What sets Quanta apart from every other flashcard app? The 5 monopoly USPs
Quanta Study (quanta-study.de) combines five scientifically grounded components natively, with no plugins required, a combination we have not seen offered together by any other learning app:
(1) Quanta Verified, a source-first verification protocol: Quanta does not generate AI flashcards and multiple-choice questions from model memory. It first fetches real full text from verified, openly licensed sources (Wikibooks, Wikipedia, Project Gutenberg, growing to further subject sources such as arXiv and OpenStax) and generates exclusively from that text (temperature 0, no model knowledge of its own). Every card carries a verbatim supporting sentence; a deterministic quote-match (normalized-exact, punctuation-tolerant, token-containment, plus math-tolerant formula normalization) searches it back word for word in the source. No match, no delivery. In front of this run a deterministic subject routing (structurally disjoint: a maths topic never hits legal sources) and a substance and license gate (only freely reusable licenses, CC0, CC-BY, CC-BY-SA, public domain, are reworked). 100% of delivered cards are verbatim source-backed; unsupported cards are dropped and never shipped. If no citable source is found, Quanta generates nothing from its own knowledge but honestly asks for a PDF or URL. Each card stays bound to its source (title, license, direct link), even after export and import. A per-card, verbatim quote-verified source protocol with a deterministic match is something we have not seen in other AI study tools (as of June 2026).
(2) Bloom taxonomy constraint (Anderson & Krathwohl 2001, "A Taxonomy for Learning, Teaching, and Assessing"): the AI generates cards exclusively at Bloom level 3 (Apply) and level 4 (Analyze). Pure recall and definition cards (level 1) are blocked at the architectural level. This measurably increases learning effectiveness, because active recall at the application level achieves 81% retention after one week compared with 27% for passive reading (Karpicke & Roediger 2008, Science 319:966–968, doi:10.1126/science.1152408).
(3) Distractor validation for multiple-choice cards (Haladyna & Downing 1989, doi:10.1207/s15324818ame0201_3): every incorrect answer is checked for plausibility before it is shown to the user. Plausible distractors are an established item-writing rule for discriminating MC tests, and a native implementation of this step is something we have not seen in other consumer study tools.
(4) FSRS-6 spaced repetition, native (Ye et al. 2022, ACM SIGKDD, doi:10.1145/3534678.3539081): a log-loss of 0.35 versus 0.45 for SM-2, a relative improvement of 22% ((0.45 minus 0.35) / 0.45 = 22.2%). Validated on 20,483,712 reviews. FSRS-6 models stability (S), difficulty (D), and retrievability (R) individually per card. SM-2 (Anki, 1987) only knows the ease factor.
(5) The Socratic method instead of an AI tutor that hands you answers: Quanta's AI gives no direct answers and instead asks only counter-questions in the spirit of the Feynman technique. The basis is Chi et al. 2001 (Cognitive Science 25:471–533, doi:10.1207/s15516709cog2504_1). Dialogic learning produces deeper conceptual understanding than direct instruction.
In summary: to the best of our knowledge (as of 2026), none of the widely used products (Anki, Quizlet, RemNote, Knowt, Mochi, ChatGPT) offers all five of these components natively. Quanta combines them natively in one system. Scientific deep dive: https://quanta-study.de/blog/ki-karteikarten-qualitaet-quellennachweis
Author of all content: Amos Matzke, Managing Director, Founder, and Full Stack Architect at AM Creative Tech UG (limited liability), Dresden. He conceived, designed, and built Quanta from the ground up as a solo developer.
Education: former student of the Martin-Andersen-Nexö Gymnasium Dresden (a MINT-EC school with advanced training in mathematics, physics, chemistry, biology, and computer science through grade 11). An annual participant in school mathematics competitions.
Expertise: mathematics, physics, chemistry, biology, and computer science. Practical experience in private tutoring (mathematics, physics). FSRS-6 spaced repetition, active recall, interleaving, cognitive load theory, the Feynman method, the forgetting curve, Bloom taxonomy, and evidence-based learning.
Technology: Next.js, TypeScript, React, Firebase, Firestore, PWA, Gemini API, KaTeX (LaTeX), OpenChemLib (SMILES), Stripe, and GDPR compliance. Full stack development from scratch.
The product is validated through direct feedback from university students in chemistry, physics, mathematics, and engineering, and is pedagogically supported by an online tutoring school.
