MEMORIA VIVUM — OFFICIAL TEST REPORT Test ID: MV-013 (Real Tags — Meaningfulness Retest) Species: Memoria vivum (The Living Archive) Creator: Wagg · Owner: Family Tree Trust Date: 2026-09-26 System: Diamond Server (Python 3, runnable engine) Document Type: Full Technical Record PURPOSE OF TEST Prior tests (MV-011, MV-012) showed that 88% of the species' connections were coincidental. Two attempted fixes failed: - v1.9 Word Weighting — no change (11.8%) - v2.0 Two-Word Rule alone — swung too far (1 connection) This test diagnoses the real cause. The problem was not the connection rule. The problem was how the input was tagged. THE DIAGNOSIS Prior tests built tags from the first 15 long words in each abstract. Research abstracts almost always start with filler phrases like "recent studies have shown..." So every paper's tags looked similar — intro words, not topic words. The species was connecting on intro words, which is why connections were coincidental. Two real papers on the same topic would start with different intro phrases and share nothing in the first 15 words — which is why the strict version (v2.0 alone) found almost nothing. THE FIX Tag extraction was changed to use term frequency instead of first-appearance. Words that appear most often in an abstract are the actual topic. Intro words appear once and drop out naturally. No engine change. Same v2.0 code. Only the input changed. METHOD 1. Fetched 25 Zenodo records (neuroscience) 2. Filtered to 13 with usable abstracts 3. Built tags using term frequency — the most-repeated words 4. Fed into v2.0 engine 5. Ran 5 growth cycles 6. Measured connection quality SAMPLE TAGS Paper 1: cryopreserved, electron, microscopy, images, cortex, thalamus, white, matter, both Paper 2: buildup, neural, self, initiated, movement, studies, intention, brain, both, invasive, preparation These are real topic words, not intro filler. ACTUAL RESULTS | Metric | Value | | Papers | 13 | | Total connections | 6 | | STRONG (3+ shared) | 1 (16.7%) | | MODERATE (2 shared) | 2 (33.3%) | | WEAK (0-1 shared) | 3 (50.0%) | | Real connections | 3 / 6 (50.0%) | VERDICT MIXED — real improvement, but not fully meaningful yet. COMPARISON ACROSS ALL TESTS | Version | Connections | Real % | | MV-011 (v1.8) | 17 | 11.8% | | MV-012 (v1.9) | 17 | 11.8% | | v2.0 alone | 1 | 100% | | MV-013 (v2.0 + real tags) | 6 | 50.0% | WHAT THIS SHOWS 1. The real cause was tag quality, not the connection rule. Better tags produced better connections without touching the engine. 2. 50% real is a 4x improvement over the 11.8% baseline. 3. The species is finding real connections. The samples show topic words doing the connecting, not filler. 4. It is not finished. 50% is "mixed" — half the connections are still coincidental. The remaining noise comes from leftover filler words (both, studies, recent) still reaching the tag list. HONEST BOUNDARY 50% real connections is better than before but below the 70% threshold for "meaningful." The species is on the right track, but the tag extraction is not yet clean enough to eliminate all coincidences. NEXT STEPS - Further refine tag extraction to strip residual filler words. - Consider using Zenodo's official subject keywords instead of derived tags. OUTPUT ARTIFACT Script: scripts/test_mv14_realtags.py Signed: Wagg · Owner: Family Tree Trust · Date: 2026-09-26 © 2026 Family Tree Trust. All rights reserved. Distributed by CTW, LLC.

