Epistemis
What's new

What has changed

Current version: 1.77.2. Only changes you can actually notice while using the app are listed here; internal maintenance is not.

v1.77.2 — Expanded 3D atlas, viewer fixes and a lighter start

  • 3D atlas: 414 scenes in 55 packs, including microbiology models and clinical correlations.
  • Fixed camera, layer, transparency, clipping, selection and quiz controls in the viewer.
  • Downloaded packs retain the metadata and translations needed for offline study.
  • Accessibility and display improvements on mobile, tablet and desktop. Update availability depends on the platform.
  • Lighter start on modest phones: when the device measures slow, the app stops preloading content “just in case” from that very launch, and search prepares its index with less work.
  • Security update for the mobile apps and server.

v1.77.0 — 3D auscultation, shift tools and clinical practice

  • 3D auscultation: place the stethoscope on a 3D chest and listen to 56 heart, lung and abdomen cases, with bell and diaphragm, manoeuvres, breathing, optional skin and a quiz mode. The listening points were measured on the model’s own anatomy.
  • Shift tools: WHO partograph, obstetric emergencies with weight-based doses, WHO contraception eligibility criteria (2025), well-child visits (Colombian Resolution 3280), guided IMCI consultation, IV drip rate, dilutions, fluid balance, a printable prescription and a genogram.
  • Practice and career: clinical note templates, the vaccination schedule (which vaccines a patient is missing), diarrhoea and rehydration, prenatal follow-up from the last period and a guide to internship, rural service and residency.
  • 64 new scales, including Score MAMÁ, Family APGAR and Herrera and Hurtado, each with its suggested course of action.
  • Custom sessions of up to 100 questions, with status filters and a timed exam mode.
  • Educational material; independent clinical review is still pending.

v1.76.0 — Starting assessment, your own syllabus and consistency without punishment

  • Starting assessment: five questions and a recommendation on where to begin. My syllabus orders your study around your own syllabus.
  • Every topic has three lenses —Learn, Practise and Clinical— and search now finds all 5,103 bank questions, which it previously did not index at all.
  • Consistency without punishment: Home counts active days instead of streaks, and the streak reminder is optional.
  • Easier for beginners: Home, search and the guided tour, with larger buttons.
  • Premium states in numbers what it includes and what is always free (clinical reference is not behind the wall); paid features are labelled as such. There is still nothing to buy: subscriptions are not active.
  • Question bank corrected: 27 USMLE questions had a shifted answer key (they marked the neighbouring option as correct), explanations no longer cite a different letter after the options are shuffled, and 292 explanations cut off mid-figure were rebuilt from the full explanation.
  • English reviewed across all content: guides, exams, atlas and glossary.
  • Educational material; independent clinical review is still pending.

v1.75.9 — The atlas, checked image by image against its source

  • The technical audit was green and nothing checked that each image showed what its label says. Each one was compared with the original source caption: 110 were withdrawn (a "small-cell carcinoma" that was not small-cell, a "pituitary" that was the pineal gland, "normal" prostate and breast that were carcinomas, 23 biopsies filed as blood smears, 7 veterinary cases and 19 diagrams published as clinical images).
  • 62 labels corrected against the right image (slide versus real specimen, stain, region, modality).
  • 10 of our own diagrams corrected: the O₂–Hb curve labelled the shifts backwards, fetal circulation sent the left atrium to the pulmonary artery and oral bioavailability sat above intravenous, among others.
  • Exams: 87 questions showed images the atlas no longer publishes; they now use only published, reviewed entries.
  • None of these checks is a clinical approval: independent medical review is still open.

v1.75.8 — AI reading of the atlas stops failing on iPhone

  • On iPhone, asking the AI to read an atlas image ended in "Load failed". Since 1.75.7 images are served on demand and the reading looked for the old path inside the app, where it no longer exists. It now uses the same route as the entry itself, and a network failure shows a readable message. On the web and desktop the atlas is bundled, which is why it never showed there.
  • Also fixed a message that was mistranslated ("No se pudo cargar la imagen").

v1.75.7 — Signing in on iPhone no longer hangs with the right account

  • With the correct password, the iPhone app sat waiting. Account initialisation hung on iOS and the last step of every successful sign-in queued behind it. A wrong password failed instantly, which hid the problem. Measured: from over 40 seconds hung to milliseconds.
  • Calculators: grey hint values no longer count as data, and the anion gap was added.

v1.75.6 — Forty Clinical defects fixed, and a sign-in that no longer sticks on "loading"

