Frequently Asked Questions (FAQ)

Tarsikh (which means "anchoring" or "consolidation" in Arabic) is a mobile learning application designed to help Quran memorizers and students (Hafidh) preserve and consolidate their memorization over the long term. It replaces passive reading with an active, personalized review system based on cognitive science.

The illusion of competence occurs when a student rereads a page they think they know: because the text is familiar, their brain gives them the misleading impression that they have memorized it. In reality, recognizing a text during reading does not activate the same neural pathways as the effort to retrieve it from memory (active recall).

Tarsikh seeks to break this illusion by hiding the text during the initial recitation phase. By forcing you to retrieve the text from your memory, the application stimulates and strengthens the neural connections necessary for long-term retention.

Tarsikh uses a spaced repetition algorithm derived from the SuperMemo SM-2 method. However, standard algorithms (like Anki) are calibrated for small individual learning cards (like vocabulary words). Tarsikh adapts this concept for continuous, long texts (portions of the Quran) by incorporating specific mathematical adjustments:

1. Error scale dampening proportional to volume: The error thresholds tolerated to obtain a score (from 1 to 5 stars) are configurable in the settings. They are defined for a reference page (about 12 verses). If you review a larger portion (for example a Hizb of 10 pages), the application applies a non-linear scale factor based on the square root of the size of the portion. This prevents assigning a bad grade to a large block for only a few minor mistakes.
2. Cognitive load scaling factor: When the student accumulates a large number of active verses (about 1.5 Hizb), the application applies a logarithmic stretch calculation to the review interval. This progressive stretching of reviews prevents the accumulation of an excessively heavy and discouraging daily workload.

No, the application integrates a rating drop dampening mechanism to protect your history:
In the database, each verse has its own memorization history (number of stars, SRS box, ease factor).

• If your verses have different mastery levels (for example, if you sometimes review by individual page and sometimes by Hizb): already well-consolidated verses (which had 4★ or 5★) are preserved. If you assign a low overall score (e.g., 2★) to the entire portion during evaluation, these strong verses will not collapse directly to 2★. The application applies a progressive drop of a single star level (from 5★ to 4★) to reflect a slight decrease in vigilance without resetting all your work.
• If all verses in the portion share the same history (for example, if you always review a uniform block from the beginning): all verses in the block will drop consistently by one star level (for example from 5★ to 4★).

• Portion Review (Memorization) manages your main study calendar (planned by your Hizbs or active pages). It directly updates the main SRS boxes, managing the due dates of your daily reviews.
• The Flashcard Training Space offers targeted mental gymnastics exercises to work on sensitive points (finding the next verse, fill-in-the-blank texts). Training sessions in Flashcard mode have their own distinct SRS boxes so that your mental gymnastics exercises do not disrupt or overload your core review calendar.

In Flashcards mode, if you make a mistake, you can categorize the nature of your slip:

• Lapse: Forgetting a word or a transition.
• Substitution: Confusion or replacement of one word with another (very common with similar verses, or Mutashabihat).
• Harakat: Error in pronunciation, vowel, or grammatical declension.
• Block: Complete blockage or memory blank in front of the card.

These classifications are saved in the history and feed your statistics to help you target your areas for improvement precisely.

The majority of teachers or Quranic schools work in environments where internet access is unstable, restricted, or non-existent. A local tracker on the teacher's tablet or phone guarantees:
1. Permanent availability of student tracking records, without network dependence.
2. Absolute privacy protection for minors (no student data, names, or progression is hosted online or accessible by third parties).
3. Simplicity of use stripped of account creation, passwords, or complex synchronization.

The main review mode of the application is organized around four successive steps:
1. Recite: Only the first verse of the portion is displayed to give you the starting point. You recite the rest from memory (or using voice listening).
2. Reclassify: Before displaying the full text, you evaluate your performance (by tapping on the incorrect words) so that the application calculates your star rating.
3. Review: The full text is displayed, allowing you to stabilize your transitions. You can also listen to the audio recitation of the portion by a chosen reciter.
4. Restart: The application records the evaluation, updates the SRS deadlines, and proposes the next portion.

