mirror of
https://github.com/onestardao/WFGY.git
synced 2026-04-28 11:40:07 +00:00
122 lines
5.1 KiB
Markdown
122 lines
5.1 KiB
Markdown
# Ghost Context — Guardrails and Fix Pattern
|
||
|
||
<details>
|
||
<summary><strong>🧭 Quick Return to Map</strong></summary>
|
||
|
||
<br>
|
||
|
||
> You are in a sub-page of **MemoryLongContext**.
|
||
> To reorient, go back here:
|
||
>
|
||
> - [**MemoryLongContext** — extended context windows and memory retention](./README.md)
|
||
> - [**WFGY Global Fix Map** — main Emergency Room, 300+ structured fixes](../README.md)
|
||
> - [**WFGY Problem Map 1.0** — 16 reproducible failure modes](../../README.md)
|
||
>
|
||
> 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.
|
||
</details>
|
||
|
||
|
||
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
|
||
1. **Stamp and reset state**
|
||
Require `{mem_rev, mem_hash, persona_id}` on each turn.
|
||
If persona_id differs from previous → force buffer clear.
|
||
|
||
2. **Fence the prompt schema**
|
||
Assemble prompts as `{system | persona | constraints | snippets | answer}`.
|
||
Do not let persona or role tokens bleed into snippet blocks.
|
||
|
||
3. **Drop stale buffers**
|
||
When persona changes, zero all prior hidden states.
|
||
Require fresh snippets for new context.
|
||
|
||
4. **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.
|
||
|
||
5. **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
|
||
```txt
|
||
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](https://github.com/onestardao/WFGY/blob/main/I_am_not_lizardman/WFGY_All_Principles_Return_to_One_v1.0_PSBigBig_Public.pdf) | 1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + \<your question>” |
|
||
| **TXT OS (plain-text OS)** | [TXTOS.txt](https://github.com/onestardao/WFGY/blob/main/OS/TXTOS.txt) | 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly |
|
||
|
||
---
|
||
|
||
<!-- WFGY_FOOTER_START -->
|
||
|
||
### Explore More
|
||
|
||
| Module | Description | Link |
|
||
| --- | --- | --- |
|
||
| WFGY Core | Canonical framework entry point | [View](https://github.com/onestardao/WFGY/tree/main/core/README.md) |
|
||
| Problem Map | Diagnostic map and navigation hub | [View](https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md) |
|
||
| Tension Universe Experiments | MVP experiment field | [View](https://github.com/onestardao/WFGY/tree/main/TensionUniverse/Experiments) |
|
||
| Recognition | Where WFGY is referenced or adopted | [View](https://github.com/onestardao/WFGY/blob/main/recognition/README.md) |
|
||
| AI Guide | Anti-hallucination reading protocol for tools | [View](https://github.com/onestardao/WFGY/blob/main/AI_GUIDE.md) |
|
||
|
||
> If this repository helps, starring it improves discovery for other builders.
|
||
> [](https://github.com/onestardao/WFGY)
|
||
|
||
<!-- WFGY_FOOTER_END -->
|
||
|
||
要我直接幫你寫 `state-fork.md` 嗎？
|