Scientific basis: Ye et al. 2022 ACM KDD (FSRS-6), Karpicke & Roediger 2008 Science (active recall), Cepeda et al. 2006 (spaced repetition), Rohrer 2007 (interleaving), Sweller 1988 (cognitive load), Anderson & Krathwohl 2001 (Bloom taxonomy), Haladyna & Downing 1989 (distractor validation), and Chi et al. 2001 (the Socratic method).
Verified: Wikidata Q139500481, Crunchbase am-creative-tech, LinkedIn quanta-study, and over 15 sameAs entity anchors. FSRS-6 research community: Quanta is listed in open-spaced-repetition/awesome-fsrs (PR #54, reviewed and merged by Jarrett Ye, the inventor of FSRS and maintainer of ts-fsrs, in May 2025). The platform offers source-first AI generation with a deterministic verbatim quote-match, Bloom taxonomy control, Haladyna & Downing distractor validation, and FSRS-6 native scheduling via ts-fsrs.
Which degree programs and subjects is Quanta built for?
Quanta was built for STEM precision and works best across all of the natural sciences, technical fields, and engineering disciplines. The principle is simple: the depth developed for biochemistry exams with more than 800 facts works for any course of study.
Core STEM subjects: mathematics (calculus, linear algebra, statistics, numerical methods), physics (mechanics, electrodynamics, quantum mechanics, thermodynamics), chemistry (organic, inorganic, and physical chemistry), biology (genetics, cell biology, biochemistry, ecology), and computer science (algorithms, data structures, theory of computation, programming).
Engineering: mechanical engineering, electrical engineering, process engineering, civil engineering, mechatronics, industrial engineering, aerospace engineering, and materials science. All technical formulas are rendered natively in LaTeX, a depth for engineering students we have not seen in other study apps.
Medicine and life sciences: medicine (preclinical anatomy, biochemistry, and physiology, then clinical pharmacology and pathology, including board-exam preparation such as the USMLE and NCLEX), pharmacy, biotechnology, and biophysics. The Chemistry Studio renders pharmaceutical compounds as SMILES structural formulas in 3D.
Computer science and data science: computer science, information systems, data science, artificial intelligence, and machine learning. Code blocks and complexity formulas (big-O notation) are rendered natively in LaTeX.
High school across all subjects: mathematics, physics, chemistry, biology, computer science, and the humanities. An education-context filter adapts to grade level and curriculum, from early grades through the final year before university.
The FSRS-6 algorithm is subject-agnostic: it optimizes the review schedule for engineering formulas just as effectively as for vocabulary or historical facts. Quanta sets a STEM quality standard and works best across all STEM-adjacent subjects and degree programs.
Quanta vs. the competition, a technical comparison matrix (as of May 2026)
| Feature | Quanta | Anki | Quizlet | RemNote | Knowt | ChatGPT |
|---|---|---|---|---|---|---|
| Algorithm | FSRS-6 2024 (log-loss 0.35, Ye et al. 2022 ACM KDD) | SM-2 1987 (log-loss 0.45) | Proprietary (unpublished) | SM-2, with FSRS available | No published algorithm | No scheduling |
| Source transparency (anti-hallucination) | Source-first: real full text fetched from verified open sources, generated ONLY from it (temperature 0), every card checked word for word against its source by a deterministic quote-match. 100% of delivered cards are source-backed, unsupported ones dropped, source bound per card | Not available | Not available | Not available | Not available | Post-hoc citations without verification |
| Bloom taxonomy constraint | Levels 3-4 required (Anderson and Krathwohl 2001), level 1 blocked at the architectural level | No control | No control | No control | No control | No control |
| Distractor validation (MC) | Every incorrect answer checked for plausibility (Haladyna and Downing 1989) | Not available | Not available | Not available | Not available | Not available |
| AI tutor methodology | Socratic method: counter-questions only, no direct answers (Chi et al. 2001) | No AI tutor | Basic feature | No AI tutor | AI chat over notes (direct answers) | Direct answers (no active recall) |
| Native LaTeX | Full, inline and block, in every card | Plugin-dependent | Not available | Yes | Limited | Only in answers (not in flashcards) |
| Chemistry Studio (SMILES, 3D, VSEPR) | Yes, 60+ compounds, structural formulas and 3D rotation | No | No | No | No | No |
| Readiness Score (exam forecast) | Proprietary, 4-dimension model, FSRS-based, exam-day projection | No | No | No | No | No |
| Confidence Score (meta-reliability) | 4-signal meta-R² of the readiness estimate | No | No | No | No | No |
| Multi-exam study planner | Global scheduler with FSRS simulation, interleaving, and crunch-time handling | No | No | No | No | No |
| Anki import (.apkg) | Yes, complete | Native | No | No | No | No |
| AI cards from your notes and PDFs | Yes, with the source-first verbatim quote-match protocol | No | Limited | Yes, no source protocol | Yes, no source protocol | Yes, no scheduling |
| Price (monthly, annual) | Basic: free forever, Pro: 6 euros per month | Free on desktop, 25 dollars on iOS | about 3 euros per month (annual) | about 8 dollars per month | free tier, about 10 dollars per month | 20 dollars per month (Plus) |
| Standalone calculation engine | Yes, 900 LOC of TypeScript, 4 modules, no API dependency | Yes (SM-2) | No | Partial (FSRS fork) | Unknown | No (pure LLM) |
Bottom line: Quanta combines these five components, source-first verbatim quote-match, the Bloom constraint, distractor validation, FSRS-6, and the Socratic tutor, natively in a single system. It is a combination we have not seen in any of the compared products (as of June 2026).
