6 months / 12,553 TL at the same level, confirmed by independent teacher

"It might be good for beginners but not for children who can already speak English fluently... She spent 4 lessons on colors and honestly didn't learn anything."

🇮🇹 — Giulia Trivero, Trustpilot Italy, 2★ — mismatch in the other direction: advanced child stuck in beginner content

"The program became much harder... all these phrases, except 'I've got,' were completely new!"

🇷🇺 — Lyudmila, tutortop.ru, 3★ — difficulty jump without pre-warning

P4 The teacher-parent channel is structurally one-way +

Parents receive feedback. They can't send any back. The product treats the parent as an audience, not a participant. Carmentxu: "they everyday give me feedback, but I can't." What's missing isn't dialogue; it's context transfer. 🇪🇸 Madrid persona: "a warm, light thread where I can say 'she's nervous about tomorrow because she had a hard day at school'." Without it, the teacher enters each lesson blind to the child's day, and the parent feels excluded from a relationship they pay to sustain. The one-way channel also compounds teacher changes: when a swap happens unilaterally (Ozge: "trauma in our lives"), the parent has no thread to object through first.

Hypothesis
A small two-way channel — the parent sends up to two sentences of pre-lesson context per week, the teacher acknowledges — turns a one-way feed into a relationship. Not chat, not homework coordination. Just context transfer

"My kid is learning a lot with Novakids. But there is one thing that I don't like: they everyday give me feedback, but I can't. I would like to write to the teachers to give them my feedback and I can't do it."

🇪🇸 — Carmentxu, Trustpilot Spain, 4★ — names the structural complaint directly

"My son cried so much and was so upset that they literally left a trauma in our lives."

🇹🇷 — Ozge, Sikayetvar.com — teacher change handled unilaterally; parent had no thread to object first

"Inexperienced teachers assign some students the wrong level, and then they get stressed that they are not progressing."

— Jelena I., SmartCustomer, 1★ — without a parent-side channel, misjudgements aren't caught early

P5 The plateau is the product's most expensive silence +

Between months 2 and 8, most learners hit a plateau — visible output stalls, but understanding keeps growing quietly underneath. Second-language-acquisition research calls this the intermediate plateau (British Council). Parents don't know this. They see the flat window, assume the product has stopped working, and start thinking about cancelling.

All eight persona interviews described a plateau moment. The 🇵🇱 Warsaw persona nearly cancelled; 🇮🇹 Milan questioned the spend; 🇩🇪 Munich was "unsure for three weeks". In the corpus, 2–3★ reviews cluster around the same signal. The platform says nothing during this window — parents are rescued instead by outsiders, like a relative who studied linguistics or an independent teacher. A pre-written message would be the cheapest fix in the product, and it lands on the highest-churn moment in the journey.

Hypothesis
A proactive plateau protocol — name it before it hits, explain the gap between understanding and speaking with authority, offer a diagnostic action (one-minute video comparison), commit to a post-plateau check-in — converts a churn-trigger moment into a credibility-building moment

"Little information each class, I feel he is not improving at all."

🌐 — Anas, Trustpilot Global, 2★ — classic plateau voice: nothing visible, cancellation under consideration

"The learning system is well thought-through and adapted to kids... I seriously dislike the subscription system -- automatic renewal is the only option... I am fed up and I already blocked future payments."

🌐 — Rafal W, Trustpilot Global, 3★ — plateau parent who says the system is good, yet exits. No proactive plateau message ever reached him

"Программа сильно усложнилась... все эти фразы, кроме I've got, были абсолютно новыми!"

🇷🇺 — Lyudmila, tutortop.ru, 3★ — the plateau-break moment: the platform should celebrate, doesn't

From open coding: across 128 quotes, no parent received a proactive plateau message. The silence is universal; the churn shows up in 2–3★ reviews in every market. This is the single highest-leverage retention fix in the product.

P6 The child's own desire is the #1 retention signal — yet invisible to the platform +

The strongest parent-cited retention signal, in every market, is the child's own desire to continue. "He's happy and wants to keep studying." "She begs for more lessons." This outweighs any platform metric, teacher report, or CEFR level in the renewal decision. 🇹🇷 Istanbul persona: "a smile after class is the best data I have."

The paradox: the platform collects rich engagement data (attendance, Game World activity, per-slide stars, homework) but never shows it to the parent in an interpreted form. The parent has to infer. JTBD names the logic: kids "spend" progress as currency. When the child's motivation is visible, retention is near-automatic; when invisible, the parent has to be convinced separately.

Hypothesis
If the Value Hub surfaces the child's enjoyment as a first-class signal (per-lesson enjoyment moments, favourite topics, teacher-flagged engagement beats) — not as a sterile engagement score, but as interpreted human moments — the parent's strongest retention evidence becomes visible in the product itself

"A year later, studies with pleasure, almost independently, has good vocabulary, constructs sentences."

🇷🇺 — ZULYAen, kursfinder.ru, 5★ — "studies with pleasure" cited before any academic marker

"Our children genuinely enjoy each class, and their English level has improved significantly."

🌐 — Diana Cohen, Trustpilot Global, 5★ — the ordering is telling: enjoyment first, skill gain second

"My daughter's avatar collection keeps her begging for more lessons — it's genius!"

🌐 — unnamed mother via easyesltips.com — gamification turns the child into the driver of her own retention

Persona-cross signal (generative only, not counted): the same pattern in every one of the 8 interviews. 🇫🇷 Lyon persona: "both children reference English content spontaneously." 🌍 MENA persona: "if he becomes resistant to going, that would trigger my attention." Child desire is not a pattern parents prefer over data — it is the pattern they use.

P7 Culture changes the proof-type, not the underlying job +

A counter-intuitive finding from open coding: parents across seven markets are doing the same jobs (Reassurance-Seeker, Auditor, Investor, Navigator, Co-Pilot, Outsourcer) — but they accept different forms of proof.

  • Russia — olympiads and school ranking.
  • Italy — value-per-euro math.
  • Türkiye — travel-confidence.
  • Poland — CEFR levels (the A1–C2 European scale).
  • Germany — peer integration with native speakers.
  • France — closing the school-curriculum gap.
  • Spain — Cambridge-exam pathway plus shyness overcome.
Grounded in Hofstede via the cultural analysis: uncertainty avoidance predicts a preference for structure (PL, DE, RU); individualism predicts peer-ranking uptake (opt-in in RU/TR, default-off in DE/FR). 🇷🇺 Moscow persona: "I want the school benchmark, not the European." Insight: not seven dashboards — one architecture with a configurable lead proof-type.
Hypothesis
One architecture; the lead proof tunes per market. Hero surface per market: RU school-alignment + opt-in olympiad track; IT explicit cost-per-outcome math; TR real-world moment capture; PL CEFR ladder; DE peer-integration milestones; FR school-gap framing; ES shyness-overcome + Cambridge pathway

"My son has been attending classes at Novakid for almost 3 years. He started at the age of 7, and is currently at level A2."

🇵🇱 Nika Ka, Trustpilot Poland, 5★ — PL: CEFR-as-proof

"When we travel abroad, they are able to communicate clearly and express their needs with ease."

🇹🇷 Ozge, Trustpilot Turkey, 5★ — TR: travel-confidence

"She speaks fluently, participates in Olympiads and international language competitions."

🇷🇺 Anelazil, kursfinder.ru, 5★ — RU: competitive-academic validation

"Ottimo rapporto qualità prezzo! Il nostro insegnante è bravissimo e empatico, è il secondo anno di corso per mio figlio."

🇮🇹 Elena B, Trustpilot Italy, 5★ — IT: qualità/prezzo framing

"After achieving 4 levels, my 12 years old son is already fluent. The content completes what they do at school (100% French school)."

🇫🇷 Mariem, Trustpilot France, 5★ — FR: school-gap framing

Two more strong corpus patterns — billing-trust failure and scheduling friction — are scoped out as trust-stack prerequisites, not design targets of the hub itself. See the full parent-voices analysis.