  • The main Clinical risk was turning a missing value into a statement that looked verified. Renal dosing filled in 60 years, 70 kg and creatinine 1.0 when there was no data; an unknown drug fell back to enoxaparin; "ara" inside "paracetamol" produced a false interaction; SBAR, I-PASS, SOAP and the discharge summary claimed stability without evidence. Forty findings fixed, each with its own test.
  • Signing in no longer sticks on "loading". Google, Apple and email could hang without an error; there is now a time limit and a message.
  • Level of evidence: the card promised grading on every statement but only 7 of 546 had it; it now states the rule that applies. And 188 duplicate atlas slides were removed, which met a quota with copies.
  • Android: the app now draws edge to edge, and the code optimiser works on the whole app again.
  • Educational material; independent clinical review is still pending.

v1.75.3 — The subscription screen, reviewed in both languages

  • The English app explained badly what it charges and where to cancel. “Suscripción activa” came out as “Active registration”, “es una función Premium” as “that’s a bonus feature” —which says the opposite— and “Epistemis charges through no other means” had become “Epistemis doesn’t pay for anything else”. Worst of all, the cancellation path pointed to “Adjustments → your name → Subscriptions”, a menu that does not exist on an English iPhone; it now says “Settings”.
  • The free trial is no longer promised blind. The app announced seven days everywhere by reading its own declared value. It now asks the store, and announces no trial at all if the store says there is no offer — which is the case today on Android and iOS, where those offers have not been created yet. On web and desktop, where they do exist, it is still announced.
  • The Founder program now says the same thing everywhere: two months (not three) and a window running to 31 December 2026 (not 31 October), for the first 200 sign-ups. The months start counting the day billing is switched on, not on your sign-up date.
  • There is still nothing to buy. Subscriptions are not active, no feature is locked and nobody can be charged anything yet.
  • Educational material; independent clinical review is still pending.

v1.75.1 — The English app stops speaking Spanish

  • 139 strings were still in Spanish in Settings with the app in English, across all eight areas —from “Momento formativo” to “Borrar todo”— and another 42 were translated and wrong: “Auto” came out as “Car”, “Listo” as “Ready” and “Cómoda” as “Chest of drawers”. The dictionary already had nearly all of them; the screen simply never asked.
  • Around 45 clinical-centre translations fixed. Among them “Guardia y reanimación” rendered as “Security and Resuscitation” and “GASA ≥1,1” as “ABG ≥1.1”, which confused the serum-ascites albumin gradient with an arterial blood gas.
  • The antibiotic spectrum matrix can be read and can be heard. Its 256 cells now state whether the antibiotic covers the organism or not: a screen reader previously announced only the organism in both cases. And the 120 “does not cover” cells are no longer drawn below the minimum legible contrast.
  • Settings tells the truth about itself. The healthcare-profession field no longer promises to adapt your study plan, because no engine read it; the self-assessment shows the 20 criteria it actually evaluates instead of 10; and the header icon returns to its line on 52 screens where it sat above the title.
  • Educational material; independent clinical review is still pending.

v1.75.0 — On-demand Atlas and whole-body 3D anatomy

  • Images and 3D models now download by module. Guides, questions, calculators, the formulary and sounds stay inside the app; the clinical and 3D Atlas are cached when opened and then work offline. Modules can be removed to recover space.
  • 3D anatomy grows to 414 scenes and 7,551 structure appearances. Its 55 packs cover body systems and regions with search, isolation, hiding, transparency, rotation and structure selection.
  • Remote delivery is audited too. Manifests are versioned and immutable, the cache verifies each pack, and the pipeline excludes 54 non-commercial objects and 3 invalid labels before publication.
  • The privacy policy, terms, store listings and website now explain first download, data use, caching and technical hosting metadata.

v1.74.6 — The atlas multiplies, with provenance and status in plain sight

  • The atlas goes from little over a thousand images to several thousand. Every case now carries its module, provenance, licence, dimensions and a SHA-256 fingerprint in a dedicated catalogue, and packs install per module from the atlas card instead of shipping inside the application.
  • Every case has its editorial status on record, and whatever fails the checks is not published. Part of the new material passed the technical and metadata checks and its status is recorded in the atlas catalogue. Another part is not published at all: it stays in a traceable quarantine for insufficient source resolution, exact duplication or a known contradiction between image and caption.
  • No image has been upscaled or interpolated. Artificially raising the resolution of clinical material can invent anatomical or diagnostic detail, so a case below the minimum is withdrawn rather than enlarged. Four assets were replaced with the official original from their source.
  • None of these checks is a clinical approval: they are software controls over files, metadata and usage rights. Independent medical review remains open.

