# Tarsikh (ترسيخ) — Complete Project & Pedagogical Documentation

> Tarsikh is an educational, offline-first mobile application and open scientific project dedicated to Quran memorization (*Hifdh*). It combines modern cognitive neuroscience with classical Quranic pedagogy to ensure solid, long-term retention.
> Website: https://tarsikh.com
> Contact: contact@tarsikh.com
> Available on: Android (Google Play)

---

## 1. The Core Problem: Overcoming the "Illusion of Competence"

When reviewing the Quran through passive reading, students frequently experience the **illusion of competence**: because the page looks familiar, the brain creates a false sense of mastery. In reality, passive visual recognition uses different neural pathways than active retrieval from memory.

**Tarsikh solves this by:**
1. **Active Recall:** Masking the Quranic text during recitation to force the brain to retrieve verses from memory.
2. **Spaced Repetition (SRS):** Scheduling reviews at mathematically optimal intervals—just before a verse begins to fade from memory.
3. **On-Device Speech Guidance:** An AI recitation assistant that progressively reveals words as they are pronounced.
4. **Total Privacy & Offline Autonomy:** Operating 100% offline with zero cloud audio processing and no account creation.

---

## 2. Key Pedagogical Modules & Features

### 2.1 The "4 Rs" Review Cycle (Memorizator)
The primary daily revision mode follows a structured cognitive loop:
1. **Recite (تلاوة غيباً):** The text remains concealed. Only the opening verse is shown as a cue. The reciter recites from memory, optionally assisted by offline voice recognition.
2. **Reclassify (تقييم الأخطاء):** The full text is revealed. The reciter taps any words where hesitations or mistakes occurred. An objective star rating (1★ to 5★) is automatically computed based on the portion size.
3. **Review (استماع وتصحيح):** Full text is displayed alongside audio recitation by renowned reciters. Both current mistakes and past problem words are visually highlighted to consolidate transitions.
4. **Restart (تأكيد):** The app records the performance, updates spacing intervals, and queues the next portion.

### 2.2 Adaptive Spaced Repetition (Adapted SM-2)
Unlike standard vocabulary flashcard algorithms (such as Anki), Tarsikh adapts the SuperMemo SM-2 algorithm specifically for continuous, long scripture:
- **Volume Error Scaling:** Error tolerance scales non-linearly with portion size (e.g., a full Hizb of ~10 pages vs. a single page) using square-root dampening so minor slips on large blocks do not unfairly reset progress.
- **Cognitive Load Protection:** When a student accumulates large active review volumes, review intervals dynamically stretch to prevent overwhelming, demotivating daily backlogs.
- **Rating Drop Cushioning:** If a reciter scores poorly on a large portion, previously well-consolidated verses (4★ or 5★) only drop by a single level rather than resetting completely.

### 2.3 Targeted Flashcards
Independent training drills that do not interfere with the main revision calendar:
- **Next Verse Drill:** Prompts the user with a verse and asks for the following one. For similar verses (*Mutashabihat*), the app provides starting context and highlights other matching locations across the Quran.
- **Cloze Tests (Fill in the blanks):** Tests missing words with three levels of difficulty (Easy, Medium, Hard).
- **Error Categorization:** Users can categorize slips into 4 types for precise self-diagnosis:
  - *Lapse:* Forgetting a word or transition.
  - *Substitution:* Confusing words with similar passages (*Mutashabihat*).
  - *Harakat:* Short vowel or grammatical declension slip.
  - *Block:* Total memory blank.

### 2.4 Free Reading (Immersive Mushaf)
- **15-Line Traditional Medina Layout:** Exact facsimile of the King Fahd Complex Mushaf (1441H) that auto-fits any phone screen with zero vertical scrolling.
- **Fluid Scrollable Mode:** Continuous vertical reading with customizable font sizes (18px to 36px) and hands-free automatic scrolling with configurable speeds (1 to 10).
- **Restricted Active Mode:** Allows filtering the Quran to only display the student's active memorized pages, with previously failed verses highlighted in translucent red for focused practice.
- **Lexical & Voice Search:** Instant offline search by typing words (TF-IDF index) or by reciting any verse directly into the microphone.

### 2.5 Offline Student Tracker (For Teachers)
Designed specifically for Quran study circles (*Halaqat*):
- Works completely offline without requiring internet, student registrations, or passwords.
- Tracks individual student milestones (Surah, verse reached, date, qualitative performance).
- Supports multi-round revisions (*Rondes de révision*).
- Displays exact memorization completion (percentage of the 604 pages).

