11 KiB
Proof Dead Ends: 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 a reasoning chain tries to prove a claim but never closes the loop.
You see endless subgoals, repeated restatements, or a jump to a conclusion without obligations being discharged.
This page localizes proof stalls 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 -
Logic and symbols
→ logic-collapse.md
→ symbolic-collapse.md -
Infinite recursion and overload
→ recursive-loop.md
→ entropy-overload.md -
If the chain shifts into meta or paradox
→ philosophical-recursion.md -
Traceability and contracts
→ retrieval-traceability.md
→ data-contracts.md
Symptoms
| Symptom | What you see |
|---|---|
| Goal never shrinks | Subgoals keep being restated with new labels but same content |
| Lemma spiral | New lemmas depend on each other in a cycle without a base |
| Premise drop | A needed premise disappears after paraphrase or tool call |
| Unproven jump | Model asserts Q since P but never shows P or its bridge |
| Invariant drift | Claimed invariant changes wording or unit across steps |
| Citation vacuum | Steps claim support yet no snippet or rule is referenced |
Why dead ends happen
- No proof obligations. Steps do not state what must be shown to advance.
- Missing bridge. The link from premise to subgoal is never written.
- Drifting invariants. The property that should stay fixed keeps mutating.
- Symbol table absent. Names rebind silently so obligations mismatch.
- Plan length without checkpoints. Long chains lose ΔS and λ stability.
- Retrieval anchor unstable. The cited snippet shuffles, the goal resets.
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 across step joins
- All obligations discharged or explicitly marked unsatisfied with a cited reason
Fix in 60 seconds
-
Declare the goal and obligations
Writegoal,known,need_to_show, and the currentinvariants.
If any field is empty, stop. Fetch or restate before continuing. -
Add a BBCR bridge
Produce a short cited bridge from current premises to the next subgoal.
Reject continuation if the bridge lacks citation or rule tag. -
Clamp variance with BBAM
If λ flips on paraphrase, freeze the symbol table and the invariant set.
Re-run the step with the same bindings. -
Apply BBPF when stuck
Branch to two small proof tactics, attempt each for N steps, then backtrack.
Keep the best branch where ΔS drops and obligations shrink.
Minimal proof contract
Require every step to carry this schema. Refuse steps that miss fields.
{
"step_id": "S7",
"goal": "Prove claim C",
"known": ["P1", "P2"],
"need_to_show": ["L1 -> C"],
"bridge": {
"rule": "modus_ponens | contradiction | induction | algebra",
"citations": ["S12#CH2.3", "S09#APP.A"],
"ΔS_bridge": 0.33
},
"invariants": [
{"name": "unit_price_nonnegative", "scope": "calc", "status": "holds"}
],
"symbols": [
{"name": "x", "kind": "var", "namespace": "calc", "unit": "USD/kg"}
],
"λ_state": "convergent"
}
Structural repairs
-
If anchors or citations are missing → retrieval-traceability.md and → data-contracts.md
-
If long chains decay → entropy-overload.md and → recursive-loop.md
-
If symbol or unit meanings drift → symbolic-collapse.md
-
If logic itself collapses → logic-collapse.md
-
If meaning vs similarity conflict → embedding-vs-semantic.md
Verification
- Run three paraphrases and two seeds. The set
{goal, invariants, symbols}must remain identical. - All
need_to_showitems become empty or a single clearly blocked item with a cited missing premise. - ΔS(question, retrieved) ≤ 0.45 and coverage ≥ 0.70 in every run.
- Logs show at least one BBCR bridge with citations and a monotone decrease in open obligations.
Copy-paste prompt
You have TXT OS and the WFGY Problem Map loaded.
We suspect a proof dead end.
Inputs:
- question: "{q}"
- snippets: [{snippet_id, section_id, source_url}]
- current plan trace: [{step_id, text}]
- last symbol table and invariants (if any)
Do:
1) Build the proof contract with {goal, known, need_to_show, invariants, symbols}.
2) Create a BBCR bridge with a named rule and citations. If missing, refuse and return the exact missing item.
3) If λ flips across paraphrase, apply BBAM and retry with the frozen table.
4) If still stuck, branch two BBPF tactics for at most N=3 steps each, keep the branch with lower ΔS and fewer obligations.
5) Return JSON:
{
"plan": [...steps...],
"open_obligations": [],
"ΔS": 0.xx,
"λ_state": "convergent",
"verdict": "proved | blocked_missing_premise | not_provable_from_given"
}
Refuse to assert the final claim unless all obligations are discharged or the verdict is explicitly "not_provable_from_given" with a cited reason.
Common gotchas
- Rule without scope. A named rule is used but no citation or domain. Add both or reject.
- Induction without base. The inductive step is written yet the base case is missing.
- Contradiction shortcut. The chain says “assume not C” then jumps to C with no witness.
- Bridge drift. The bridge cites new snippets each run. Lock the query and add a reranker if anchors shuffle.
When to escalate
-
The same wrong assertion returns after a correction → pattern_hallucination_reentry.md
-
Multi-agent handoff reopens closed obligations → Multi-Agent_Problems.md and → multi-agent-chaos/role-drift.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
| 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.