v1.74.2 — Settings that only claim what the app actually does

  • Switches with nothing behind them are gone. The app does not verify healthcare credentials: the professional role is self-declared, unlocks nothing, and the verified-physician badge is removed. Nor does it connect to any hospital system: FHIR R4 here is the format used to export and import a study case, and it now says so.
  • The figures in Settings are measured. Editorial status is computed over the entries with a declared review date —current, due for review, expired, and median age— instead of four fixed badges; the storage breakdown comes from the space actually used; and configuration rows read the setting that is really applied. If something cannot be measured, the screen says so.
  • Report a specific answer. Every Clio answer now has a report button that opens the form with the answer already attached and sends a real email to support, with category, version and content revision. Previously the report was stored on the device and nobody ever read it.
  • Screen privacy now works. The blur-on-leaving setting was stored as a preference but did nothing; it is now applied when the app goes to the background. Conversations are not used to train models, and the consent toggle that suggested otherwise has been removed.
  • Spanish leaking into the English interface. Fixed the founder-badge notice, the vision readings, the library counters and several clinical labels —“Coma” was being translated as the punctuation mark— and removed from the dictionaries the zero-width spaces injected by machine translation.

v1.74.0 — More robust offline support and a verifiable release

  • Account sync now uses Firestore’s modern persistent cache and coordinates state across tabs; if it is unavailable, Epistemis keeps local progress and its pending queue.
  • The English Home interface and changelog received editorial corrections to prevent mixed-language translations.
  • Currency and offline-pack indexes were regenerated, and the same build 112 bundle was propagated to web, Android, and iOS.
  • External review remains open: this version does not imply independent clinical validation or automatic store availability.

v1.73.2 — Reading refinements and coordinated native release

  • Spacing and text refinements in Home, Study, Clinical and the atlas.
  • Optional iOS dictation with microphone and speech recognition permissions. Apple may process audio; offline recognition is not guaranteed.
  • Release candidate for testing; this does not imply store availability or independent clinical validation.

v1.73.1 — Navigation, study and a bilingual interface

  • Smoother navigation, controls adapted to small screens and improved readability in Home, Study and Clinical.
  • Saved items retain deletions during sync; search opens biomarkers and comparisons require explicit units.
  • Spanish and English interface and widgets; all six web library collections in both languages, with English app screenshots.
  • Educational content remains subject to review: translation coverage does not mean clinical validation, and some passages remain in Spanish.
  • Updated Android files. Google Play availability depends on the store release process.

v1.71.0 — A connected Medical OS, with no subject left thin

  • A connected, local Medical OS: an initial clinical ontology of 120 concepts links knowledge, learning, clinical work, the atlas, evidence and research without duplicating the corpus. Relationships are labelled as editorial drafts, never as causality or clinical advice.
  • Reproducible Dataset Lab: it opens CSV files locally, detects types and missingness, builds a dictionary, computes a SHA-256 fingerprint and generates charts without uploading raw rows. Exports retain provenance and decisions.
  • Evidence Watch and guideline comparison: records what changed, which fields moved, source currency and review status. The queue works in software, but no change ships without editorial review and no real-time literature feed is simulated.
  • Effect measures with explicit limits: the lab keeps RR, OR and HR separate and will not turn a hazard ratio into NNT without a compatible time-specific baseline risk. A p value is not presented as effect size or clinical importance.
  • +193 questions, taking the bank to 5,103. They are not filler sprinkled around: it cleared the backlog of the 18 subjects that sat below 20 questions, and none is under 25 today. Revising a whole subject with twelve questions is not revising, and that gap hit the smaller subjects hardest.
  • All of them come through the full pipeline: per-option reasoning — why each one fails, not just which is right —, declared difficulty, a dated source, a learning objective and a risk level.
  • 100% of questions now carry an explanation. Fifty had one written and were being served mute: twenty behind a broken <script>, twenty because their explanation used a format the extractor did not recognise, and ten because the author had only written per-option reasoning and no general explanation. Not one was rewritten: they were there and were not reaching the screen.

