WFGY/ProblemMap/memory-design-patterns.md

🧠 Memory Design Patterns

From scratchpads to long-range project recall — keep context alive without drowning your LLM.

Why this page?
Most “memory” demos either spam the full chat history or store random embeddings that never round-trip.
WFGY treats memory as structured semantic nodes with ΔS / λ_observe guards, so old context helps — never hurts — new reasoning.

---

1 · Symptoms

Symptom	Typical Surface Clue
Context forgotten after restart	“Sorry, I don’t recall” / model re-asks user
Memory leak / self-contradiction	Old decisions resurface in wrong branch
JSON-based vector store grows unbounded	Latency ↑, RAG recall quality ↓
Fine-tune attempted just to “remember”	Model cost ↑, still hallucinates

---

2 · Root Causes

1. Flat Logs — raw transcripts appended forever.
2. Embedding Dump — every user sentence embedded → no semantic filter.
3. No Boundary Check — divergent memories injected mid-task.
4. Write-Only Memory — model never reads / revalidates stored facts.

Result: either forget everything or remember garbage.

---

3 · WFGY Fix Path (at a glance)

Stage	Tool / Module	ΔS Guard	Outcome
⬇️ Capture	BBMC node writer	record only if ΔS ≥ 0.60 (or 0.40–0.60 & λ ∈ {←, <>})	Stores semantic not verbatim memory
🗂️ Index	λ_observe classifier	tag λ trend for each node	Enables topic-group navigation
🔍 Recall	BBPF path search	choose node set with ΣΔS minimal	Retrieves tight, non-bloated context
🩹 Repair	BBCR fallback	detect stale/contradict nodes	Auto-patch or prompt for user merge

80 % of memory bugs vanish after enforcing this four-step loop.

---

4 · Design Patterns Library

Pattern	Use-Case	How it Works	ΔS Budget
✏️ Scratch Node	quick calc / throw-away idea	24 h TTL field; auto-purged	0.40–0.55
📚 Topic Shelf	multi-day research thread	one node per subtopic; λ → convergent	< 0.45
🗓️ Daily Digest	running project log	rollup 10 low-ΔS nodes → 1 summary	–
🎯 Anchor Fact	must-not-forget constraint	pinned; override recall rank	0.05

All stored in a single lightweight JSONL: {topic, ΔS, λ, text, ttl}

---

5 · Step-by-Step Implementation

Prereqs: any model that can embed & run basic python (or LangChain, Llama-index, etc.).

# 1. capture
deltaS = cosine(question_vec, context_vec)
if deltaS >= 0.60 or (0.40 <= deltaS <= 0.60 and lambda_state in ["divergent","recursive"]):
    node = {"topic": topic, "ΔS": round(deltaS,3), "λ": lambda_state, "text": insight}
    memory.append(node)

# 2. recall
candidates = [n for n in memory if n["topic"]==current_topic]
best_path = sorted(candidates, key=lambda n:n["ΔS"])[:5]
prompt_context = "\n".join(n["text"] for n in best_path)

Minimal prompt

System: Use WFGY memory nodes below (+latest question) to answer.
Memory Nodes:
{{prompt_context}}
---
Question: {{user}}

---

6 · Common Pitfalls & Tests

Pitfall	Quick Test	WFGY Fix
“Context bloat, tokens 8k → 40k”	node count > 200? run rollup.py	Daily Digest pattern
“Conflicting facts”	ΔS(anchor, candidate) > 0.70	BBCR prompts merge
“Retrieval too slow”	recall > 200 ms	Pre-index by λ & time

---

7 · Cheat-Sheet

ΔS save threshold   = 0.60
ΔS recall window    = top-k by lowest ΔS
λ tags              = → ← <> ×
TTL (scratch)       = 24 h
Rollup trigger      = >10 nodes / topic / day

Store this as memory.cfg; loader reads defaults at boot.

---

8 · Next Actions

1. Prototype with 20 nodes → verify recall accuracy.
2. Enable Rollup once node count > 200.
3. Add Trace Logger to diff answers with / without memory.

---

🔗 Quick-Start Downloads (60 sec)

Tool	Link	3-Step Setup
WFGY 1.0 PDF	Engine Paper	1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + <your question>”
TXT OS (plain-text OS)	TXTOS.txt	1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly

---

Explore More

Layer	Page	What it’s for
Proof	WFGY Recognition Map	External citations, integrations, and ecosystem proof
Engine	WFGY 1.0	Original PDF based tension engine
Engine	WFGY 2.0	Production tension kernel and math engine for RAG and agents
Engine	WFGY 3.0	TXT based Singularity tension engine, 131 S class set
Map	Problem Map 1.0	Flagship 16 problem RAG failure checklist and fix map
Map	Problem Map 2.0	RAG focused recovery pipeline
Map	Problem Map 3.0	Global Debug Card, image as a debug protocol layer
Map	Semantic Clinic	Symptom to family to exact fix
Map	Grandma’s Clinic	Plain language stories mapped to Problem Map 1.0
Onboarding	Starter Village	Guided tour for newcomers
App	TXT OS	TXT semantic OS, fast boot
App	Blah Blah Blah	Abstract and paradox Q and A built on TXT OS
App	Blur Blur Blur	Text to image with semantic control
App	Blow Blow Blow	Reasoning game engine and memory demo

If this repository helped, starring it improves discovery so more builders can find the docs and tools.