A cross-read of three sources that had drifted apart: the management documents (emealia-mgmt/), this documentation set, and the shipped implementation (emealia-app/). Everything below is verified against code and the implementation plan, not against intent.
The product is further along than any of its documents say, and its market is worse than any of its documents say. Those two facts pull in opposite directions and they define this moment.
On the build side: the core loop is complete. Free-text logging, honest ranges, band states, the no-failure guarantee, all-details, history, statistics, weight, day context, offline queue, trial + Stripe, GDPR export/erasure — shipped and tested. Two features the August competitor analysis flags as strategic and unmatched in the field — nutrient contributions and multi-meal logging — are not roadmap items. They are in production. A third, the weekly nutrient view, which the same analysis lists as a "strong candidate" to scope, shipped as stories 2.11a–2.11c before the analysis was written.
On the market side: between the July and August analyses, MyFitnessPal absorbed Cal AI, Apple began building nutrition logging into the platform, and easy AI input stopped being a differentiator. The strategy documents, the landing page, and the pitch all still lead with easy. That is now the one claim the two largest distributors in the category can match for free.
Stop leading with ease. Lead with the frame. "Meal tracking made easy" is a claim MyFitnessPal can now match. "Eat by feel, not by math" cannot be matched by anyone whose product is a number — and Apple, whose entire behavioural vocabulary is rings and goal closure, structurally will not ship it.
The implementation plan is the only source that reflects reality. This documentation set described a July product; emealia-mgmt/3-design.md still specifies a colour palette the app replaced weeks ago; the August competitor analysis assesses "feature additions" as if two of them were proposals. All three have now been reconciled here.
| Claim in the older documents | Reality in the code |
|---|---|
| Band contributions are a promising new addition to assess | Shipped core/nutrient_contributions.py, GET /nutrients/{n}/contributions, NutrientContributionPanel.vue — day and week windows, midpoint shares, top-N + "the rest" |
| Multi-meal logging is a new addition to assess | Shipped stories 18.7–18.9: one AI call segments a text into up to 8 meals, N drafts, one credit per batch, per-card confirm and discard |
| A weekly nutrient view is a "strong candidate — medium cost" | Shipped GET /weekly, Today ↔ This-week tabs, and a last-7-days tab on All details |
Design palette #1A201C forest-green ground | Replaced Deep Oceanic Slate: #11161B ground, #1C242C surface, peach accent retained. This doc set now matches. |
| Suggestions layer "underdeveloped" | Accurate — generation exists, delivery (story 4.5) is unbuilt. The one place the documents were not too pessimistic. |
The July analysis treated Cal AI as proof that a low-friction upstart beats incumbents. That window closed by acquisition, not by competition. MyFitnessPal now owns both the largest food database and the best consumer AI-input funnel, and Apple is commoditising basic logging to zero for a very large installed base.
What survived intact is the frame — ranges instead of point numbers, bands instead of a budget, no red, no streaks, no verdict. Nobody credible competes there. Eight rows of the competitive feature matrix are uncontested, and those eight rows are the entire product thesis. Full analysis →
MyFitnessPal moved photo logging, recipe import and per-meal macro targets behind Premium in May 2026 and caught a consumer class action for still marketing itself as "free". Yazio's German list price reached €83.90/yr with rotating discounts and a trial available only on the web. Both write emealia's first two comparison pages for it, honestly. Marketing plan →
Ordered by how much damage each does if it is not addressed. Everything here is actionable now.