Six parent jobs

archetypes.mdopen-coding-analysis.md
Method. Derived bottom-up via open coding of 128 parent quotes (quote → theme → job mapping in the linked file, not re-rendered here), then anchored in SDT/EVT/JTBD. A job = what a parent hires the Value Hub to do, not a personality type — the same parent occupies Navigator at onboarding, Reassurance-Seeker during a plateau, Investor at renewal. Limitation: six is a judgement call on where to draw cluster boundaries; the full quote → job mapping in archetypes.md is auditable.
Job title Design implications
Reassurance-Seeker
"Tell me I'm doing the right thing as a parent."		 One specific falsifiable observation beats warm summaries. Teacher voice present. Proactive plateau-naming. Child's enthusiasm captured.
Auditor
"Show me the evidence so I can judge for myself."		 Depth layer mandatory. Calibrated honesty (one strength + one area to reinforce). Falsifiability as design principle. Triangulation hooks.
Investor
"Defend the money to me."		 Renewal-defence artefact. Cost-per-outcome math (market-configurable, default OFF in FR/JP/DE). Clean exit path. Transparent billing.
Navigator
"Show me the path ahead so I can prepare."		 Curriculum roadmap as first-class surface. Placement test at onboarding. Pre-plateau warning. Difficulty-jump signalling. External alignment hooks (school, Cambridge).
Co-Pilot
"Give me a role without becoming the homework parent."		 Bi-directional teacher channel, bounded. "Capture-a-moment" affordance (10–30s). Pre-lesson prompts. Shareable family milestones.
Outsourcer
"Just work — don't make me manage this."		 Invisible operation default. Anomaly-based alerts only. Weekly digest IS the surface (not dashboard). No engagement-seeking UX, no parent gamification.

Cultural variation: universal vs adaptive

cultural-analysis.md
Why cultural layer. I analysed seven markets (PL, IT, TR, RU, DE, FR, ES) separately to avoid a one-size-fits-all hub. Inputs: regional Trustpilot corpora, Hofstede dimensions, cross-cultural anxiety research, competitor localisation. Finding: jobs are universal; proof-types are not. Some design decisions travel; others need per-market tuning. Full layer →
Design decision ScopeUniversalThe need is invariant across the 7 markets studied. Format may localise (channel, tone), but the design decision applies identically everywhere — skipping it hurts parents in every market.
AdaptiveThe need is market-specific: the same implementation helps in one market and harms in another. Shipping one version worldwide is an anti-pattern — the design must be tuned per market (default on / off / opt-in per locale).		 Note
AI progress narrativeUniversalFormat varies, need is constant
Progressive disclosureUI patternA short headline at the top; detail on demand. One specific sentence, then the evidence, then the deep audit trail. Matches what all 8 personas picked as their preferred dashboard.UniversalHeadline first, evidence on demand
Plateau detection + reassuranceUniversalSee pattern P5 above
Can-do statements over abstract levelsPlain-English progress"Mia can order food in a restaurant" beats "Mia reached A2.3". Can-do statements name what the child can actually do — the opposite of CEFR-ladder labels. All 8 persona interviews preferred them.UniversalDescribe what the child can do, not a level label
Weekly push summaryUniversalChannel varies (WhatsApp / Telegram / KakaoTalk / email)
Interface languageAdaptiveMust ship in the parent's language in 10+ markets
Lead proof-type per marketAdaptiveOlympiads in RU, travel-confidence in TR, value-per-euro in IT, CEFR in PL, peer-integration in DE, school-gap in FR, Cambridge pathway in ES — see P7 above
Sharing artifact toneAdaptiveRestrained in DE, FR; ranked/certified in RU, PL; celebratory/family-forward in IT, TR, ES
Peer benchmarkingAdaptiveOpt-in only; default off in DE, FR (high individualism)
Cost-per-skill metricAdaptiveDefault in IT, TR (Investor-heavy); hidden in DE, FR (harms the comfort-seeking and co-piloting jobs)
Pre-K vs advanced reportingAge-split templatesPre-K (ages 4–6): visual moments + short teacher voice notes + "favourite topic this week" — no letter grades. Primary (7–10): a few can-do statements + recorded lesson clip. Older (11–12): can-do statements aligned to CEFR + Cambridge-pathway signal.AdaptiveDifferent template per age bracket — tap for detail

The trust stack

synthesis §4 tensionssynthesis §7 trust stack
Own framing, not from literature. The trust stack is a framing I constructed to map the five tensions above against the five points in the product where parent trust can break. Not a published model. Useful because it makes scope explicit: the Value Hub is layers 4–5 by design, layers 1–3 are prerequisites — a beautiful hub on broken billing is still a sinkhole.

The five tensions running through the six jobs: Honesty ↔ Anxiety (resolved in P2), Control ↔ Convenience (Weekly Digest for Outsourcers + opt-in teacher channel for Co-Pilots), Data ↔ Story (P1), Platform ↔ Real World (Moment Capture), System ↔ Child (P6).

Layer What breaks it · evidence · fix scope
5. Narrative & interpretation AI summaries disconnected from source data. 🇮🇹 Milan persona: "if there's no 'based on what,' I don't trust the sentence." Fix in the Value Hub.
4. Data & interpretation Metrics without interpretation; uniform-positive feedback. realjaew: "notifications are always positive." Fix in the Value Hub.
3. Teacher continuity Unilateral teacher changes. Ozge: "trauma in our lives." Ksenia: "told to choose others after 4 lessons." Prerequisite to the Value Hub — needs a teacher-change workflow with parent consultation.
2. Platform reliability Unreachable platform; broken progress features. Filiz: "inaccessible for days." Oya Canli: "progress button requests audio permission." Prerequisite — if the hub is gated behind a bug, no narrative above reaches the parent.
1. Billing (foundation) Auto-renewal surprises, hostile cancellation, pause-doesn't-pause-billing. Rafal, Rafi Camhi, Jeanne Alexandre, Andrey Svechnikov — billing complaints dominate 1–2★ across every market. When billing trust collapses, educational value is erased retroactively. Foundational prerequisite — not a Value Hub deliverable.

Implication for the brief: the Value Hub, strictly scoped, covers layers 4–5. Layers 1–3 must be fixed in parallel — or the hub ships on top of a trust sinkhole and fails regardless of its own quality.

From synthesis to design. The patterns, jobs, and trust stack above translate into ten concrete design imperatives across five bridges (AI, teacher, child, real world, school) + two edge-case categories (multi-child, non-English-speaking parents). Each imperative becomes a row in the Hypothesis Atlas (§05) with its own evidence, reference, and verification plan — the atlas is the canonical list. Prerequisite for all of them: trust-stack layers 1–3 must be fixed in parallel, or the hub ships on top of a sinkhole.

04 Thesis

Novakid tracks 1,500 micro-skills per child. The dashboard shows none of them.

In 128 real parent reviews, the dashboard is mentioned in ~2%. Seven of eight interviewed parents build their own tracker (phone notes, spreadsheets, video). Per-slide star ratings, teacher vocabulary notes, full lesson recordings: all collected, all invisible. The bottleneck isn't measurement. It's translation.

Five sources hold part of the picture — AI, teacher, child's experience, real world, school. None reaches the parent in a form they can act on. The Value Hub is the translation layer.

One architecture. Per-role defaults. Per-market screens.

One parent plays different roles at different moments — Navigator at onboarding, Reassurance-Seeker through a plateau, Investor at renewal. The hub has to show the same data through the lens that matches the moment. A separate dashboard per role fragments the product and breaks against real parenting, which is never one role at a time.

Five sources → six jobs

Every screen in the hub answers one question: which source is it translating, which job is it serving? If a screen can't answer both, it's decoration.

