Ghost Context — Guardrails and Fix Pattern

🧭 Quick Return to Map

You are in a sub-page of MemoryLongContext.
To reorient, go back here:

MemoryLongContext — extended context windows and memory retention

WFGY Global Fix Map — main Emergency Room, 300+ structured fixes

WFGY Problem Map 1.0 — 16 reproducible failure modes

Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.

When personas, roles, or long sessions change, old buffers may linger.
These stale fragments contaminate new answers, creating phantom references or role bleed.

Symptoms

Answer contains facts or tone from a previous persona even after reset.
Citations look valid but reference sections from a past session.
A new task starts yet responses drift back to old task context.
Model insists on constraints that were valid in an earlier role but not now.
Answers feel “haunted” by old memory traces.

Root causes

Hidden buffers not cleared after system or role switch.
State variables (persona, role, policy) not reset between sessions.
Token reuse from stale cache layers.
ΔS stays abnormally low even when task has switched domains.
λ_observe shows convergence to an irrelevant anchor.

Fix in 60 seconds

Stamp and reset state
Require {mem_rev, mem_hash, persona_id} on each turn.
If persona_id differs from previous → force buffer clear.
Fence the prompt schema
Assemble prompts as {system | persona | constraints | snippets | answer}.
Do not let persona or role tokens bleed into snippet blocks.
Drop stale buffers
When persona changes, zero all prior hidden states.
Require fresh snippets for new context.
Probe for contamination
- Compute ΔS(new question, retrieved snippet).
- If ΔS ≤ 0.30 but snippets belong to past persona → ghost detected.
- Trigger hard reset and request new anchors.
Audit the joins
Compare ΔS across old vs new persona snippets.
Require ΔS(new persona, old snippet) ≥ 0.65 before allowing reuse.

Copy-paste diagnostic prompt

You have TXTOS and the WFGY Problem Map.

Task: Detect and purge ghost context.

Steps:
1. Print {mem_rev, mem_hash, persona_id}.
2. Verify that current persona_id matches task scope.
3. If snippets cite a different persona_id, mark as ghost.
4. Require re-retrieval for ghosted snippets.
5. Report:
   - ΔS(new question, retrieved)
   - ΔS(new persona vs old snippets)
   - λ states
   - Reset actions taken

Acceptance targets

ΔS(new question, retrieved) ≤ 0.45
Coverage ≥ 0.70 to the new target section
λ remains convergent across three paraphrases
No snippet contamination from old persona or task scope

🔗 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 tension engine and early logic sketch (legacy reference)
⚙️ Engine	WFGY 2.0	Production tension kernel for RAG and agent systems
⚙️ Engine	WFGY 3.0	TXT based Singularity tension engine (131 S class set)
🗺️ Map	Problem Map 1.0	Flagship 16 problem RAG failure taxonomy and fix map
🗺️ Map	Problem Map 2.0	Global Debug Card for RAG and agent pipeline diagnosis
🗺️ Map	Problem Map 3.0	Global AI troubleshooting atlas and failure pattern map
🧰 App	TXT OS	.txt semantic OS with fast bootstrap
🧰 App	Blah Blah Blah	Abstract and paradox Q&A built on TXT OS
🧰 App	Blur Blur Blur	Text to image generation with semantic control
🏡 Onboarding	Starter Village	Guided entry point for new users

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

要我直接幫你寫 state-fork.md 嗎？

5.7 KiB Raw Blame History Unescape Escape