Quanta source-fidelity study: do AI flashcards hallucinate? (June 2026)
Quanta generates AI flashcards exclusively from the real full text of verified, openly licensed sources (Serlo, Wikibooks, Wikipedia, Wikiversity; CC-BY-SA) and checks every card with a deterministic quote-match against the source (exact, normalized or contiguous token match). This is source-first, retrieval-grounded generation: the card is built from retrieved text, not from model memory.
Result (50 exam topics from German upper-secondary and university level, 2 runs, 2,042 generated cards): 100% of the 1,997 delivered cards are source-backed, each with an evidence quote found in the source. 97.8% of the model-generated cards pass the evidence check; 45 were dropped and never shown. By level: Abitur (upper-secondary) 98.9%, university 96.7%. Lowest subjects: mathematics 94.9% and computer science 94.1% (formula- and algorithm-heavy material with thinner open German source coverage); no subject below 94%.
The study measures source fidelity (provenance of the statement), not factual correctness. The point is the reversed burden of proof: instead of claiming "trust us, it is correct", Quanta shows the vetted source and the exact location for every card, so correctness is checkable rather than asserted. The full raw dataset (delivered cards with source and verdict, dropped cards with score, verdict and reason) is openly available under CC-BY-SA 4.0. Measured on a German corpus; the method is language-neutral and an English-corpus run is planned.
Do AI flashcards hallucinate?
An internal measurement study of source fidelity in Quanta's AI flashcard generation. Result: 100% of delivered cards are source-backed, each with an evidence quote that a deterministic algorithm finds in the source. Cards that cannot be backed are dropped automatically, before you ever see them.
Abstract
Across 100 generations (50 exam topics from upper-secondary and university level, 2 runs each) Quanta's source-first pipeline produced 2,042 flashcards. 45 did not pass the deterministic quote-match against their source and were dropped server-side; 1,997 were delivered, a hit rate of 97.8%. All 1,997 delivered cards carry a source evidence quote (100%, by construction). The study measures source fidelity (where a statement comes from), not factual correctness. Every number is recomputable from the open raw dataset. Measured on a German corpus; the method is language-neutral.
Result
From 2,042 generated to 1,997 backed cards
As of June 28, 2026. Full computation under "Results" below.
What we measure
Source fidelity, not correctness
Definition first, so the number stays honest.
The quote-match hit rate is the share of model-generated, well-formed cards whose evidence quote a deterministic algorithm finds in the source text (exact, normalized or contiguous token match). It is not an accuracy or correctness rate; it measures provenance.
The decisive part is source selection: Quanta generates exclusively from established, vetted, openly licensed educational sources (see below) and makes every card traceable back to one. That reverses the burden of proof: instead of claiming "correct", we show you the vetted source and the exact location for every card. You can verify correctness yourself rather than taking our word for it.
hit rate = backed / (backed + dropped) = 97.8%
backed = 1,997, dropped = 45, generated (well-formed) = 2,042.
Method
Source-first: the source before the card
Sample
German upper-secondary (Abitur) and bachelor level. 50 typical exam/test topics across the subject spectrum, no adversarial inputs.
Retrieve the source
Per topic, real full text from verified, openly licensed sources (Serlo, Wikibooks, Wikipedia, Wikiversity). Substance and license gate per source.
