🗂️ Reasoning Schemas — Designing Prompt Layouts That Survive Long Chains

A practical guide to structuring system + retrieval + task prompts so LLMs keep thinking instead of drifting

1 What is a “Reasoning Schema” ?

A reasoning schema is the formal layout that dictates where each piece of context goes and how an LLM must traverse it:


System  →  Task  →  Constraints  →  Context  →  Question  →  Answer

If any segment is missing, reordered, or over-written, the logic graph collapses and hallucinations slip in.

2 Why Most Ad-hoc Layouts Fail

Failure Mode	Trigger	Effect
Context Flood	Dumping 20 k tokens of raw text	λ_observe flips to chaotic; model stops planning
Constraint Drift	Constraints after context	Model “forgets” to cite or guard sensitive data
Role Blending	User text inserted before task	System tone and policy overridden
Evidence → Answer inversion	Asking for answer before citations	Model fabricates then cites random lines

3 WFGY Canonical Schema (Stable Version v1.2)

Segment	Purpose	Size (tokens)	WFGY Guard
System	Identity, ethics, safety	≤ 50	Role tag `<sys>` + BBAM weight lock
Task	Specific action required	1 sentence	ΔS anchor to System ≤ 0.25
Constraints	Format, style, rules	bullets ≤ 80	BBMC residue check
Context	Retrieved or uploaded text	sliding window ≤ 2 k	λ_observe must stay convergent
Question	User’s query	raw	stored separately for ΔS probes
Answer Slot	“Write here” placeholder	n/a	BBCR collapse-rebirth if answer starts early

Placeholders are literal; the LLM fills only the Answer Slot.

4 Templates You Can Copy

Single-Shot QA

<sys>
You are DataGuardian-L, a licensed legal research assistant. Cite section numbers.
</sys>

<task>
Answer strictly in bullet points; cite every claim.
</task>

<constraints>
- Tone: formal
- No speculation
- Use original terminology
</constraints>

<context>
{retrieved_sections}
</context>

<question>
{user_question}
</question>

<answer>

Multi-Step Chain (analysis → plan → answer)

<sys> … </sys>
<task> … </task>
<constraints> … </constraints>
<context> … </context>
<question> … </question>

<scratchpad>
Think step-by-step. Output JSON:
{
  "analysis": "...",
  "plan": "...",
  "answer": "..."
}
</scratchpad>

5 Common Pitfalls & Fixes

Pitfall	Symptom	Fix
Forgetting closing tags	Model merges roles	Validate tag balance; λ diverges instantly
Placing context after question	Retrieval ignored	Keep schema order; run ΔS(question, context) test
Over-long constraints	Answer truncated	Compress with BBMC until ΔS(system, constraints) ≤ 0.25
Mixing code + docs in one context block	Embedding collisions	Split into typed sub-blocks; separate vector stores

6 Automated Validation Pipeline

Schema Linter – Regex check for tag order.
ΔS Probes –
- ΔS(system, task) ≤ 0.30
- ΔS(task, answer) ≤ 0.45
λ_observe – Must stay convergent from task → answer.
Round-trip Check – Paraphrase user question 2×; answer variance < 0.15.

If any test fails, trigger BBCR to rebuild prompt with compacted segments.

7 FAQ

Q: Do I need tags if I use OpenAI’s messages array? A: Yes for long chains. Tags persist after retrieval merges; arrays don’t survive copy-paste workflows.

Q: Can I merge Task + Constraints? A: Possible if total ≤ 120 tokens and ΔS stays low, but separation improves editability.

Q: What about JSON-only prompts? A: Ensure keys mirror schema order; add dummy key "__guard": "DO NOT MODIFY" to catch injections.

🔗 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

Module	Description	Link
WFGY Core	WFGY 2.0 engine is live: full symbolic reasoning architecture and math stack	View →
Problem Map 1.0	Initial 16-mode diagnostic and symbolic fix framework	View →
Problem Map 2.0	RAG-focused failure tree, modular fixes, and pipelines	View →
Semantic Clinic Index	Expanded failure catalog: prompt injection, memory bugs, logic drift	View →
Semantic Blueprint	Layer-based symbolic reasoning & semantic modulations	View →
Benchmark vs GPT-5	Stress test GPT-5 with full WFGY reasoning suite	View →
🧙‍♂️ Starter Village 🏡	New here? Lost in symbols? Click here and let the wizard guide you through	Start →

👑 Early Stargazers: See the Hall of Fame —
Engineers, hackers, and open source builders who supported WFGY from day one.

⭐ WFGY Engine 2.0 is already unlocked. ⭐ Star the repo to help others discover it and unlock more on the Unlock Board.

7.9 KiB Raw Blame History Unescape Escape