---

## 3. Audio & Recitations

Tarsikh supports multi-reciter audio playback with offline pre-downloading:
- **Hafs Reciters:** Saad Al-Ghamdi, Abdul Rahman Al-Sudais, Mohamed Siddiq Al-Minshawi, Mahmoud Khalil Al-Husary, Mishary Alafasy, Maher Al-Muaiqly.
- **Warsh Reciters:** Ibrahim Al-Dosary, Yassin Al-Jazaery.
- **Loop & Repetition Modes:** Verses can be looped indefinitely ($\infty$) or repeated 1x, 3x, or 5x for audio-based memorization.
- **Screen Wake Lock:** Prevents screen timeout during active recitation and audio listening.

---

## 4. Internationalization & Riwayat

### Supported Languages (8)
The interface is fully localized in:
- **Arabic (`ar`)** (Native RTL)
- **Urdu (`ur`)** (RTL)
- **English (`en`)**
- **French (`fr`)**
- **Indonesian (`id`)**
- **Turkish (`tr`)**
- **Bengali (`bn`)**
- **Russian (`ru`)**

### Supported Readings & Scripts
- **Hafs 'an 'Asim:** Classic Uthmanic Medina script (QCF V4 & DigitalKhatt typography).
- **Warsh 'an Nafi':** Authentic North & West African Uthmanic Moroccan script.
- **Phonetic Latin Transliteration:** Complete Latin transliteration aligned word-for-word for non-Arabic speakers learning phonetics.

---

## 5. Privacy, Offline Architecture & Security

Tarsikh was architected from day one with a strict **Privacy-First & Offline-First** philosophy:

1. **Zero Cloud Audio:** The microphone audio stream is processed strictly in temporary device RAM by a local AI engine and discarded immediately. No voice recording is ever saved to the disk or transmitted over the internet.
2. **Zero Commercial Trackers:** No advertising, no third-party telemetry, no Google Analytics, and no behavioral SDKs.
3. **No Mandatory Account:** Users never provide an email, name, or phone number to use the app.
4. **Local Data Isolation:** All learning history, student progress, and personal notes remain exclusively on the user's phone in a local sandbox. Uninstalling the app permanently erases all stored data.

---

## 6. Scientific Research & Voluntary Contribution

Tarsikh hosts an open scientific research initiative aimed at modeling human forgetting curves and memory retention mechanisms on long structured texts:
- **Strictly Voluntary & Manual:** No background data synchronization exists. Users must actively click "Share review data" in the settings.
- **Anonymous UUID:** Each device generates an anonymous random identifier (RFC4122 v4 UUID, e.g., `usr-...`) on install. This ID is never linked to personal identity or hardware fingerprints (no IMEI or MAC address).
- **Web Demographic Survey:** Users can optionally visit [tarsikh.com/study](https://tarsikh.com/study) and paste their anonymous UUID to provide demographic metadata (age range, native language, memorization background, sleep habits).
- **Right to Erasure (GDPR):** Synced research records can be permanently deleted upon request by emailing `contact@tarsikh.com` with the UUID.

---

## 7. Published Technical Papers

- **Title:** *Quranic Speech Recognition: Studio vs. Phone Microphone & Phonetic Error Modeling for Tarsikh*
- **Publication Date:** 2026-06-16
- **Model:** Zipformer 65M INT8
- **DOI:** [10.5281/zenodo.21770922](https://doi.org/10.5281/zenodo.21770922)
- **PDF Download:** https://tarsikh.com/assets/research/asr-report-2026.pdf
- **BibTeX:**
```bibtex
@techreport{tarsikh2026asr,
  title={Quranic Speech Recognition: Studio vs. Phone Microphone & Phonetic Error Modeling for Tarsikh},
  author={Tarsikh Project},
  year={2026},
  institution={Tarsikh},
  doi={10.5281/zenodo.21770922},
  url={https://doi.org/10.5281/zenodo.21770922}
}
```

---

## 8. Frequently Asked Questions (FAQ)

### What makes Tarsikh different from typical Quran apps?
Most Quran applications are digital readers or audio players. Tarsikh is an active learning and retention system based on cognitive science: it hides verses to force active memory retrieval and spaces reviews mathematically so memorized portions are never forgotten.

### Why doesn't voice recognition penalize my mistakes automatically?
Tarsikh's speech recognition serves as a **reading assistant**, revealing words as you pronounce them to keep your natural rhythm. Because it runs 100% offline on mobile processors, automated error judgment could cause false penalties on pronunciation nuances. The student manually and honestly tags their mistakes during the Reclassify step.

### Does the app evaluate Tajweed rules (such as Ghunnah or Madd)?
No. Tarsikh focuses strictly on textual memorization (forgotten words, inverted verses, word substitutions). This corresponds to the primary challenge of long-term scripture retention.

### Is Tarsikh free or are there in-app purchases?
The application is 100% free, non-profit, and open, with zero advertisements and no paid subscriptions.

### Can I review without any internet connection?
Yes. The complete Quranic text, multiple font families, offline search index, and AI speech recognition engine are fully embedded in the app package. You can review in remote locations, on airplanes, or in areas with no cellular coverage.
```