This configuration aims to prevent any passive visual cheating. The first verse serves only as an indicator and starting guide to identify the portion to recite, forcing you to retrieve the rest from memory.

Yes, this is completely normal and it is an important design choice.
Speech recognition (ASR) in Tarsikh should be seen as a concentration and recitation assistant. It serves to reveal words as you pronounce them from memory to help you follow your own reading rhythm.

For the application to function 100% offline, it uses a compact and lightweight speech recognition model (quantized in INT8 for mobile). Although effective, this local model does not have the industrial power required to perform error analysis or fine pronunciation correction.

This is why, at the end of your recitation, you go through the "Reclassify" phase where the full text is displayed: it is up to you to touch the words on which you stumbled to record your errors manually. The application will then faithfully adjust your score and your SRS calendar.

Yes, the application natively supports Hafs and Warsh readings (widely used in North and West Africa). The user can switch from one to the other in the settings. The application also supports the Latin transliteration of the Hafs recitation.

This technical feat is possible thanks to two key elements implemented in our native bridge:
1. INT8 Quantification: The Zipformer deep learning model (quran_phoneme_zipformer.int8.onnx) is compressed to 8-bit precision. This reduces its size and drastically decreases the required mathematical calculations.
2. The Sherpa-ONNX Engine: Written in highly optimized C++, this execution engine loads the model directly into memory and decodes the microphone audio stream on only 2 CPU threads. This guarantees real-time processing (latency below 100 ms) while preserving battery life.

The application focuses on the correctness of textual memorization (forgotten, inverted, substituted, or mispronounced words), which corresponds to the primary objective of active review.

Absolutely not.
The audio stream recorded by the microphone (16-bit PCM sampled at 16 kHz) is normalized on the fly into a floating-point array and sent directly to the volatile RAM of the local ONNX decoder.
No audio file is written to the phone's hard drive, no recording is kept, and no voice data is sent over the internet. Your recitation remains strictly private.

The application integrates an entirely anonymous scientific research protocol studying human learning and memory mechanisms applied to Quran memorization. If the user consents, their review history is analyzed to help researchers model long-term forgetting curves more accurately.

No, there is no automatic or continuous background synchronization.
To participate, you must perform a manual action:
1. Sending data (in-app): In the Settings screen, you must actively click the "Share review data" button. At that precise moment, the application extracts your review history (dates, boxes reached, number of errors, reaction times, and estimated review durations) in chunks to send it to the research server. The application stores the ID of the last row sent locally. If you perform new reviews after this click, they will remain strictly on your phone, unless you decide to click the button again.
2. Demographic questionnaire (optional, web): To provide more scientific value to the data, you can copy the unique anonymous identifier (UUID) from your application, go to tarsikh.com/study and fill out a quick demographic questionnaire (age, mother tongue, learning method).

Upon installing the application, a random identifier (UUID v4 conforming to RFC4122) prefixed with usr- is generated locally. This identifier is not linked to any civil identity information (name, e-mail) or unique hardware identifiers of the phone (like MAC address or IMEI). It is therefore mathematically impossible for researchers to trace the physical identity of a participant.

No. The application is completely free, with no ads or in-app purchases. It is developed in a spirit of community service and scientific research, preserving the user from any visual or commercial distraction during their moments of study and memorization.

Tarsikh adopts an offline-first architecture. The Quranic texts (according to Hafs or Warsh readings), your history, and spacing algorithms are directly managed by an internal SQLite database stored locally on your phone. Likewise, the speech recognition engine runs locally, allowing you to review even in the middle of the desert, on a plane, or in areas not covered by mobile networks.

This is the necessary trade-off for the complete absence of third-party servers:

• The application embeds the complete SQLite database containing the entire Quran (6,236 verses, written in multiple scripts, transliterations, and phonetic targets).
• It integrates ONNX engine compilation binaries specific to your mobile processor architecture (ARM64, x86_64) to run AI locally.
• It stores the Zipformer acoustic model locally.

To help you maintain your rhythm, the app schedules local reminder notifications after 24 hours and 7 days of inactivity; these notifications are disabled by default in the settings.