AI1,500 micro-skill trackers, per-slide stars, difficulty adjustment. The dashboard shows none of it.
TeacherIn-lesson chat, vocabulary notes, live observation. Lost when the lesson ends; post-lesson notes arrive bland and uniformly positive.
ChildEngagement, Game World activity, per-slide stars, emotional state. Never reaches the parent — yet the child's own will to continue is the #1 retention signal.
Real worldOrdering food on holiday, relatives, shifts in confidence. Captured nowhere. 7 of 8 personas run their own tracking; all 7 would stop if the platform captured these moments.
SchoolRussia's school ranking, Baccalauréat (FR), Cambridge YLE (IT/ES), CEFR (PL), peer integration (DE). The platform is blind to what each market uses as proof.

Four dimensions of progress

Reducing "progress" to one number fails parents. The hub treats it as four dimensions, and every surface makes clear which one it's showing: quantitative (CEFR, vocabulary, accuracy — the what), qualitative (teacher observations, child-specific strengths — the how), behavioural (real-world moments, school-teacher remarks — the so-what), and emotional (child's desire to continue, parent's peace of mind — the feeling).

Business math — a rough sketch

Novakid serves ~940K cumulative parents in 50+ countries at $9–20 per lesson. If the active cohort is ~100K and average lifetime value (LTV) is ~€1,200/year, a five-point lift in month-3 retention is worth ~€6M annual recurring revenue. CampusESP 2024 reports +8.3% retention from progress alerts in higher ed — different segment, transferable mechanism. Even at one-tenth of that lift, the ROI on a dashboard redesign is clear.

Months 2–8 are the highest-churn window. This is corroborated by British Council research on second-language acquisition, by 2–3★ cancellation stories across the quote corpus, and by every persona interview. A pre-written message at the right moment costs almost nothing and hits the highest-risk point in the journey — the cheapest high-impact intervention the product can ship.

Approach note. This is a five-day test assignment. With more time I would add: (1) 5–8 interviews with real parents to check the persona findings; (2) a pass on Novakid's actual month-2-to-8 retention curve to calibrate the plateau claim; (3) two weeks of technical discovery to find out whether the AI can generate weekly per-child narratives at acceptable cost and latency. What follows is a strategy and a design, not a production-ready specification.

05 Hypothesis Atlas

The atlas holds 35 hypotheses — 13 P0, 17 P1, 5 P2. Each row cites at least one of: the 128-quote corpus, a framework, a persona interview, or a reference pattern.