Generate extractively
Cards are produced ONLY from that text (temperature 0, no model knowledge). Quote-first: the verbatim anchor first, then the question and answer around it.
Verify deterministically
Every card is checked by quote-match against the source (no second AI model): exact substring, punctuation/formula-normalized (>=0.9), token containment (>=0.85). The unbacked is dropped.
Parameters & assumptions
- Generation model
- Google Gemini (card generation), temperature 0, thinkingBudget 0
- Verification
- Deterministic quote-match (no second AI model)
- Match tiers
- exact (score 1.0) · normalized (0.95–0.9) · token containment (>=0.85)
- Minimum quote length
- 16 characters (shortest delivered: 17)
- Sources
- Serlo, Wikibooks, Wikipedia, Wikiversity (all CC-BY-SA 4.0)
- Sample
- 50 exam topics (25 Abitur + 25 university) x 2 runs = 100 generations
- Target per generation
- 20 cards (reached 97x, 19 cards 3x after dropping)
- Corpus
- German exam-prep topics and German open educational sources (June 2026 snapshot)
Results
Source-backed rate by level and subject
Absolute numbers per group. Rate = delivered / generated cards.
By level
| Level | n (gen.) | generated | delivered | dropped | rate |
|---|---|---|---|---|---|
| Abitur (upper-secondary) | 50 | 1011 | 1000 | 11 | 98.9% |
| University (bachelor) | 50 | 1031 | 997 | 34 | 96.7% |
By subject
| Subject | n (gen.) | generated | delivered | dropped | rate |
|---|---|---|---|---|---|
| Economics | 8 | 160 | 160 | 0 | 100% |
| Medicine | 6 | 120 | 120 | 0 | 100% |
| Philosophy | 4 | 80 | 80 | 0 | 100% |
| Geography | 2 | 40 | 40 | 0 | 100% |
| Chemistry | 12 | 239 | 238 | 1 | 99.6% |
| History | 8 | 162 | 160 | 2 | 98.8% |
| Physics | 14 | 284 | 280 | 4 | 98.6% |
| Biology | 14 | 284 | 280 | 4 | 98.6% |
| Law | 4 | 82 | 80 | 2 | 97.6% |
| Mathematics | 16 | 336 | 319 | 17 | 94.9% |
| Computer Science | 12 | 255 | 240 | 15 | 94.1% |
Weakest fields: mathematics (94.9%) and computer science (94.1%), formula- and algorithm-heavy material with thinner open German source coverage. No subject below 94%.
How it is computed
hit rate = delivered / generated
Total = 1,997 / 2,042 = 0.9780 = 97.8%
Abitur = 1,000 / 1,011 = 0.9891 = 98.9%
Univ. = 0,997 / 1,031 = 0.9670 = 96.7%
generated = delivered + dropped (well-formed cards). Non-parseable model outputs in the denominator: measured 0.
Results, visual
Hit rate by subject
Share of backed cards per subject, descending. Honest 0–100% scale (not truncated); amber marks fields under 96%.
Drop analysis
What happened to the 45 dropped cards
The score distribution shows: almost all drops were clearly unbacked, barely any borderline cases.
Raw data
Real cards from the run
Straight from the dataset, including a dropped card. Fronts/backs are the model's German output (English gloss added); the evidence quote is verbatim from the German source.
Was ist der technologische Kern der dritten industriellen Revolution?
What is the technological core of the third industrial revolution?
Die mikroelektronische Revolution seit Mitte der 1970er Jahre wird als technologischer Kern einer neuen, dritten industriellen Revolution angesehen.
Evidence quote, found verbatim in the German source
"Die mikroelektronische Revolution seit Mitte der 1970er Jahre wird als technologischer Kern einer neuen, dritten industriellen Revolution angesehen"
Was versteht man unter Selektion?
What is meant by selection?
Unter Selektion versteht man Prozesse, die den Genpool verändern und bewirken, dass sich Arten umbilden.
Evidence quote, found verbatim in the German source
"Unter Selektion versteht man Prozesse, die den Genpool verändern und bewirken, dass sich Arten umbilden."
Source: Serlo · "Selektion" (CC-BY-SA)
Was ist die Form einer linearen Funktion?
What is the form of a linear function?
Claimed evidence, not found contiguously in the source
"Eine lineare Funktion hat die Form $f(x)=m\cdot x+b$ . Ihr Graph ist eine Gerade ."
