Symbolic Collapse: Guardrails and Fix Pattern

When symbols, roles, units, or variables drift in meaning across steps, the chain collapses even if individual sentences look fluent.
This page localizes symbolic failures and gives a minimal, testable repair plan using ΔS, λ_observe, and E_resonance.

Open these first

Visual map and recovery
→ rag-architecture-and-recovery.md
Cite first and make it auditable
→ retrieval-traceability.md
→ data-contracts.md
→ citation_first.md
Lock constraints and unlock safely
→ pattern_symbolic_constraint_unlock.md
If logic also drifts
→ logic-collapse.md

Symptoms

Symptom	What you see
Variable alias collision	`x` means “price” at step 3 and “index” at step 7
Unit flip	km vs miles, USD vs EUR, or 0-1 vs 0-100 score scales without notice
Role leak	Tool output fields are reused as different roles in later steps
Type drift	A list becomes a dict halfway, downstream steps still “pass”
Anchor rename	Entity “A Inc.” becomes “Alpha” with no traceable mapping
Quantifier slip	“Some” turns into “all” when summarizing two steps later
Schema split	Same object has different required fields across steps

Why symbolic collapse happens

No symbol table. Meanings live only in prose and mutate under paraphrase pressure.
Missing namespace. Agent or tool outputs write into global scope.
Unit contract absent. The pipeline accepts values without unit or scale tags.
Constraint unlock without fences. The model invents aliases to escape constraints.
Header drift flips λ. Reordered headers produce distinct symbol bindings per run.
Hybrid retrieval shuffle. The anchor snippet changes and symbols rebind silently.

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 at joins and handoffs
Zero symbol drift across steps when checked against a symbol table

Fix in 60 seconds

Create a symbol table
Add an explicit table with name, kind (var, role, unit, entity), namespace, definition, source_snippet, allowed_values.
Enforce cite-first and schema-locked steps
Require citations before any symbol is used. Enforce unambiguous fields.
See retrieval-traceability.md and data-contracts.md.
Clamp variance with BBAM
If λ flips after paraphrase, apply BBAM to hold one binding set stable.
Bridge with BBCR
Produce a short, cited bridge that restates the symbol table and the current state, then continue reasoning on top of that bridge only.
Lock namespaces
Prefix every symbol with a scope agent.role.symbol or tool.name.field. Reject writes outside the declared namespace.

Minimal symbol table contract

Every plan step must carry the table and refuse execution on mismatch.

{
  "symbols": [
    {
      "name": "x",
      "kind": "var",
      "namespace": "calc.pricing",
      "definition": "unit price per kg",
      "unit": "USD/kg",
      "source_snippet": "S12#CH2.3",
      "allowed_values": "real >= 0"
    },
    {
      "name": "R",
      "kind": "role",
      "namespace": "agent.verifier",
      "definition": "checks unit and citation before approval",
      "allowed_values": ["approve","reject","fix"]
    }
  ],
  "schema_version": "v1"
}

Reject the step if any field changes without a cited reason.

Structural repairs

Unit and scale discipline → data-contracts.md → retrieval-traceability.md
Ordering stability → retrieval-playbook.md → rerankers.md
Constraint unlock with fences → pattern_symbolic_constraint_unlock.md → prompt-injection.md
Long window stability → context-drift.md → entropy-collapse.md
If meaning vs similarity conflict → embedding-vs-semantic.md
If logic also fails → logic-collapse.md

Verification

Three paraphrases, two seeds. All runs must keep the same symbol table.
ΔS(question, retrieved) ≤ 0.45 and coverage ≥ 0.70 in every run.
No unit or namespace change without a cited justification that points to a specific snippet.
Report a diff of the symbol table between steps; the diff must be empty or fully justified.

Copy-paste prompt

You have TXT OS and the WFGY Problem Map loaded.

We suspect a symbolic collapse.

Inputs:
- question: "{q}"
- current snippets: [{snippet_id, section_id, source_url}]
- last symbol table (if any)
- last steps with {claim, citations, λ_state, ΔS}

Do:
1) Build a symbol table with {name, kind, namespace, definition, unit, source_snippet, allowed_values}.
2) Cite first, then restate the claim using only table symbols.
3) If λ flips across a paraphrase, apply BBAM. If content diverges, produce a BBCR bridge that freezes the table.
4) Output JSON:
   { "symbols": [...], "steps": [...], "final_answer": "...",
     "ΔS": 0.xx, "λ_state": "convergent", "table_diff": [] }
Refuse the final answer if any step uses a symbol that is not in the table.

Common gotchas

Alias creep. The model introduces “alpha” for A without binding it to the table. Reject and force a mapping row.
Silent unit conversion. Numbers change scale between steps. Require unit and scale fields.
Cross-agent overwrite. Handoffs write to shared names. Use strict namespaces and a write fence.
Hybrid retrieval reorder. Top-k changes on rerun. Lock query and tie breaks, or add a reranker.

When to escalate

Table stays unstable after BBAM and a bridge → audit logic and role discipline: logic-collapse.md and → multi-agent role isolation: Multi-Agent_Problems.md and role-drift.md
The same wrong claim returns after correction → pattern_hallucination_reentry.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