Ranking rubric. A hypothesis is P0 if it scores high on all four: (a) evidence strength — cited by ≥2 independent sources (corpus + persona, or corpus + reference); (b) impact — moves retention, trust, or activation directly; (c) cost — shippable within the assignment's default scope; (d) reversibility — testable and rollback-safe. P1 loses on one dimension (usually cost or dependency). P2 loses on two, or only surfaces in a single edge-case persona. Full per-row scoring in synthesis.md.
ID Headline — tap for evidence Rank → Design surface Impact
Bridge 1 — AI → Parent (13 hypotheses)
ai-narrative-hero One specific, checkable sentence a week replaces the empty dashboardJob: Reassurance-Seeker + Auditor (all six benefit).Evidence: corpus — parents track their own progress on every market; theory — SDT competence need + F-pattern dashboard reading; all 8 personas picked sentence-first.Reference: GitHub Primer sentence-first disclosure. Apple Screen Time two-line summary.How to check: A/B vs card layout, 20 parents × 14 days. Metric: 7-day return rate. Kill if >1 in 10 narratives disputed. P0 → Hero Narrative Retention
progressive-disclosure-by-job One sentence → evidence → audit trail, three layers in one screenJob: all 6 (each stops at natural depth).Evidence: dashboard UX research — 46.7% hit info-overload; non-experts need interpretation, not raw data. All 8 personas chose layered over pure-story or pure-data.Reference: Linear triage-vs-focus toggle. GitHub repo insights: headline → chevron → detail.How to check: heatmap click-depth per role in pilot. Outsourcers rarely past L1; Auditors often reach L3. P0 → Information Architecture Trust
plateau-protocol-proactive Name the plateau before it hits, explain it, give a bounded action, check in afterJob: Reassurance-Seeker + Investor.Evidence: British Council calls the intermediate plateau a documented phase of language learning. 2–3★ cancellation stories cluster here. All 8 personas described a plateau; 4/8 considered cancelling.Reference: Noom "consolidation phase". Strava "rest day". Both treat flat stretches as expected.How to check: A/B vs silent control, 200 parents × 60 days. Target ≥5 pp churn drop. Cheapest high-impact fix — a pre-written message. P0 → Plateau Protocol Retention
curriculum-roadmap-visible Curriculum as a navigable screen — where the child is, what's next, where difficulty jumpsJob: Navigator + Investor.Evidence: corpus — Nejat (TR, 12,553 TL stuck 6 months), Lyudmila (RU, unannounced difficulty jump), ALFAZANI TALHA ("no clear plan"). Personas: 🇵🇱 Warsaw + 🇫🇷 Lyon asked for "the full staircase".Reference: Duolingo Path, Khan Academy skill tree, Strava training plan.How to check: does "where is my child?" ticket volume drop ≥20% after launch? P0 → Curriculum Roadmap Retention
progress-button-fix-ux Fix the broken "Progress" button — it blocks behind an audio-permission promptJob: all 6 (prerequisite).Evidence: Oya Canli (Google Play); GAPS.md. Parents tapping "Progress" hit an unrelated audio-permission prompt — blocks the most-asked-for feature.Reference: hygiene, not a pattern.How to check: ship before launching the hub; a dashboard gated behind a bug won't be read. P0 → Engineering prerequisite Trust
progress-narrative-depth Every narrative claim links to traceable evidence (the "based on what")Job: Auditor + Investor.Evidence: corpus — 🇮🇹 Milan persona "if there's no 'based on what,' I don't trust the sentence." Synthesis §2.Reference: GitHub audit log pattern (each line traces to commit). Strava segment breakdown (each segment traces to data).How to check: parent trust score pre/post seeing the evidence trail; should shift ≥1.5 points on 5-point scale. P0 → Hero Narrative Trust
Bridge 2 — Teacher → Parent (3 hypotheses)
teacher-feedback-calibrated One strength + one area to work on + next-week focus — never uniformly positiveJob: Reassurance-Seeker + Auditor + Co-Pilot.Evidence: realjaew (TR) "always positive" = trust-break; Pınar (UK) praises specificity. PEAS: calibrated honesty lowers anxiety more than uniform positivity.Reference: Grammarly green/red score. LinkedIn Learning post-quiz breakdown. Peloton teacher form feedback.How to check: pilot 30 teachers / 300 parents × 30 days. Teacher admin ≤3 min/lesson; parent trust goes up; kill if teacher NPS drops >5 pts. P0 → Post-Lesson Card Trust
Bridge 3 — Child → Parent (3 hypotheses)
child-enjoyment-signal Per-lesson enjoyment, favourite topics, teacher-flagged moments — not a sterile engagement scoreJob: Reassurance-Seeker + Investor (top retention signal).Evidence: corpus — "happy and wants to continue" (Diana Mikel, ZULYAen, Diana Cohen, Aylin Gokturk). 🇹🇷 Istanbul: "a smile after class is the best data I have"; 🇪🇸 Madrid: "if she's laughing...". JTBD: kids "spend" progress as currency.Reference: Seesaw artefact capture. Duolingo character engagement. Peloton teacher shoutouts.How to check: reassurance score with enjoyment signal vs metrics-only should shift ≥1.5 pts on a 5-pt scale. P0 → Hero Narrative + Digest Retention
Bridge 4 — Real world → Parent (1 hypothesis)
real-world-moment-capture Voice note, photo, or short text → structured record → threaded into the AI narrativeJob: Co-Pilot + Reassurance-Seeker.Evidence: corpus — Greek menu, Marmaris café, olympiads, travel confidence. Every persona keeps their own tracking (phone notes, spreadsheets, videos); every one said they'd stop if the platform captured well.Reference: Seesaw artefact capture. Apple Journal voice/photo. Day One journal.How to check: activation rate in 28-day pilot. ≥50% active = win; <30% = redesign capture flow. P0 → Moment Capture Retention
Cross-bridge (5 hypotheses)
information-bridge-five-silos The hub architecture connects 5 silos (AI / Teacher / Child / Real-world / School) into the parent narrativeJob: all 6 .Evidence: synthesis §1 — the thesis; open-coding Theme A "evidence outside dashboard."Reference: Stripe role-based views of one data layer (owner / developer / support see different). Linear triage/focus toggle.How to check: meta — every surface in section 06 must trace to at least one silo and at least one job. P0 → Information Architecture Trust
Foundation — prerequisites for any hub surface to work (9 hypotheses)
trust-stack-billing-transparency Clear pause / cancel path; no auto-renewal surprises; visible next charge dateJob: all 6 (layer 1 of trust stack).Evidence: corpus — most common 1★ reason across all markets (Rafal, Rafi Camhi, Jeanne Alexandre, Andrey Svechnikov). 10 distinct trust-breaks coded. Synthesis §7.Reference: Netflix cancel-anytime baked into UX. Stripe subscription management clarity.How to check: 1–2★ review volume on billing topic should drop ≥50% in 90 days post-fix. P0 → Settings / account area Trust
teacher-continuity-policy Teacher-change workflow: parent consultation before change, not notification afterJob: Reassurance-Seeker + Co-Pilot.Evidence: corpus — Ozge (TR) "trauma in our lives"; Ksenia "told to choose others after 4 lessons." Theory: SDT relatedness violation.Reference: Hinge "introducing your next match" framing pattern. Spotify DJ introduces new voices.How to check: teacher-change complaint volume; retention delta at month 3–4 for affected cohort. P0 → Settings + notification Trust
dashboard-rare-use-assumption Design for weekly use by non-expert parent — interpretation, not raw dataJob: all 6 .Evidence: theory — dashboard UX research: 46.7% info-overload as primary failure mode; rare-use + non-expert users need interpretation, not raw data.Reference: Apple Screen Time (weekly summary, not daily log). Nest home summary.How to check: session length ≤60 sec for Outsourcer cohort; interpretability test with 12 parents. P0 → Design principle Trust
17 P1 hypotheses — strong evidence, high impact
can-do-statement-operationalized Describe what the child can do, not a level label ("can order food" > "A2.3")Job: Reassurance-Seeker + Navigator + Investor.Evidence: Cambly Kids already operationalises can-do at course granularity (6 outcomes per course). Theory: CEFR for-parents research. Persona: all 8 said "can do X" made sense, "A2.3" did not.Reference: Cambly Kids course-page can-do bullets. Lingokids skill-mastery framing.How to check: card-sort of "can order food" vs "A2.3" with 12 parents — which do they remember 48 hours later? P1 → Hero Narrative + Roadmap Trust
trajectory-over-point-in-time Show 4-week rolling averages as canonical; weekly numbers as noisy contextJob: Reassurance-Seeker + Auditor.Evidence: theory — weekly numbers bounce and trigger anxiety spikes; trends are interpretable and harder to misread. Reference: Strava, Apple Fitness, Noom all default to trendline, not daily.Reference: Strava training load rolling average. Apple Fitness trends. Noom weight trajectory vs daily number.How to check: qualitative — does a 4-week trendline correlate with parent reassurance scores better than the week number? P1 → Hero Narrative Trust
invisible-metric-during-plateau During a plateau, surface the metrics that are moving (comprehension, nuance)Job: Reassurance-Seeker.Evidence: theory — the plateau is a gap between understanding and speaking; comprehension keeps developing while production stalls. Reference: Duolingo adds league rank invisible yesterday, visible today; Noom adds CBT-lesson completion when weight plateaus.Reference: Duolingo league layer. Noom mindset-lesson count.How to check: during plateau cohort, does adding "comprehension signals" card reduce mid-month cancellation attempts? P1 → Plateau Protocol Retention
peer-benchmarking-opt-in Peer-ranking opt-in only; default on in RU/TR, default off in DE/FRJob: Investor (where welcome); potentially harms Reassurance-Seeker.Evidence: theory — Hofstede individualism-vs-collectivism; 7-market cultural analysis. Reference: Duolingo leagues as opt-in-by-market. Strava leaderboard privacy zones.How to check: per-market opt-in rate >15%? If <5% in any market, default remains off globally for that market. P1 → Settings & Defaults Expansion
cost-per-outcome-math Value-per-euro math — default in IT/TR (Investor-heavy); hidden in FR/DEJob: Investor (where welcome).Evidence: corpus — 🇮🇹 Giulia explicitly calculates €/lesson. Cultural-analysis: IT/TR investor-proof-type; FR/JP/DE cost framing instrumentalises non-instrumental value.Reference: Spotify "you saved £X vs buying songs" pattern (opt-in). Amex year-in-review rewards calc.How to check: per-market adoption rate + renewal delta among opt-in cohort. P1 → Settings & Defaults Retention
teacher-bidirectional-channel Parent sends up to 2 sentences of pre-lesson context per week; teacher acknowledgesJob: Co-Pilot.Evidence: corpus — Carmentxu (ES) "they give me feedback but I can't". Persona: 🇪🇸 Madrid described the exact mechanism.Reference: Superhuman reply-later note pattern. Intercom internal note (bounded, not chat). Front side-conversation.How to check: teacher-load cap: ≤2 parent notes / teacher / week. Kill if teacher NPS drops or parents escalate into chat expectations. P1 → Post-Lesson Card Trust
lesson-recording-accessible Recorded-lesson replay with timestamp controls and optional transcriptJob: Auditor + Co-Pilot.Evidence: corpus — Juan Ramon (ES) "re-watch and follow improvements" — one of the few dashboard features universally praised. Cambly Kids already ships per-lesson replay.Reference: Cambly Kids lesson archive. Peloton on-demand replays. Zoom recording with chapters.How to check: replay watch-rate vs control; Auditors should hit replay for >30% of sessions they flag. P1 → Post-Lesson Card Trust
school-alignment-per-market Map curriculum to local-school benchmarks per market (RU ranking, FR Bac, IT/ES Cambridge, PL CEFR)Job: Navigator + Investor + Reassurance-Seeker.Evidence: corpus — Dorotea (IT) "school teachers noticed," Natalya (RU) school-ranking, Mariem (FR) "100% French school." Persona: 🇷🇺 Moscow "I want the school benchmark, not the European." Cultural-analysis §Part I.Reference: Khan Academy grade-level alignment per country. Duolingo school-curriculum hooks.How to check: per-market retention uplift after adding school-alignment card; strongest expected lift in RU, FR, IT. P1 → Curriculum Roadmap Retention
year-review-artefact Annual synthesis, adapted per market, shareable to family (not public)Job: Reassurance-Seeker + Investor.Evidence: corpus — "what we tell grandma" pattern across markets. Persona: universal proud-moment narrative, explicit in 🇷🇺 Moscow, 🇮🇹 Milan, 🇫🇷 Lyon.Reference: Spotify Wrapped. Duolingo Year. Strava Year. Reddit Recap. Insight: annual recap is licensed emotional signalling.How to check: share-rate (family, not public); renewal-rate delta among cohort that received year-review vs control. P1 → Year-in-Review Expansion
proof-layer-culturally-tuned One architecture; lead proof-type tunable per market (Olympiads RU, €/lesson IT, travel TR, CEFR PL...)Job: all 6 (via different proof).Evidence: Cultural analysis, part I: 7 markets × proof-type mapping. Theory: Hofstede dimensions. Reference: Stripe role-based IA. Duolingo per-market leaderboard defaults.How to check: per-market retention by hero proof-type; win-rate of the market-matched hero. P1 → Hero Narrative + Year-in-Review Retention
anomaly-alerts-outsourcer Outsourcer receives only plateau / teacher-change / absence-pattern / milestone alertsJob: Outsourcer (legitimate strategic delegation, not failure mode).Evidence: corpus — long-term loyal users leave minimal reviews (Ahmed Zelikha "excellent choice," Pantyusha5 "zero problems"). Jobs §6 (Outsourcer).Reference: Apple anomaly-only notifications (unusual heart rate, long-period no activity). Slack @channel-only-for-critical pattern.How to check: Outsourcer-cohort alert open rate (should be ~70%+); cohort retention should match or exceed engaged cohort. P1 → Weekly Digest + alerts Retention
placement-test-rigorous Accurate starting-level assessment at onboarding (feeds the roadmap)Job: Navigator + Auditor.Evidence: corpus — Nejat (stuck level), pelma pelma (no test), Maria Angels (wrong level). Reference: Duolingo placement test. Khan Academy diagnostic.How to check: mis-placement rate (level-change requests within 4 weeks) should drop >50%. P1 → Onboarding flow Activation
parent-anxiety-reduction-foreground Every surface tested: does this raise or lower parent anxiety?Job: Reassurance-Seeker .Evidence: theory — Parental Educational Anxiety Scale (Guo et al.); high-anxiety markets RU, TR. Uniform positive feedback paradoxically amplifies anxiety.Reference: Noom behavioural framing (anxiety-aware copy). Headspace calming tone defaults.How to check: PEAS-lite 3-question survey at month 1 and month 3; score should decrease, not increase. P1 → Design principle Trust
trust-audit-before-launch 5-question "what broke trust?" survey of churned parents before hub launchJob: all 6 (foundation).Evidence: synthesis §7 + §11 validation roadmap. Reference: Y Combinator cancellation survey template. Netflix churn survey.How to check: pre-ship research, not product feature. P1 → Research (pre-ship) Trust
platform-reliability-transparency Surface system status — "was the platform down?", lesson reschedules, permission-bug resolutionsJob: all 6 (foundation layer 2).Evidence: corpus — Filiz (TR) "inaccessible for days"; Oya Canli progress-button bug. GAPS.md P0. Trust-stack §7 layer 2.Reference: Stripe status page. GitHub status page. Atlassian statuspage.How to check: trust-score lift after incident transparency vs silent recovery. P1 → Settings + notifications Trust
first-week-experience First-week experience is a distinct design target, not inheritance from onboardingJob: Navigator + Reassurance-Seeker (activation window).Evidence: corpus — first-impression signals scattered; no single synthesis of ideal first-week journey. Cambly Kids structured onboarding → trial → plan pattern.Reference: Duolingo first-7-days activation path. Headspace first-session guided tour. Superhuman 30-min onboarding.How to check: day-7 return rate; trial-to-paid conversion rate. P1 → Onboarding + first-week states Activation
activation-metric-definition Define the single KPI that proves the hub worked (7-day return + first-moment-shared)Job: all 6 (measurement layer).Evidence: synthesis §10 — candidates: 7-day return rate, time-to-first-progress-share, parent NPS at month 3. Currently no single primary metric.Reference: Facebook's "7 friends in 10 days" activation metric. Duolingo's day-7 return. Slack's 2000-messages-in-team activation.How to check: pick one; align product + analytics; lock pre-launch. P1 → Analytics (pre-launch) Activation
5 P2 hypotheses — valid but lower priority or edge case
multi-child-support Side-by-side child panels; no child-switcher as primary UXJob: all 6 (households with two or more children).Evidence: persona — 🇫🇷 Lyon has two children at different levels; child-switcher dropdown actively hurts. Corpus has less visible signal (single-child reviews dominate) but design-logical extension.Reference: Seesaw default-multi-child view. Apple Health family view. Spotify family account.How to check: per-child opens vs per-family; households with two or more children should scan both children per session. P2 → Information Architecture Activation
non-english-parent-digest Weekly digest in parent's language, not child's (do not assume both parents speak English)Job: all 6 (non-English-fluent parents).Evidence: persona — 🇩🇪 Munich edge case, school English, cannot read English weekly digest. Theory: accommodate the weakest-position parent, raises floor for all.Reference: Duolingo language-of-instruction parent setting. Strava locale-based notification copy.How to check: digest open rate by market — spike in DE/FR/IT if localised copy shipped. P2 → Weekly Digest Activation
game-world-narrative-elevated Surface Alex / Luna / Astro / Wisword character arcs to the parent narrativeJob: Reassurance-Seeker + Co-Pilot.Evidence: Cambly Kids competitor audit: character-driven curriculum narratives (Ann/Peter, Daisy/Emma/Sammy) make the child's week legible to the parent.Reference: Cambly Kids character thread. Duolingo character story mode.How to check: A/B character-thread digest vs metrics-only; does shareability with grandparents/family improve? P2 → Weekly Digest Retention
streak-with-recovery Child-side streaks with forgiveness built-in (Duolingo pattern, not Headspace)Job: Reassurance-Seeker (parent sees the child is consistent).Evidence: Duolingo streak + streak-freeze keeps emotional investment without punitive feel. Headspace streak-reset-to-zero creates shame, pushes cancellations.Reference: Duolingo streak freeze + weekend freeze. Strava "rest day" vs "missed day" framing.How to check: streak-break emotional response — if parents report child crying over broken streaks, redesign recovery path. P2 → Child-side (not hub) Retention
reassurance-investment-toggle One interface, two reading modes — quick reassurance vs deep investmentJob: Reassurance-Seeker (default) + Auditor + Navigator (on demand).Evidence: parents operate in two modes; most dashboards force a choice. Linear triage/focus; Stripe role views.Reference: Linear triage/focus toggle. Asana personal/team view.How to check: mode-switch usage pattern over 30 days — do the same parents switch mode across journey stages? P2 → Information Architecture Trust