MEMORIA VIVUM — OFFICIAL TEST REPORT Test ID: MV-014 (v2.2 — Anchor Model with Heart) Species: Memoria vivum (The Living Archive) Creator: Wagg · Owner: Family Tree Trust Date: 2026-09-26 System: Diamond Server (Python 3, runnable engine) Document Type: Full Technical Record PURPOSE OF TEST The v2.2 engine was built to match the profile for the first time. This is the first run of the Anchor Model with the Heart, Relationship Levels, Release-as-Archive, and No Cascade. The test asks one question: does the engine now do what the profile says it does? WHAT WAS BUILT The v2.2 engine implements: 1. The Heart — a sealed identity anchor set before the first memory arrives. Every memory connects to the heart through its identity anchor. 2. The Anchor Model — every memory carries anchors. Connections form on shared anchors. 3. Relationship Levels — 1 shared anchor weak, 2 real, 3+ strong bond. 4. Release as Archive — releasing a memory moves it to the archive. The trace remains. The connections stay. 5. Three States — Active, Archived, Destroyed. Destroyed is only user-authorized. 6. No Cascade — releasing one memory does not disturb any other. METHOD 1. Sealed the heart with the identity anchor wagg-family-archive 2. Absorbed 7 real memories with realistic anchors — dates, people, places, themes 3. Ran 3 growth cycles 4. Read every thread by level 5. Tested recall with context 6. Tested release 7. Tested destroy THE 7 MEMORIES - photo_thanksgiving_2024 - photo_grandma_birthday_2024 - msg_mom_about_thanksgiving - note_grandma_cornbread - photo_family_reunion_2024 - voice_grandma_story - doc_grandpa_military ACTUAL RESULTS AFTER WEAVING | Metric | Value | | In memory | 7 | | Anchors indexed | 15 | | Weak threads | 8 | | Real threads | 12 | | Strong bonds | 8 | | Verified | 6 | SAMPLE RELATIONSHIPS | Memory A | Memory B | Level | Shared Anchors | | photo_thanksgiving_2024 | photo_grandma_birthday_2024 | STRONG | 2024, family, home, wagg-family-archive | | photo_thanksgiving_2024 | msg_mom_about_thanksgiving | STRONG | 2024, family, thanksgiving, wagg-family-archive | | photo_grandma_birthday_2024 | note_grandma_cornbread | STRONG | grandma, home, wagg-family-archive | | doc_grandpa_military | photo_family_reunion_2024 | WEAK | wagg-family-archive | RECALL WITH CONTEXT Query: ["grandma"] - Core: voice_grandma_story, photo_grandma_birthday_2024, note_grandma_cornbread - Context: doc_grandpa_military, msg_mom_about_thanksgiving, photo_family_reunion_2024, photo_thanksgiving_2024 Recall returned three memories about grandma, plus four more memories that were connected through the family network. The context came with the answer, exactly as the profile describes. RELEASE TEST - Before: 7 active, 0 archived - After releasing doc_grandpa_military: 6 active, 1 archived - The trace remains. The connections stay. DESTROY TEST - Before: 6 active - After destroying note_grandma_cornbread: 5 active, 1 archived - Complete removal. No trace. Only user authorization produced it. VERDICT PASS — the engine now matches the profile. WHAT THIS SHOWS 1. The Heart works. Every memory carries the identity anchor, so every memory is connected to the heart. Nothing floats outside the tree. 2. Relationship levels work. Strong bonds formed where three or more anchors overlapped. Weak threads formed where only the identity anchor was shared. The engine reports exactly what the profile says it should. 3. Recall with context works. Asking for "grandma" returned three direct matches and four surrounding memories. The answer came with the shape of the moment, not just the pieces. 4. Release as Archive works. The memory moved from active to archived. The trace remained. No cascade. Nothing else changed. 5. Destroy works. Complete removal, only on request. The remaining memories were not affected. 6. The engine and the profile finally line up. Every claim in the profile — Heart, Anchor Model, Relationship Levels, Release as Archive, Three States, No Cascade — is now implemented and demonstrated. HONEST BOUNDARY The test used 7 hand-authored memories with well-distributed anchors. Real data from a real device will have messier anchors, missing fields, and unknown relationships. This test proves the engine matches the profile. It does not prove how the engine performs on real personal data at scale. That is the next test. WHAT THIS TEST PROVES 1. Memoria vivum v2.2 exists as a runnable engine. 2. Every layer of the profile is now implemented in code. 3. The engine produces exactly the behavior the profile describes. 4. The Heart, the Anchor Model, and the three-state memory system all work as designed. OUTPUT ARTIFACT Engine: memoria_vivum_v2_1.py (contains v2.2 code) Result file: test_mv14_two_word.txt Signed: Wagg · Owner: Family Tree Trust · Date: 2026-09-26 © 2026 Family Tree Trust. All rights reserved. Distributed by CTW, LLC.