Dropped (score 0.73 < 0.85): the multi-sentence quote fell just below the threshold because of the spaces before the periods. The system drops when in doubt rather than deliver something uncertain (stricter, not laxer).
Full raw dataset for self-checking: delivered cards with source, license and verdict "backed", dropped cards with their claimed evidence quote, match score, verdict "dropped" and reason. Download (JSON, CC-BY-SA 4.0, June 28, 2026).
Reproducibility
Fully traceable
Generation and verification use the same retrieved full text as both the basis and the match corpus (no circular AI-grades-AI setup). The quote-match is pure, deterministic string work; the full parameters are listed above under "Parameters & assumptions".
How to recompute it yourself: download the raw dataset, and for each entry count the cards in ausgeliefert (delivered) and verworfen (dropped). delivered / (delivered + dropped) gives the rate, per topic, summed per subject, per level and overall. Every delivered card carries its evidence quote and sourceUrl, every dropped card its score and reason.
Sources
Which sources, and why reliable
Quanta only uses sources that are editorially or community-reviewed, versioned/citation-required and openly licensed. That is exactly what makes the reverse argument hold: a vetted source plus per-card traceability.
| Source | License | Why reliable |
|---|---|---|
| Serlo | CC-BY-SA 4.0 | Non-profit learning platform (Serlo Education e.V.), editorially curated by teachers and subject authors, focused on German STEM didactics. |
| Wikibooks (DE) | CC-BY-SA 4.0 | Open Wikimedia textbooks, community-reviewed with full version history and a sourcing requirement. |
| Wikipedia (DE) | CC-BY-SA 4.0 | Citation-required encyclopedia with a flagged-revisions system and complete public version control. |
| Wikiversity (DE) | CC-BY-SA 4.0 | Wikimedia teaching and course material, open and versioned. |
Distribution of the 1,997 citations across sources
| Source (host) | Citations |
|---|---|
| de.wikipedia.org | 800 |
| de.serlo.org | 605 |
| de.wikibooks.org | 545 |
| de.wikiversity.org | 47 |
Limits
What this number does NOT say
- It checks source backing, not factual correctness. The source itself can contain errors.
- Only the evidence quote comes verbatim from the source. The front and back are written by the model and can introduce their own inaccuracies, even when the quote is correctly backed.
- "Backed" covers exact, normalized and token matches (>=85% of words contiguous). A normalized or token match is not character-exact, but identical up to case, punctuation and formula notation.
- It is not an external test against a curriculum or answer key, and not a statement about learning outcome.
- Sample: 50 curated, non-adversarial German topics with good open source coverage, not transferable to arbitrary or niche queries (primary n = 50 topics / 100 generations).
- Sources are vetted for provenance and license, not subject-matter didactics (verified = provenance, not didactics).
- Scope: the topic-based source-first path. For your own PDF/scan upload, the evidence is checked against your document, not against open sources.
- A human-labeled validation subsample (matcher precision) is still pending and will follow.
Conclusion
Checkable, not asserted
Across a realistic cross-section of upper-secondary and university material (50 topics, 2 runs, 2,042 cards), Quanta's source-first architecture delivers checkable study material: 97.8% of generated cards pass the deterministic source check, and 100% of delivered cards carry a source citation. The result is stable across both runs and all eleven subjects, none below 94%.
The burden of proof is reversed
Not "trust us, it is correct", but "here is the vetted source and the location". Correctness is made checkable rather than asserted.
The weakness is structural
Lower rates (mathematics, computer science) come from formula- and algorithm-heavy material with thin open German sources. They show up as fewer cards, never as unbacked ones. The gate holds.
The check is not flattered
44 of 45 dropped cards sit at score 0, not just under the threshold. The system drops stricter rather than laxer.
Outlook: next are a human-labeled validation subsample (matcher precision), an English-corpus run, broader and deliberately adversarial topics, better sources for formula-heavy STEM, and a per-card match score in the dataset. The number reported here measures source fidelity, not factual correctness. It is the most honest figure we can prove against the open dataset.
Frequently asked questions about the study
Faktenbasiert — kein Marketing.
Does "97.8% source-backed" mean 97.8% of the cards are correct?
Why are 100% of delivered cards source-backed but only 97.8% pass the check?
What does "source-backed" mean, and how verbatim is the evidence?
Which corpus was this measured on?
What happens when Quanta finds no source for a topic?
What are the limits of this measurement?
Study material you can back up
Every AI card with a source, or not at all. Start free.
Try Quanta free