06 Eight Surfaces

The atlas lists 35 live hypotheses. They don't become 35 screens. Of the 35: 25 fuel the 8 surfaces below; 3 stay as design principles (they shape every surface — so can't be one); 7 stay as cross-cutting concerns (onboarding, research protocol, nav scaffold). Each surface bridges specific data sources to specific jobs. §07 shows how these surfaces arrange into navigation; §08 shows how they fire across the parent journey.

01 Hero Narrative — the top-left 5-second read +

The first thing every parent sees. One specific, checkable sentence about the child's week, with a collapsed evidence trail below for Auditors. Defaults to the market-appropriate proof (Moscow → school ranking, Milan → €/lesson, Istanbul → travel confidence, Warsaw → CEFR). During a plateau, auto-shifts to the comprehension signals that are moving.

Hypotheses.ai-narrative-hero (P0) · progress-narrative-depth (P0) · can-do-statement-operationalized (P1) · trajectory-over-point-in-time (P1) · child-enjoyment-signal (P0) · proof-layer-culturally-tuned (P1).

Reference. GitHub Primer "sentence-first disclosure" + Apple Screen Time two-line summary + Duolingo Path next-step framing.

Default. Collapsed to headline sentence; chevron reveals evidence; further chevron opens full audit trail.

02 Plateau Protocol — a pre-written message sequence +

Not a screen — a triggered protocol. Detects a plateau (engagement normal, visible output stalled 14+ days) and fires a pre-registered sequence: in-app banner + email + WhatsApp, naming the plateau, explaining the gap between understanding and speaking, offering a bounded diagnostic (one-minute video comparison), committing to a post-plateau check-in. Cheapest high-impact fix in the product.

Hypotheses.plateau-protocol-proactive (P0) · invisible-metric-during-plateau (P1).

Reference. Noom "consolidation phase" reframe. Strava "rest day" messaging. Duolingo streak-freeze framing.

Default. Silent until triggered; one-time per plateau event; user can request check-in earlier via single click.