| # | Gap | Why it matters | Severity |
|---|---|---|---|
| 1 | The public site advertised a 7-day trial the product does not offer. Six occurrences of "7 Tage" on the German landing page against a shipped UsageMeter of 30 AI credits. Corrected 2026-08-06 — the rebuilt landing page states the credit trial everywhere. The design doc's "7-day-trial CTA" copy standard still needs the same correction. |
A direct competitor is currently facing a class action over inaccurate "free" marketing. This was a legal exposure, not a copy nit — and it undermined the one-price-honest positioning the whole strategy rests on. | Fixed |
| 2 | Contribution copy has no shipped guardrail. The feature is live; the ordering, labelling and "still estimating" rules exist as analysis, not as acceptance criteria. | Attribution is one word away from blame. "71% of your fat came from the pesto pasta" is an accusation if the copy implies fault. This is the highest ED-safety risk surface in the product and it shipped without a written copy contract. | High |
| 3 | Suggestions do not reach the user. GenerateDailySuggestion and the eligibility policy are built and tested; story 4.5 — persistence and delivery — is not. |
Daily tip open rate is a named KPI (≥15% at M3) and a kill condition. It cannot be measured, because no tip is delivered. Meanwhile every competitor is shipping an AI coach. | High |
| 4 | Both human gates are unrun. G1 (5 testers understand bands unaided) and G2 (30 real meals, correction rate <40%). | G1 is stated as required before the Today UI is finalised — the UI is finalised. G2 gates acquisition spend. Neither can be delegated to an agent. | High |
| 5 | The medical / general-wellness disclaimer is blocked on counsel (H10). Research and draft EN/DE copy exist; sign-off does not. | Launch blocker. Positioning as a general-wellness product rather than a medical device depends on wording nobody has approved. | Blocker |
| 6 | EU AI Act Art. 50 has no line item. Transparency obligations became enforceable 2 August 2026. | emealia complies in substance already — every estimate is labelled an estimate and shown as a range. It lacks the explicit disclosure placement at onboarding and at the meal-input surface. Cheap to fix, and a strong DE marketing argument next to DSGVO. | Medium |
| 7 | Auth is documented two ways. Strategy says social-only, no email/password (founder decision 2026-07-08). The design doc's login-screen standard still says "email + Google + Apple together". | Social-only is the anti-abuse mechanism behind the soft paywall. An email path reopens it. One of the two documents is wrong and the code should settle it. | Medium |
| 8 | Three features nobody else has are not marketed. Contributions, multi-meal logging and the weekly view are invisible on the public site. | "Log your whole day in one sentence" is verifiably unique in the field right now and directly answers the category's best-documented complaint. Zero build cost. | Medium |
| 9 | Eating out, shared dishes and non-Western cuisines already work and are never claimed. Free text handles "the chicken tikka masala at the Indian place" natively; a database product structurally cannot. | The structural failure mode of every competitor, answered by accident, unmentioned. In Germany the cuisine point reaches a large population no competitor addresses. | Medium |
| 10 | Observability and AI-cost optimisation are thin. Epic 15 is 2/4; epic 21 is 1/8. | "AI cost > €0.08/user/month" is a kill condition. Detecting it requires the instrumentation that is not finished. | Medium |
daily_summaries. No AI call, no tokens, no hallucination surface. Features that raise perceived intelligence while lowering COGS are rare.In order. The first three cost nothing to build.
| # | Move | Cost | Owner |
|---|---|---|---|
| 1 | 3-design.md's copy standard | Minutes | Founder |
| 2 | Done | Founder | |
| 3 | Done | Founder | |
| 4 | Run G1 and G2 — and add the contribution list to the G1 comprehension test. If testers read it as blame, the copy is wrong and shipping it as-is undoes the band work. | Two sessions | Founder |
| 5 | Ship the deterministic tip library as story 4.5's payload — 20 evidence-backed rules that cost nothing to run, cannot hallucinate, and are auditable under AI Act Art. 50 | ~1 epic | Build |
| Source | What it holds |
|---|---|
emealia-app/ | Implementation monorepo — backend, frontend, and the 29-epic implementation plan that is the single source of truth for what gets built |
emealia-mgmt/ | Strategy, design, onboarding and market-analysis documents |
emealia-web/ | Public landing page, the interactive prototype, and this documentation set |
| Prototype | 9-screen clickdummy |