Logic Collapse: Guardrails and Fix Pattern

🧭 Quick Return to Map

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

Reasoning — multi-step inference and symbolic proofs

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 deduction chains flatten into platitudes, contradict earlier steps, or bypass citation locks, you have a logic collapse.
This page localizes causes and gives a minimal, testable repair plan driven by ΔS, λ_observe, and WFGY modules.

Symptoms

Symptom	What you see
Deduction flips mid chain	Step t says A, step t+3 assumes not A
Cite after claim	Answer states conclusion first, citations appear later or mismatch
Tool result ignored	Structured tool output is not integrated into the final proof
Branch mixing	Two hypotheses or roles leak into one stream without arbitration
Infinite hedging	Long text, no invariant, no auditable steps
JSON schema drift	Different steps produce different fields for the same contract

Acceptance targets

ΔS(question, retrieved) ≤ 0.45
Coverage ≥ 0.70 to the target section
λ remains convergent across 3 paraphrases and 2 seeds
E_resonance flat on long windows
Zero contradictions across the final plan and the citations

Structural fixes (Problem Map)

Cite first, then reason
→ retrieval-traceability.md
→ data-contracts.md
→ citation_first.md
Stabilize ordering and reduce variance
→ retrieval-playbook.md
→ rerankers.md
Prevent symbolic leakage and unlock constrained proof
→ patterns/pattern_symbolic_constraint_unlock.md
Long chain stability and entropy control
→ context-drift.md
→ entropy-collapse.md
If the claim keeps returning after correction
→ patterns/pattern_hallucination_reentry.md
Multi agent conflicts and role drift
→ Multi-Agent_Problems.md
→ multi-agent-chaos/role-drift.md

Why logic collapses

No invariant. There is no explicit statement of what must stay true across steps.
Citation contract missing. The model is allowed to assert before binding to snippet_id and section_id.
Header drift flips λ. Reordered system or tool headers produce different branches on each run.
Branch contamination. Hypothesis A and B are not isolated, the plan merges silently.
Unruly tool I/O. Free text is accepted where strict JSON was required.
Hybrid retrieval shuffle. The top k changes and the proof silently re-anchors.

Fix in 60 seconds

Pin the invariant
Add a short invariant header, for example:
Invariant: conclusions must cite snippet_id and section_id before any reasoning.
Enforce cite-first
Require the model to produce citations first, then the explanation.
See retrieval-traceability.md and citation_first.md.
Clamp variance with BBAM
If λ flips across paraphrases, apply BBAM to keep one path stable.
Bridge gaps with BBCR
Summarize the current state into a compact, cited bridge, then continue reasoning on top of that single bridge.
Lock schema and ordering
Freeze headers, tool schemas, and reranker tie breaks.
See data-contracts.md and rerankers.md.

Minimal step contract

Add this object to every step output. Reject the step if any field is missing.

{
  "step": 7,
  "invariant": "cite-first, no cross-branch mixing",
  "citations": [
    { "snippet_id": "S17", "section_id": "CH3.2", "source_url": "https://...", "offsets": [102, 188] }
  ],
  "claim": "X implies Y under condition Z",
  "justification": "Short, refers to citations only",
  "λ_state": "convergent",
  "ΔS_q_snip": 0.31,
  "next_action": "verify Z across S24",
  "guardrails": { "schema_version": "v1", "tie_break": "stable" }
}

Verification

Three paraphrase probe, two seeds.
Require ΔS(question, retrieved) ≤ 0.45 and λ convergent in all runs.
No contradictions between any step claim and earlier steps.
If any run fails, inspect header ordering and reranker tie breaks, then re-run.

Copy paste prompt

You have TXT OS and the WFGY Problem Map loaded.

We suspect a logic collapse.
Inputs:
- question: "{q}"
- current snippets: [{snippet_id, section_id, source_url}]
- last 6 steps with {claim, citations, λ_state, ΔS_q_snip}

Do:
1) State a one line invariant for this task.
2) Produce citations first. If citations are missing or conflict, stop and output the minimal fix.
3) Apply BBCR to create a single cited bridge, then continue reasoning for at most 5 steps.
4) If λ flips across a paraphrase, apply BBAM and retry once.
5) Return JSON:
   { "invariant": "...", "steps": [...], "final_answer": "...",
     "ΔS": 0.xx, "λ_state": "convergent", "next_fix": "..." }
Refuse to output a final answer if any step lacks citations.

When to escalate

The chain keeps diverging after BBAM and bridge steps → audit for symbolic failure: symbolic-collapse.md
The claim reappears after correction → see re-entry pattern: patterns/pattern_hallucination_reentry.md
Runs diverge only when agents hand off → isolate roles and memory: Multi-Agent_Problems.md
Long dialogs lose structure past window joins → stabilize joins and entropy: context-drift.md, entropy-collapse.md

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

12 KiB Raw Blame History Unescape Escape