03 Post-Lesson Card — the moment after the lesson ends +

Appears after every completed lesson. Three elements in one card: (a) calibrated teacher feedback — one strength, one area to work on, a one-sentence focus for next week, teacher's name always present; (b) a bounded thread for the parent to send up to two sentences of pre-lesson context per week, teacher acknowledges; (c) the recorded lesson replay with timestamp controls. The only place the teacher bridge becomes visible to the parent.

Hypotheses.teacher-feedback-calibrated (P0) · teacher-bidirectional-channel (P1) · lesson-recording-accessible (P1).

Reference. Grammarly post-quiz breakdown (calibrated green/yellow/red) + Superhuman reply-later note (bounded async) + Cambly Kids lesson archive (replay model).

Default. Surfaces automatically in the lesson-history list and via push notification after each lesson. Parent reply limited to 2 sentences per week; teacher "acknowledged" (not free-form reply) to protect teacher workload.

04 Weekly Digest — the push product surface +

For the 30% of parents (Outsourcers) who never open the dashboard, the Friday/Sunday digest is the Value Hub. Channel per market — WhatsApp (MENA, LATAM, ES, IT, TR), email (DE, FR, UK), Telegram (RU-diaspora). Non-English-speaking parents get it in their language, not the child's. Content: the same Hero Narrative + three can-do bullets + one teacher highlight.

Hypotheses.anomaly-alerts-outsourcer (P1) · non-english-parent-digest (P2) · game-world-narrative-elevated (P2).

Reference. ClassDojo Friday weekly report (donut chart + written summary). Lingokids AI-generated weekly report. Strava weekly summary email.

Default. On by default in parent's stated channel preference; Outsourcer cohort receives only anomaly alerts (plateau / teacher change / absence / milestone).

05 Moment Capture — real-world → record +

The parent's way to record "something happened in English today." Voice note, photo, or short text → structured record → threaded into the AI narrative and year-in-review. Removes the reason 7 of 8 personas keep their own tracking. One tap from the dashboard; also via the mobile share sheet if the parent is at the Greek restaurant right now.

Hypotheses.real-world-moment-capture (P0).

Reference. Seesaw artefact-capture model (child-facing, same mechanic). Apple Journal voice/photo entry. Day One journal.

Default. Surface exists from day one; first prompt fires after first travel/weekend the child has logged a lesson near.

06 Curriculum Roadmap — the Navigator's first-class surface +

Where the child is, what's next, where difficulty steps up, how it maps to the local school. Turns the curriculum from a black box into a navigable staircase. Fed by a placement test at onboarding; per-market school-alignment overlay on top (RU grades, FR Bac, IT/ES Cambridge YLE, PL CEFR).

Hypotheses.curriculum-roadmap-visible (P0) · school-alignment-per-market (P1) · placement-test-rigorous (P1).

Reference. Duolingo Path (visual forward journey). Khan Academy skill tree. Strava training plan.

Default. Visible from week one; school-alignment overlay activates based on parent-reported school country.

07 Year-in-Review — the shareable annual artefact +

Annual synthesis that closes the renewal debate for hesitant parents and gives Reassurance-Seekers something to send to grandma. Tuned per market — subtle (DE, FR), ranked/certified (RU, PL), celebratory/family-forward (IT, TR, ES). Shareable to family by default; public sharing opt-in, off by default in JP/FR.

Hypotheses.year-review-artefact (P1) · proof-layer-culturally-tuned (P1, contributes).

Reference. Spotify Wrapped, Duolingo Year, Strava Year, Reddit Recap. Shared pattern: celebrates identity, not just performance.

Default. Generated once per year in December; shareable within family via link; public share requires explicit opt-in.

08 Settings & Defaults — the trust-stack foundation +

Where trust is earned or lost. Transparent billing (pause, cancel, next charge date visible). Teacher-continuity policy (consult before change, not notify after). Interface language set for the parent, not the child. Market-gated defaults for peer-ranking and cost-per-outcome. System-status transparency ("was the platform down?"). None of the surfaces above succeed if this leaks.

Hypotheses.trust-stack-billing-transparency (P0) · teacher-continuity-policy (P0) · platform-reliability-transparency (P1) · peer-benchmarking-opt-in (P1) · cost-per-outcome-math (P1).

Reference. Netflix cancel-anytime UX. Stripe billing clarity. Atlassian status page. Hinge "introducing your next match" framing.

Default. Always accessible from account menu; pause/cancel one tap away, no dark patterns.

Three hypotheses stay as design principles, not surfaces: progressive-disclosure-by-job, information-bridge-five-silos, dashboard-rare-use-assumption. Seven more are cross-cutting: parent-anxiety-reduction-foreground, trust-audit-before-launch, activation-metric-definition, first-week-experience, reassurance-investment-toggle, multi-child-support, streak-with-recovery. They shape onboarding, research protocol, and the IA, not a single screen.

07 IA

All 8 surfaces from §06 arrange into 5 navigation entries plus a push layer — full coverage of the atlas, not a subset. Current state traced from the §02 audit; proposed tree below. The 3 principles and 7 cross-cutting concerns from §06 shape defaults and behaviour inside these entries — they don't get their own tabs.

Today · current parent account
Six tabs, two apps, progress missing
Parent Account ├── Home ← no progress data ├── Schedule ← booking only ├── Library ← recorded lessons, buried ├── Achievements ← two different UIs in one shell ├── Speaking Practice ← separate product, no hub ├── Referrals └── Account ├── Billing ← 1★-review hotbed └── Settings
Diagnostic · the dashboard is mentioned in ≈2% of 128 real parent reviews. Nothing answers "how is my child progressing?"
Proposed · Value Hub
Five entries, one push layer
Value Hub ├── Home ← 5-second read: how's my child? ├── Roadmap ← curriculum, next steps, school map ├── Lessons & Teacher ← post-lesson card + history ├── Moments & Stories ← real-world capture + year review └── Settings ← trust stack foundation + Push layer ├── Weekly Digest ← Outsourcers ├── Plateau Protocol ← triggered └── Anomaly alerts ← exception-only
Every entry answers one question. Home — "how?"; Roadmap — "where?"; Lessons — "who and what?"; Moments — "does it matter beyond class?"; Settings — "can I trust this?".

Surface × Nav coverage matrix

Where each of the 8 surfaces lives in the navigation. A surface can be primary in one entry and supporting in another — supporting means it appears there as context, not as the entry's reason for existing.

Surface Home Roadmap Lessons & Teacher Moments & Stories Settings Push
01 Hero NarrativeAI · Child Primary — — — — Echo
02 Plateau ProtocolTriggered In-app banner — — — — Primary
03 Post-Lesson CardTeacher — — Primary — — Notification
04 Weekly DigestCross-silo — — — — — Primary
05 Moment CaptureReal-world Entry point — — Primary — Prompt
06 Curriculum RoadmapAI · School — Primary — — — —
07 Year-in-ReviewCross-silo — — — Primary — Delivery
08 Settings & DefaultsTrust layer — — — — Primary Policy-gated
Primary The entry's reason for existing — where the surface is the load-bearing element Support Appears as supporting context, entry-point, notification, or echo — Not present in this entry

The full customer journey map — six stages from onboarding to renewal, showing today's pain and the Value Hub intervention at each step — lives in §08. Three hi-fi screens (Hero Narrative, Post-Lesson, Plateau) plus multiple concept directions live in §09.

08 CJM

Six stages, before-and-after, across roughly a year of product relationship. Parents don't fail because the layout is wrong — they fail because the right moment has no interface on it. The CJM shows which of the 8 surfaces fires when; windows come from the quote corpus and the persona interviews.

Per-stage deep dive

Each stage: what breaks today (with corpus quote), what the Value Hub replaces it with, which surfaces fire, the metric that confirms the intervention. Plateau (months 2–8) grounded in British Council research, corpus cancellation stories, and 4 of 8 persona interviews where cancellation was seriously considered.

