Modeling File Recovery

📋 Overview

Mode 0 is the most flexible seed recovery method. You provide a model file that describes candidate word groups for each position in the mnemonic. wrecover explores all valid combinations satisfying the constraints and tests each against the target address.

Use Mode 0 when you have multiple candidate words per position, or need complex constraints (required words, fixed positions). For simple missing-word recovery with known positions, Mode 1 is simpler.

💡 GPU Supported

Mode 0 supports GPU acceleration (CUDA and OpenCL) in the Pro version, providing up to ~1000× speedup over a single CPU thread for 24-word mnemonics.

📝 Model File Format

The model file is a plain text file (.txt). Each line represents one word slot in the mnemonic. Multiple words on a line are candidates for that slot — wrecover tries all of them. The number of lines must be ≥ the mnemonic length (-l).

Basic example — 12-word seed, position 4 uncertain

# model.txt — one line per mnemonic position
# Multiple words on a line = candidates for that slot 

abandon 

ability 

able 

about above absent absorb absurd 

accuse 

accident 

account 

zoo 

zone 

zebra 

young 

yellow

Line modifiers

Prefix	Meaning	Example
(none)	Regular group — can go in any free slot	`abandon ability able`
`@`	Required — one of these MUST appear (any position)	`@ accuse account`
`@#N`	Required AND fixed at position N (1-based)	`@#3 about above`

⚠️ Modifier syntax

The modifier (@, @#N) must be the first token on the line, followed by the candidate words. Position indices are 1-based. The number of lines must be ≥ the value of -l.

Example with modifiers

# 12-word seed: know words 1-3 and 6-12, position 5 is
# mandatory "accuse" or "account", positions 4 is one of 3 options

abandon

ability

able

about above absent       # 3 candidates

@#5 accuse account       # position 5: required, fixed here

accident

absorb

zoo

zone

zebra

young

yellow

🚀 Commands

Mode 0 always requires work-file flags: -ws for a new search or -wi to resume. The address type and default derivation path are auto-detected from the address format.

Start a new search

wrecover -l 12 -m 0 -cf model.txt -addr "1A1zP1eP5QGefi2DMPTfTL5SLmv7DivfNa" -wlang en -ws -wf recovery.wrec -wp "this workfile password"

With GPU (Pro version)

wrecover -l 12 -m 0 -cf model.txt -addr "1A1zP1eP5QGefi2DMPTfTL5SLmv7DivfNa" -wlang en -ws -wf recovery.wrec -wp "this workfile password" -gpu -grid 2048 -gblock 256

Resume from work file

 # CPU

            wrecover -wi -wf recovery.wrec -wp "this workfile password"

            # GPU (Pro)

            wrecover -wi -wf recovery.wrec -wp "this workfile password" -gpu -grid 2048 -gblock 256

With a known BIP39 passphrase

wrecover -l 12 -m 0 -cf model.txt -addr "1A1zP1eP5QGefi2DMPTfTL5SLmv7DivfNa" -wlang en -pp "MyKnownPassphrase" -ws -wf recovery.wrec -wp "this workfile password"

Create work file only (for later distribution)

wrecover -l 12 -m 0 -cf model.txt -addr "1A1zP1eP5QGefi2DMPTfTL5SLmv7DivfNa" -wlang en -ws -wf recovery.wrec -wp "this workfile password" -save-only

📖 Parameter Reference

Option	Description	Required
`-l 12\|18\|24`	Mnemonic length	Yes
`-m 0`	Select Mode 0	Yes
`-cf <file.txt>`	Path to model file (.txt extension required)	Yes
`-addr <address>`	Target blockchain address (type auto-detected)	Yes
`-wlang <code>`	BIP39 wordlist language (see table below)	Yes
`-ws`	New search — start recovery and save progress	Yes*
`-wi`	Resume from existing work file	Yes*
`-wf <file.wrec>`	Work file path (.wrec extension recommended)	Yes
`-pp <passphrase>`	Known BIP39 passphrase, if used	No
`-dpath <path>`	Override derivation path (default: auto from address)	No
`-t <N>`	CPU threads (Free: capped at 10; Pro: unlimited)	No
`-gpu`	Enable GPU (Pro version — CUDA or OpenCL)	No
`-grid <N>`	GPU grid size (Pro, default: 256)	No
`-gblock <N>`	GPU threads per block (Pro, default: 256, max: 256)	No
`-wp <password>`	Encrypt the work file with a password	No
`-save-only`	Create work file without starting recovery	No
`-startp <N>`	Start at combination N — distributed recovery (Pro)	No
`-endp <N>`	Stop at combination N — distributed recovery (Pro)	No

* Use either -ws (new search) or -wi (resume) — not both.

