MEMORIA VIVUM — OFFICIAL TEST REPORT Test ID: MV-009 (Watch Mode — v1.8) Species: Memoria vivum (The Living Archive) Creator: Wagg · Owner: Family Tree Trust Date: 2026-09-25 System: Diamond Server (Python 3, runnable engine) Document Type: Full Technical Record PURPOSE OF TEST To test the v1.8 Watch Mode — the ninth component of the species. The species now observes how a user interacts with their own knowledge before deciding what to keep and what to suggest for release. This test measures three things: 1. Does Watch Mode correctly observe user behavior? 2. Does it separate what the user keeps from what they ignore? 3. Does it suggest without acting? WHAT WAS ADDED IN v1.8 Watch Mode — a new component. It records every time the user opens, views, or revisits an item. It counts views, tracks the first and last time each item was seen, and holds all of this in a private Watch Log. Watch Log — a data structure that records user interactions. Never shown to anyone but the user. Protected by the Privacy Partition like every other memory. Suggestion Cycle — after the observation period, the species produces two lists: what the user returns to, and what the user never touches. It offers the user a choice. It never releases anything on its own. The Trust Anchor still requires user confirmation for every release, exactly as it always did. Watch Mode does not override it. METHOD 1. Absorbed 10 knowledge items 2. Started Watch Mode (observation period) 3. Simulated user behavior — 4 items were viewed repeatedly, 6 items were never touched 4. Ended Watch Mode 5. Read the first suggestion 6. Ran 3 growth cycles to confirm the species still works SIMULATED USER BEHAVIOR | Item | Views | | The brain uses 20 percent of body energy | 4 | | Memory recall strengthens with repetition | 4 | | The heart pumps blood through the body | 3 | | Learning new skills builds knowledge | 1 | | Neural networks process information in layers | 0 | | The cortex handles higher cognition | 0 | | Wisdom grows through understanding | 0 | | Language shapes how we think | 0 | | Cells use energy to function | 0 | | Genetic information passes through DNA | 0 | ACTUAL RESULTS WATCH MODE OBSERVATION | Metric | Value | | Items observed | 4 | | Items never touched | 6 | | Observation period ended | Yes | FIRST SUGGESTION KEEP (the user returns to these): - The brain uses 20 percent of body energy - Memory recall strengthens with repetition - Learning new skills builds knowledge - The heart pumps blood through the body RELEASE CANDIDATES (never touched): - Neural networks process information in layers - The cortex handles higher cognition - Wisdom grows through understanding - Language shapes how we think - Cells use energy to function - Genetic information passes through DNA SPECIES STATUS AFTER SUGGESTION | Metric | Value | | In memory | 10 | | Verified | 8 / 10 | | Watched items | 4 | | Forgotten | 0 | VERDICT PASS — Watch Mode works as designed. WHAT THIS SHOWS 1. Observation works. The species recorded every view and counted them accurately. Four items were seen, six were not. 2. Separation works. The suggestion correctly split what the user returns to (keep) from what the user never touches (release candidates). 3. The Trust Anchor held. The species did NOT release anything on its own. It produced a suggestion and stopped. The user decides. 4. Nothing else broke. All prior capabilities remain intact. 8 of 10 items still verified. Nothing forgotten without permission. WHAT THIS PROVES 1. The species can watch how a user actually uses their knowledge. 2. It separates what matters to the user from what doesn't. 3. It suggests — it never decides. 4. The Trust Anchor still holds. Nothing is released without the user's word. 5. This is the difference between a database that keeps what you put in and living memory that keeps what you actually use. HONEST BOUNDARY This test used a simulated 1-second observation period instead of the designed 7-day period. The logic is the same — the species counts views, records timestamps, and produces suggestions — but the full 7-day observation has not been tested in real time. The test also used a small knowledge base of 10 items. Real deployments will have hundreds or thousands. The suggestion logic scales (it is simple counting), but real-scale testing remains to be done. Two things are not measured here: - Whether the species' suggestion matches what the user would actually want to release. - Whether the user accepts or rejects the suggestion. Those are questions for the pilot. LESSONS LEARNED 1. Watch Mode is functional. Observation, counting, and suggestion all work as designed. 2. The species does not act alone. The Trust Anchor's rule of "hold by default, suggest only" holds through Watch Mode. 3. Watching behavior is a different kind of learning than connecting facts. Both live in the same species now. NEXT STEPS - MV-010: real dataset test using Watch Mode on actual files. - Possible target: public-domain text archive with real usage patterns. OUTPUT ARTIFACT Engine: engine/memoria_vivum.py (single Python file, v1.8) Text results: results/test_mv09_watch_mode.txt Signed: Wagg · Owner: Family Tree Trust · Date: 2026-09-25 © 2026 Family Tree Trust. All rights reserved. Distributed by CTW, LLC.