01 Discover Days–weeks before signup Consideration
Today · what breaks

Proof of outcomes is missing from the marketing surface. The signals parents trust — other families' stories, shareable artefacts, public milestones — don't exist in a form they can find before signing up. The trust stack isn't visible either: billing terms, teacher-change policy, cancellation clarity all live behind the paywall. Parents decide on vibes, then discover the reality.

"Parents often rely on report cards to monitor their children's academic progress. If they're getting A's and B's, they don't look at anything else." 🇺🇸 Araceli Simeón — Parent Organization Network, EdSource
Value Hub · intervention

Family-shared Year-in-Review artefacts and a public trust-stack policy (billing, pause, teacher-continuity) bleed the hub outward into the discovery channel. Hero Narrative and Roadmap appear as marketing previews.

Year-in-Review Settings · policy Hero · preview Roadmap · preview
Metric Trial sign-up rate from family-shared Year-in-Review links. Secondary: policy-page dwell time as an indicator of trust-stack scrutiny.
02 Trial One free lesson, decision window Activation
Today · what breaks

The trial lesson is generally strong — teacher quality appears in ≈65% of reviews, overwhelmingly positive. But the post-trial screen doesn't show what the paid experience will feel like between lessons. Parents see a booking flow and a pricing page, not a dashboard they can judge.

Billing terms at trial-end are the second-largest 1★ topic in the corpus. Parents say "yes" at trial, then hit terms they didn't expect.

Value Hub · intervention

Immediately post-trial, the Hero Narrative renders with the child's first data point ("Mia can introduce herself and name three objects"), the Post-Lesson Card fires as a real artefact, and Settings exposes billing transparently — one tap from the "start your plan" CTA.

Hero Narrative Post-Lesson Card Settings · trust stack Roadmap · after trial
Metric Trial-to-paid conversion rate. Secondary: post-trial return-to-dashboard rate within 48 hours.
03 Onboard Week 0–1 Early drop-off
Today · what breaks

Placement is loose. Parents report the child being placed at a wrong level ("stuck" or "too easy"), which erodes confidence in the platform before a routine forms. The "Progress" button is reportedly gated behind an audio-permission prompt — a literal wall between the parent and the feature they most want.

"When I click on the progress button in the bottom menu to see the progress of my child, it automatically launches a permission request for audio recording?" 🌍 Oya Canli — Google Play

Learning plan expectations are set vaguely — ALFAZANI TALHA and others note "no clear plan" as a trust-break within the first weeks.

Value Hub · intervention

A rigorous placement test feeds the Roadmap — parent sees the staircase with position + next three steps on day one. Hero renders a first-week state (calibrated for sparse data). Settings is configured actively: teacher-continuity policy, parent language, market-appropriate defaults. Moment Capture seeds the habit after the first weekend.

Curriculum Roadmap Hero Narrative Post-Lesson Card Settings Moment Capture · seed Digest · first delivery
Metric Day-7 return rate; mis-placement rate (level-change requests within 4 weeks); first-digest open rate.
04 Week 1–4 Routine forming Settling
Today · what breaks

Lessons happen, but the channel between them is empty. Notifications are uniformly positive regardless of how the lesson went — the clearest trust-break named in the corpus. Parents who do check the dashboard find it "little information" and begin to doubt.

"Little information each class, I feel he is not improving at all." 🌍 Anas — Trustpilot Global, 2★

The one praised feature is recorded lessons — parents use it as a DIY verification tool. That signal is real but unscalable: it puts the burden of interpretation on the parent.

Value Hub · intervention

Three firing surfaces shape the weekly rhythm: Hero on open, Post-Lesson Card after each lesson with calibrated teacher feedback (one strength + one area + one next-week focus — never uniformly positive), and the Weekly Digest Fri/Sun in the parent's channel. Moment Capture stabilises here — prompts fire after weekends or travel.

Hero Narrative Post-Lesson Card Weekly Digest Moment Capture
Metric 7-day return rate; Post-Lesson read rate; Moment Capture activation (target ≥ 50% in 28 days); digest open rate per market channel.
05 Plateau Month 2–8 · highest-churn window Highest churn
Today · what breaks

The platform is silent. The plateau is a documented phase of second-language acquisition — visible output stalls while comprehension keeps developing — and it's the single moment every interviewed parent described. Four of the eight persona interviews said they almost cancelled during it. Reassurance came from outside the product: a relative, an outside teacher, a school meeting.

In the corpus, this is the stage where parents write stories like Nejat's (🇹🇷, kursfinder) — "12,553 TL, stuck six months" — and cancel.

Value Hub · intervention

The cheapest, highest-impact intervention in the product. The Plateau Protocol fires across in-app + email + WhatsApp: names the plateau, explains understanding-vs-speaking, offers a one-minute video diagnostic, commits to a post-plateau check-in. Hero shifts to metrics that are moving during a plateau (comprehension, recognition); Moment and Digest hold the line with real-world wins.

Plateau Protocol Hero · shift Digest · reframe Moment · holds line Roadmap · next-step map Settings · pause option
Metric Mid-month cancellation attempts during the plateau cohort (target ≥5 pp drop vs silent control). Secondary: Plateau Protocol click-to-diagnostic rate.
06 Renew Annual or monthly decision moment Decision moment
Today · what breaks

Parents who renew today do it because external proof showed up — a school teacher noticed, a relative commented, a summer trip confirmed the progress. That's a win for the product and a warning: the renewal defence happens outside the platform. For parents whose external proof doesn't arrive, there is nothing on the dashboard to hold them.

"My daughter has been taking this course for over three years and I must say the improvements are visible, even acknowledged by her school teachers." 🇮🇹 Dorotea — Trustpilot Italy, 5★

Billing transparency becomes load-bearing in this stage: auto-renewal surprises are a top-three 1★ reason across markets.

Value Hub · intervention

Year-in-Review is the in-product renewal defence — shareable to family, market-tuned (celebratory in IT/TR/ES; subtle in DE/FR; certified/ranked in RU/PL). Hero telescopes to a 12-month trend so decisions rest on trajectory, not the last bad week. Settings keeps billing transparent up to the renewal date.

Year-in-Review Settings · transparent renewal Hero · 12-mo Moment · feeds review Roadmap · next year
Metric Month-3 retention uplift (target +5 pp, worth ≈ €6M ARR at current active cohort). Secondary: Year-in-Review share rate to family; renewal-billing 1★ reason frequency (target −50% in 90 days).
Reading per-stage together

The map has one obvious tell: the silent stages in today's column are exactly the moments where the product loses parents. The Value Hub's job is to put a surface at every silent moment — not to prettify existing screens.

09 Hi-Fidelity

All 8 surfaces drawn at fidelity — Hero Narrative, Post-Lesson Card, Plateau Protocol, Weekly Digest, Moment Capture, Curriculum Roadmap, Year-in-Review, Settings & Defaults. Every parent job (see §03) now has first-class surface coverage; §08 CJM shows which surface fires when.

Each surface is explored as Primary + 3 concept variants — not style dressing, but genuinely different approaches to the same job. 32 hi-fi frames total. The Primary is the proposal; concepts stress-test alternative voices, channels, and formats before committing.

