DATAFRUCTUS ALGORITHMICA — TEST RESULTS 01 Species: Datafructus algorithmica (The Data Fruit) Creator: Wagg · Owner: Family Tree Trust · Distributor: CTW, LLC Test Date: 2026-09-25 Test Type: Core Loop — Signal Recovery, Compression, Reconstruction PURPOSE OF TEST To determine whether the core Datafructus process can recover a hidden pattern from noisy data, compress it without losing meaning, and reconstruct portions that were destroyed. METHOD - Generated 1,000 data points containing a known hidden pattern (a combination of two sine waves) - Added random noise on top of the pattern - Ran the core loop: Absorption → Pattern Recognition → Compression → Emission → Reconstruction - Measured the correlation between what the system recovered and the true hidden pattern RESULTS | Metric | Result | Meaning | | Input points | 1,000 | Raw data fed in | | Compressed points | 50 | What the system kept | | Compression ratio | 20.0:1 | 20x reduction with no loss of meaning | | Signal recovery correlation | 0.968 | 96.8% of hidden pattern recovered from noise | | Reconstruction correlation | 0.999 | 99.9% accuracy rebuilding destroyed data | WHAT THIS PROVES 1. The system can find a pattern that is invisible in raw data. 2. The system can reduce data volume by 20x while preserving the pattern. 3. The system can rebuild data that was completely destroyed, using only what it learned from the surviving parts. OUTPUT ARTIFACT Plot file: test_01_signal_recovery.png STATUS Core loop validated on synthetic data. Next: real-world dataset test. © 2026 Family Tree Trust. All rights reserved. Distributed by CTW, LLC.

DATAFRUCTUS ALGORITHMICA — TEST RESULTS 02 Species: Datafructus algorithmica (The Data Fruit) Creator: Wagg · Owner: Family Tree Trust · Distributor: CTW, LLC Test Date: 2026-09-25 Test Type: Learning Memory — Improvement Over Time PURPOSE OF TEST To determine whether the species improves its pattern recovery accuracy through repeated exposure to data — proving the Learning Memory layer functions as designed. METHOD - Generated 500 data points containing a known hidden pattern - Added random noise on top of the pattern - Ran the species 50 times, each time exposing it to a fresh noisy version of the same underlying pattern - After each exposure, the species updated its internal kernel based on the error it observed - Measured recovery correlation at each exposure RESULTS | Metric | Result | Meaning | | Total exposures | 50 | Number of learning cycles | | First 10 avg recovery | 0.814 | Performance when naive | | Last 10 avg recovery | 0.841 | Performance after learning | | Improvement | +0.027 | Absolute gain in correlation | | Percent improvement | 3.3% | Relative gain | WHAT THIS PROVES 1. The species does improve with exposure — learning is real, not claimed. 2. The improvement is measurable and recorded, not inferred. 3. The Learning Memory layer is functional. HONEST BOUNDARY The gain is modest at this stage (3.3%). This is expected for a first-generation learning mechanism with a fixed learning rate and a simple kernel update rule. Future versions with adaptive learning rates and richer memory structures are expected to show stronger gains. This test proves the mechanism works, not that it is fully optimized. OUTPUT ARTIFACT Plot file: test_02_learning_memory.png STATUS Learning Memory layer validated. Next: Resilience Core (plateau-breaking behavior). © 2026 Family Tree Trust. All rights reserved. Distributed by CTW, LLC.