🌐 Supported Languages & Address Types

Languages (`-wlang`)

Code	Language
`en`	English
`es`	Spanish
`fr`	French
`it`	Italian

Address types (auto-detected)

Address format	Type	Default derivation path
Starts with `1`	Bitcoin Legacy (P2PKH)	`m/44'/0'/0'/0`
Starts with `3`	Bitcoin SegWit P2SH	`m/49'/0'/0'/0`
Starts with `bc1q`	Bitcoin Native SegWit	`m/84'/0'/0'/0`
Starts with `0x`	Ethereum	`m/44'/60'/0'/0`

🔄 Distributed Recovery

Copy the .wrec work file to multiple machines and run simultaneously. Each instance automatically claims and processes different chunks — no coordination required.

# Create the work file on one machine

wrecover -l 18 -m 0 -cf model.txt -addr "bc1qxy2kgdygjrsqtzq2n0yrf2493p83kkfjhx0wlh" -wlang en -ws -wf recovery.wrec -wp "this workfile password" -save-only 

# For manual range splitting (Pro version):

wrecover -l 18 -m 0 -cf model.txt -addr "bc1qxy2kgdygjrsqtzq2n0yrf2493p83kkfjhx0wlh" -wlang en -ws -wf recovery.wrec -wp "this workfile password" -startp 0 -endp 5000000000 -save-only 

wrecover -l 18 -m 0 -cf model.txt -addr "bc1qxy2kgdygjrsqtzq2n0yrf2493p83kkfjhx0wlh" -wlang en -ws -wf recovery.wrec -wp "this workfile password" -startp 5000000000 -endp 10000000000 -save-only 

wrecover -l 18 -m 0 -cf model.txt -addr "bc1qxy2kgdygjrsqtzq2n0yrf2493p83kkfjhx0wlh" -wlang en -ws -wf recovery.wrec -wp "this workfile password" -startp 10000000000 -save-only



    
    
        ⚡ Performance
        
            Hardware 12-word seed 18-word seed 24-word seed
            
                CPU — per thread (Free & Pro) 3k–4k seeds/sec 11k–13k seeds/sec 30k–32k seeds/sec
                NVIDIA Tesla T4 (Pro) ~2.6M seeds/sec ~9.8M seeds/sec ~21M seeds/sec
                NVIDIA RTX PRO 6000 (Pro) ~31.4M seeds/sec ~117.2M seeds/sec ~264M seeds/sec
                N× GPUs (Pro) Linear scaling: N × single-GPU throughput
            
        
        
            💡 Free vs Pro
            Free version: CPU only, capped at 10 threads. Pro version: unlimited CPU threads + GPU acceleration (CUDA & OpenCL) + distributed range control (-startp/-endp). CPU throughput scales linearly with thread count.
        
    

    
    
        🔧 Troubleshooting
        
            Error: "phrase length is greater than available words list"
            The model file has fewer lines than the value of -l. Add more lines or reduce the mnemonic length.
        
        
            Error: "Invalid or inaccessible Modeling file"
            Verify the file path, read permissions, and UTF-8 encoding. All words must be valid BIP39 words for the selected language.
        
        
            Error: "Corrupted data or incorrect password"
            The work file is encrypted with PBKDF2-SHA512. If you used -wp when creating it, provide the same password on resume with -wi -wp <password>.
        
        
            Error: "-startp/-endp not available for free version"
            Range control for distributed recovery is a Pro-only feature. The Free version processes the full search space sequentially.
        
    

    
        ← Back to Documentation

Hardware	12-word seed	18-word seed	24-word seed
CPU — per thread (Free & Pro)	3k–4k seeds/sec	11k–13k seeds/sec	30k–32k seeds/sec
NVIDIA Tesla T4 (Pro)	~2.6M seeds/sec	~9.8M seeds/sec	~21M seeds/sec
NVIDIA RTX PRO 6000 (Pro)	~31.4M seeds/sec	~117.2M seeds/sec	~264M seeds/sec
N× GPUs (Pro)	Linear scaling: N × single-GPU throughput