  • Hero Narrative (Home — the 5-second read) — Primary Editorial Evolution (Fraunces headline + one checkable sentence + collapsed evidence trail), plus Ask Mia (child-voice framing), Conversation Q&A (parent questions answered inline), Journey Map (trajectory-over-snapshot).
  • Post-Lesson Card (the moment after the lesson ends) — Primary + four concepts: Editorial, Chat, Portfolio, Audio. Calibrated teacher note, bounded parent↔teacher thread, recorded replay.
  • Plateau Protocol (the triggered sequence) — Primary + four concepts. Detects a plateau, names it, offers a one-minute diagnostic, commits to a check-in. Not a screen — a protocol.
  • Weekly Digest (the push that reaches non-openers) — Primary Data Postcard (WhatsApp/TG image unfurl with email fallback) + Editorial Letter, Story Scroll, Channel-Native. Reaches the 70%+ of parents who never open the dashboard.
  • Moment Capture (real-world evidence) — Primary One-Tap Moment (IG-story viewfinder) + Journal Weave, Teacher-Caught, Snap & Say. Threads parent-captured moments into the AI narrative.
  • Curriculum Roadmap (Navigator's map) — Primary Staircase (Duolingo Path model) + Skill Forest, Term Calendar, School Alignment. Position + next 3 steps, per-market benchmark overlay.
  • Year-in-Review (the renewal artefact) — Primary A Letter to You (Fraunces long-form) + Wrapped Reel, Certificate, Year in Moments. Shareable to family; renewal defence inside the product.
  • Settings & Defaults (trust stack) — Primary Transparent Dashboard (Stripe-style levers) + Conversational Setup, Policy-First, System Status. Billing transparency + teacher-continuity policy + market-tuned defaults.

Open the Hi-Fi page in Figma →  ·  Coverage Matrix →

Design system

All visual tokens — color, typography, effects, spacing, radii — extracted into Novakid-DS, a separate Figma library published as a Team Library and consumed in the product file. Forked from Figma SDS (Community, April 2026); Brand primitives reverse-engineered from novakidschool.com. Novakid-specific rebuilds of Brand / Highlight / Green / Slate / Yellow primitive ramps for monotonic progression and unified hue; added Brand Deep (identity purple), data-viz palette, Focus Ring effect (WCAG 2.4.11), Pressed / Selected / Skeleton / Inverse / On-Brand-Deep tokens, LH/LS primitives, responsive Typography with Mobile/Tablet size overrides. 200+ variables across Light/Dark modes, 19 text styles, 14 effect styles. All status surfaces pass WCAG AA contrast.

Open Novakid-DS · Foundations →

10 Live prototype

Statics stop answering questions the moment the reviewer wonders “what happens when…?” So the Hi-Fi moves into code: a working, keyboard-navigable prototype that replays eight weeks of one child’s journey, adapts to five markets, and keeps every design decision tied to the quote that justified it.

→  novakid-prototype.pages.dev
Desktop-first · 1440+ · opens into a market picker. No sign-in.

What it is

  • Five surfaces live and interactive — Hero Narrative (Dashboard), Post-Lesson Card, Plateau Protocol, Weekly Digest (three-channel), Curriculum Roadmap (staircase).
  • Five markets — 🇷🇺 Russia / 🇹🇷 Turkey / 🇮🇹 Italy / 🇵🇱 Poland / 🇪🇸 Spain. Currency, school-grade alignment, primary digest channel, cultural proof type, and teacher-accent hint all re-bind per market. Same Mia, five market realities.
  • Eight-week scrubber — replay Mia’s trajectory from W1 onboard through the W4 plateau trigger to the W6 recovery and W8 milestone. Data shifts live: CEFR %, speaking minutes, lesson count, vocabulary, teacher notes, plateau banner.
  • Rationale mode (hotkey ?) — every design decision is outlined and hover reveals the parent quote from the 128-quote corpus that justifies it. Research bound to pixels; no decoupled deck.

How to walk it in 90 seconds

  1. Open the URL, pick a market (Turkey is closest to Mia’s profile).
  2. Scrub to W4 — plateau banner fires on the dashboard. Click it to enter the Plateau Protocol.
  3. Press ? — every bound element highlights. Hover the hero narrative, the plateau SVG, the cost-per-skill stat.
  4. Switch markets via the top-right chip — school alignment, currency, and digest defaults re-bind. Same Mia.
  5. Open Digest — three channels (email, WhatsApp, Telegram) side-by-side. The default one reflects the active market.

Keyboard shortcuts

1 2 3 4 5  switch market  ·  ← →  scrub week  ·  ?  toggle rationale mode

Why build it

Because the assignment evaluates strategy and IA before pixels, and a prototype is the fastest way to prove that the design is a system, not a set of mockups. It forces every surface to resolve real data, every interaction to have a contract, every claim about “moment-first design” to be testable by clicking. The prototype also serves as an honesty check: if a persona toggle would have required a magic “the product knows who you are” layer, it got cut. Market toggles stayed because locale is a real signal a product can read.

Stack: vanilla ES modules, CSS custom properties extracted from Novakid-DS (Figma), no build step. Deployed to Cloudflare Pages. Source and data are on disk; quotes, timelines, and surface specs are JSON fed into render functions. Open the prototype →

— Appendix

Expand full research trail
01-audit/
heuristic-evaluation.md 6-lens firsthand audit, 23 Novakid screenshots, overall 1.5/5
trial-lesson-notes.md Real Novakid trial lesson: auto-assigned Level 3, no baseline assessment, Game World observations
formal-ux-audits.md Nielsen 10 + Don Norman 7 + Cognitive Walkthrough (1/5 tasks fully completed, 2 partial, 2 failures)
parent-voices-analysis.md 128 real parent quotes, 15+ platforms, 6 design imperatives, cultural analysis
quote-verification.md 4-tier honesty audit: Verified / Plausible / Paraphrased / Synthetic
trustpilot-systematic.md 200+ Trustpilot reviews, 15 specific quotes with page URLs, teacher perspective
trustpilot-parent-quotes.md 40 verified Trustpilot Global quotes, 2024–2026
trustpilot-regional-parent-quotes.md 22 quotes across 6 regional Trustpilot domains (TR, ES, PL, IT, DE, FR)
russian-platform-parent-quotes.md 12 quotes from Russian-language platforms (kursfinder, progbasics, tutortop, azbukakursov)
alt-platform-parent-quotes.md 19 quotes from Google Play, App Store, sikayetvar.com, Ekşi Sözlük
platform-current-state.md Novakid facts: 940K parents, 1,500 micro-skills (cited), curriculum, Game World
competitor-analysis.md 10 steal-worthy patterns, 7 verified + 4 partial + 1 retracted
competitors/INDEX.md 86 screenshots across 4 competitors + master index
competitors/breadth-scan.md Public-source verification of 12 pattern claims
02-research-inputs/
synthesis.md Primary synthesis. Information Bridge axis: five silos, six jobs, five tensions, trust stack, plateau protocol, inline validation
archetypes.md 6 JTBD jobs (Reassurance-Seeker / Auditor / Investor / Navigator / Co-Pilot / Outsourcer), SDT/EVT-anchored
cultural-analysis.md Standalone cultural layer: 7 markets × proof-types, Hofstede dimensions, market-specific hero surfaces
interview-guide.md 18-question interview — mental models of progress, no Novakid in stimulus
additional-parent-quotes.md 35 parent quotes from EdSource, Russian aggregators, Ekşi Sözlük
theoretical-frameworks.md 9 frameworks: SDT, EVT, JTBD, PEAS, British Council plateau, Hofstede, dashboard UX, Baumrind, behavioural economics
open-coding-analysis.md Inductive open coding of 128 quotes → 7 emergent themes, 6 jobs, 5 narrative-axis candidates
persona-interviews/ 8 persona interviews: Warsaw/PL, Istanbul/TR, Milan/IT, Moscow/RU, Munich/DE, Madrid/ES, MENA, Lyon/FR
parent-psychology.md EdTech parent psychology, plateau theory, churn drivers
dataviz-patterns.md Dashboard visualization patterns (Apple Rings, progressive disclosure, Tufte)
inspiration/skyeng/ Skyeng reference: 40 screenshots of kids/parent dashboards + deep analysis (skyeng-deep-analysis.md)
Credibility audit trail
credibility-map.md 62 claims checked: 28 verified, 13 partial, 14 unverified, 7 retracted
GAPS.md 12 audit gaps from Yannis-persona critique (P0 + P1)