v1.70.0 — Epistemis connects learning, clinical work and evidence

  • Personal Radar on Home: it picks one next action from overdue reviews, weak areas, mistakes, coverage and clinical practice. The recommendation, its signals and confidence are computed locally; Clio only receives context when you tap “Consult”.
  • Cross-cutting medical graph: searching for a topic no longer ends on an isolated card. Epistemis connects guides, questions, cases, atlas, drugs, calculators, Clio and sources to understand, visualise, practise, apply, remember and verify evidence. A connection means semantic relation, not causality or a clinical recommendation.
  • Evidence and Research workspace with a source registry, local PDF analyser, PICO projects, a reproducible PubMed query, RR/OR/HR/CI/ARR/NNT interpretation and a diagnostic lab for sensitivity, specificity, predictive values and likelihood ratios.
  • Clinical Clio adds structured case intake and flows for differential diagnosis, management-changing tests, medication safety and triage. Answers separate data, inference and uncertainty, while an audit catches non-existent citations or unsupported figures and regimens.
  • Up to three images inside chat, with orientation, compression, quality checks and privacy warnings. Images are not stored or synced; every response offers a traceability card with mode, requested engine, personalisation, sources and estimated citation coverage.
  • 417 questions (8% of the bank) now carry the X-ray, ECG, slide or lesion the stem talks about, taken from the app's own atlas. No new material was needed: the 6,555 plates were already licensed and catalogued — what was missing was the bridge. The image is added only when the diagnosis appears literally in the stem, never by matching against the options, so it cannot give the answer away. And every one carries a caption: it is an atlas example, not that patient's own study.
  • Spaced review now tunes itself to your pace: the optimiser recomputes the 17 FSRS weights from your own history and only changes them if they beat the defaults by at least 2%. The whole computation happens on your device, and you can always reset to the originals.
  • Questions and guide corrections now update without waiting for a new release. Guides are patched by replacing text, never by injecting server-side code: the promise that no outside code runs holds by design, not by trust.
  • Percentiles against the cohort: contribute your mock-exam score — the score alone, nothing that identifies you — and see where you land. It requires an account, and a minimum number of attempts before the comparison shows up.

v1.68.2 — Desktop caught up, and dictation in Clio

  • You can dictate to Clio instead of typing. Useful exactly when you actually ask things: hands busy, or on your way somewhere.
  • The desktop app had drifted behind while the app grew: the sidebar showed 7 of the 15 atlases and the command palette waited for the whole clinical corpus to index before it would open "Settings". Now the sidebar comes from the same navigation as mobile, and the palette answers instantly.
  • In a narrow window the desktop app no longer forces the sidebar: under 900 px it uses the same compact interface as mobile, so a small window is usable again.
  • Auscultation sounds and video are served in ranges: you can seek within a track without waiting for the whole thing to load.

v1.68.1 — Epistemis on the desktop, and Clio included

  • A desktop app for Windows and macOS: the same app, with its content inside the package and no browser in the way.
  • Clio no longer needs you to bring an API key: sign in and it is included, with a monthly allowance. If you prefer your own key, it stays unlimited.
  • A public website with the open library — symptoms, formulary, calculators, case atlas and glossary — readable without installing anything.

v1.68.0 — Security hardening and sync coverage

  • Guides no longer load any external resource: all content is served from the device itself.
  • The mistakes notebook and the emergency and clinic records now sync with your account.
  • Public website released and the iOS build prepared.

v1.66.0 — A new icon set across the app

  • Generic emoji are gone: line icons consistent with the brand, including organs and specialties.
  • Fixed icons that rendered blank or showed their raw name.

v1.65.0 — Restructured navigation

  • New bar: Home · Study · Clinic · Atlas · More. The Atlas becomes a tab of its own.
  • Fewer hops to reach what you use daily.

v1.63.0 — Clio, the AI tutor, arrives

  • The AI tutor is now called Clio and answers from the app’s guides, citing which one each claim comes from.
  • A temporary chat that is stored nowhere.
  • Included when you sign in, or with your own API key with no cap.

v1.62.0 — Google sign-in modernised

  • Migrated to Credential Manager: Google sign-in is faster and no longer fails on recent Android.

v1.47.0 — Triage, renal dosing and the Founder programme

  • Triage tools and dose adjustment in renal impairment.
  • Founder programme for those arriving before subscriptions switch on.
  • Startup and search performance improvements.

v1.40.0 — Exam analytics and audio guides

  • Mock-exam analytics that reinforce what you get wrong most.
  • Audio-only mode for listening to the guides.
  • Goal-based home, learning paths and the "My day" flow.
  • Search turned into a command palette with grouped results.

Content updates without reinstalling

Fixing a guide or adding questions is a content update, not a new app version: it arrives on its own and does not show up in this list. If you find an error, send the reference to soporte@somatryx.com.

Get up to date

If you installed it a while ago, update from the store to get everything above.

Download

Free. No account required. Offline core; the Atlas downloads by module.