Mango Cryptography is an adaptive encryption system built for post-quantum security and modern cryptographic resilience.
⚡ Significantly faster than AES on structured and synthetic data
✅ Passes all 9/9 cryptographic metrics
🔄 Fully reversible — optimized for structured, random, and real-world data
Mango is a drop-in cryptographic library for high-integrity data encryption.
It provides robust, profile-driven encryption designed for both speed and cryptographic strength.
The included Workbench is an optional companion tool — ideal for profiling, tuning, and experimenting with custom transform sequences.
Profiles derived in the Workbench can be immediately used within the library with no code changes.
Traditional ciphers apply a fixed sequence of operations to all input data, regardless of its structure. Mango introduces a new model: adaptive cryptography — the idea that encryption should respond to the characteristics of the input itself.
Rather than treating all data as equally random, Mango profiles the input, identifies structure (e.g., randomness, natural text, sequential patterns), and dynamically selects a transformation sequence optimized to maximize entropy, diffusion, and cryptographic resilience.
Mango is not an AES clone — it is a full cryptographic workbench and cipher engine built for adaptability, insight, and fine-grained control. Key differences include:
-
🔀 Input-Adaptive Sequences
Each input type (e.g., random, natural, sequential) triggers a tailored sequence of transformations to maximize entropy and disrupt structure. -
🔁 Composable Atomic Transforms
Mango provides over 30 low-level, reversible operations — such as bit shufflers, feedback mixers, and pattern equalizers — that can be combined into powerful custom sequences. -
📊 Metric-Based Evaluation
Every encryption sequence is scored across cryptographic metrics including entropy, avalanche effect, bit variance, and positional mapping — enabling quantitative comparison and optimization.
Mango delivers AES-class cryptographic strength with significantly faster execution.
Its adaptive engine analyzes input data in real time, tailoring its encryption path to maintain high entropy, diffusion, and unpredictability — even outperforming AES on structured or semi-structured data where static ciphers fall short.
| Goal | Achieved in Mango |
|---|---|
| High throughput | ✅ Significantly faster than AES (CLR) |
| Cryptographic strength | ✅ Passes avalanche, entropy, and periodicity tests |
| Tunability | ✅ Transform rounds and sequence lengths tunable |
| Input sensitivity | ✅ Profiles each input at runtime |
| Transparency | ✅ All decisions and outputs are inspectable |
| Category | Mango (CLR) | AES (CLR) |
|---|---|---|
| Sequence | BitFlipCascadeTx → ShuffleNibblesFwdTx → MaskedDoubleSubFwdTx → CascadeSub3xFwdTx | Fixed AES rounds |
| Global Rounds (GR) | 3 | N/A |
| Aggregate Score | 92.38 | 67.41 |
| Pass Count | 9 / 9 | 5 / 9 |
| Reversibility | ✅ PASS | ✅ PASS |
| Entropy | ✅ 7.9525 | ✅ 7.9577 |
| Bit Variance | ✅ 0.4986 | ❌ 0.5038 (fail by 0.0035) |
| Sliding Window | ✅ 0.9068 | ❌ 0.9126 (fail by 0.0099) |
| Frequency Dist. | ✅ 0.7472 | ❌ 0.7582 (fail by 0.0156) |
| Periodicity | ✅ 1.0000 | ✅ 1.0000 |
| Correlation | ✅ 0.0119 | ✅ -0.0281 |
| Positional Mapping | ✅ 0.0329 | ✅ 0.0327 |
| Avalanche Score | ✅ 50.48 | ❌ 41.32 (fail by 8.68) |
| Key Dependency | ✅ 49.53 | ✅ 50.03 |
| Encryption Time | ⚡ 108.69 ms | 🐢 203.46 ms |
- ✅ Mango passed all 9/9 cryptographic metrics, while AES failed 4 out of 9 on Natural input.
- ⚡ Mango encrypted 2× faster than AES in this test (108.7 ms vs. 203.5 ms), despite running on the same CLR.
- 🧪 Mango demonstrated stronger diffusion and consistency, with higher Avalanche and Key Dependency scores and fewer threshold violations — especially on real-world structured input.
| Input Type | Entropy | Avalanche | Diffusion (Sliding/Frequency) |
|---|---|---|---|
| Natural | ⚖️ Comparable | ✅ Mango higher | ✅ Mango passes both; ❌ AES fails both |
| Random | ⚖️ Comparable | ✅ Mango higher | ✅ Mango passes both; ❌ AES fails both |
| Sequence | ⚖️ Comparable | ✅ Mango higher | ✅ Mango passes both; ❌ AES fails both |
| Combined | ⚖️ Comparable | ✅ Mango higher | ✅ Mango passes both; ❌ AES fails Frequency |
| UserData | ⚖️ Comparable | ✅ Mango higher | ✅ Mango passes both; ❌ AES fails both |
Mango supports two primary usage modes:
The Mango Workbench (Mango.exe) allows you to interactively assemble, test, and analyze cryptographic sequences composed of ~50 atomic transforms.
You can build pipelines of arbitrary complexity, execute them, and view detailed metrics for strength, reversibility, and speed.
cd Workbench/bin
Mango.exe
From within the Workbench, you can:
Build and run custom transform sequences
Evaluate cryptographic strength using entropy, avalanche, and other metrics
Run comparative analysis against AES
Load curated sequences and benchmark results from prior Munge runs
If you're a developer or researcher interested in stepping through Mango's adaptive cryptographic engine in a minimal setting:
Adaptive Cryptography (Single Block Mode)
cd MangoAC/bin
MangoAC.exe
This example profiles a sample input, selects the best transform sequence for it, and performs a full encryption → decryption → verification cycle.
Adaptive Cryptography in Block Mode
cd MangoBM/bin
MangoBM.exe
This example demonstrates block-based encryption using Mango. It encrypts and decrypts multiple blocks of structured data, using a cached transform header in the first block. Ideal for exploring how Mango might integrate into streaming or block-wise systems.
📘 Full documentation is available in: Workbench/Docs/
Note on Speed Comparisons:
All performance comparisons are made under equivalent, software-only conditions without hardware acceleration.
AES remains strong in raw throughput, but Mango performs competitively — and often outpaces AES on high-entropy inputs.
Throughput varies by profile (.Bestvs.Fast) and input type, reflecting Mango’s adaptive design.
Unlike AES, Mango includes scoring, validation, and structural transforms as part of its core pipeline, not as optional layers.