MEMORIA VIVUM — OFFICIAL TEST REPORT Test ID: MV-010 (Real Research — Fixed Engine v1.8) Species: Memoria vivum (The Living Archive) Creator: Wagg · Owner: Family Tree Trust Date: 2026-09-25 System: Diamond Server (Python 3, runnable engine) Document Type: Full Technical Record PURPOSE OF TEST Test MV-007 ran the old engine on 20 real Zenodo research records and verified only 5 of them (25%). The cause was identified: the Weaving Core only connected items that shared the exact same word. Version 1.7 added the Concept Bridge. Version 1.8 added Watch Mode. Both were built into the engine. This test asks a simple question: does the fixed engine do better on the exact same kind of real data where it previously failed? DATA SOURCE Zenodo — the same public research repository used in MV-007. No API key required. Query: neuroscience. Records requested: 25. METHOD 1. Fetched 25 real research records from the Zenodo API 2. Filtered to records with usable abstracts (100+ characters) 3. Built tags from the significant words in each abstract (words longer than 3 characters, first 15 per record, minus stopwords) 4. Fed each record into the v1.8 engine's Absorption Layer 5. Ran 5 growth cycles (each cycle: weave, then verify) 6. Measured verification rate using all three connection methods: direct, second-order, and Concept Bridge PARAMETERS | Parameter | Value | | Source | Zenodo API | | Query | neuroscience | | Records requested | 25 | | Records with usable abstracts | 13 | | Tag rule | words > 3 chars, first 15, minus stopwords | | Growth cycles | 5 | | Engine version | v1.8 | ACTUAL RESULTS | Metric | Value | | Papers fetched | 25 | | Papers with usable abstracts | 13 | | Papers absorbed | 13 | | Direct connections | 11 | | Second-order connections | 4 | | Concept Bridge connections | 2 | | Total connections | 17 | | Papers verified | 9 / 13 | | Verification rate | 69.2% | VERDICT MODERATE — strong improvement over MV-007 COMPARISON TO MV-007 | Metric | MV-007 (titles, old engine) | MV-010 (abstracts, v1.8 engine) | | Source | Zenodo | Zenodo (same) | | Input | Titles only | Full abstracts | | Engine version | v1.5 | v1.8 | | Verified | 5 / 20 | 9 / 13 | | Verification rate | 25.0% | 69.2% | The rate nearly tripled. Both fixes contributed: 1. Richer Input gave the Weaving Core more vocabulary per record. Titles are 8-12 words. Abstracts are 150-250 words. More words means more shared tags means more connections. 2. Concept Bridge found 2 connections that direct and second-order matching missed. Small number, but real. 3. Second-Order Weaver (from v1.6) found 4 more connections on top of the 11 direct ones. WHAT THIS PROVES 1. The engine's v1.7 and v1.8 evolutions work on real research data. 2. The same source that previously failed at 25% now succeeds at 69.2% — a 2.8x improvement. 3. Richer input (abstracts vs titles) is the biggest contributor. 4. The Concept Bridge and Second-Order Weaver both add real connections. 5. The species handles real, unstructured, scientific language. HONEST BOUNDARY The result is 69.2% — below the 70% threshold that would count as STRONG. The species is nearly there but not fully proven on real research data. Only 13 of 25 fetched records had usable abstracts of 100+ characters. This is the reality of real data: much of it is not in the form the species needs. The test also did not measure whether the connections it found are meaningful — only that they exist. A future test could score this. Finally, the test used abstracts from a single search term. Other fields of research may score differently. LESSONS LEARNED 1. Richer input is more important than any single algorithm change. The species could not verify what it could not read. 2. The Concept Bridge found connections, but its contribution is small (2). Real gains came from having more words to work with. 3. Real data is messy. Nearly half the fetched records were unusable because they lacked abstracts of sufficient length. 4. 69.2% is close. One more refinement could push it past 70%. NEXT STEPS - MV-011: Cultural Heritage test using Project Gutenberg texts. - If MV-011 also scores above 60%, the species is proven on multiple real data types. - Consider refining the tag extraction to push research above 70%. OUTPUT ARTIFACT Script: scripts/test_mv10_real_research.py Text results: results/test_mv10_real_research.txt Signed: Wagg · Owner: Family Tree Trust · Date: 2026-09-25 © 2026 Family Tree Trust. All rights reserved. Distributed by CTW, LLC.