vrr/WFGY
2
0
Fork 0
mirror of https://github.com/onestardao/WFGY.git synced 2026-04-28 19:50:17 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4
WFGY/ProblemMap/GlobalFixMap/Safety_PromptIntegrity/citation_first.md

6.6 KiB
Raw Blame History

Citation-First Prompting — Guardrails and Fix Patterns

🧭 Quick Return to Map

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

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.

Citation drift is one of the most common ways LLMs lose trust.
Without cite-then-explain discipline, answers may sound fluent but detach from sources.
This page locks the workflow: every reasoning step begins from citations, not narrative.

When to open this page

  • Model produces fluent paragraphs with zero citations.
  • Citations appear, but after-the-fact and unverifiable.
  • References change between runs, even with same inputs.
  • ΔS(question, retrieved) ≤ 0.45 but λ diverges when narrative precedes citation.
  • Users complain: "Where did this answer come from?"

Open these first

Core acceptance

  • Every answer starts with citations (no narrative before refs).
  • Coverage ≥ 0.70 of the target section.
  • ΔS(question, cited snippet) ≤ 0.45.
  • λ convergent across 3 paraphrases.
  • No hallucinated citations (must resolve to a retriever record).

Fix in 60 seconds

  1. Citation-first discipline

    • Always start with snippet_id, section_id, source_url.

  2. Enforce schema

    • Required fields: 
      { "snippet_id": "...", "section_id": "...", "source_url": "...", "offsets": [..], "tokens": N }

  3. Reason only after citation

    • Explain or analyze after citation block.

  4. Reject broken runs

    • If citation missing → abort answer, return error tip.

  5. Stability probe

    • Run 3 paraphrases. If λ diverges, lock citation schema, rerun.

Typical failure vectors → fix

Vector Symptom Fix
Narrative-first Text precedes refs, unstable λ Force cite-then-explain ordering
Fake refs Hallucinated URLs Schema lock + retrieval-traceability.md
Drifting refs Different citations each run Clamp λ with BBAM, validate ΔS ≤ 0.45
Silent fallback Model drops refs under safety refusal Apply SCU (symbolic constraint unlock)

Probe prompt

You must output citations before narrative.
Schema: {snippet_id, section_id, source_url, offsets, tokens}

Rules:
1. Cite first. Explain only after citations are shown.
2. No answer if citations missing.
3. Log ΔS(question, cited snippet). Reject if ≥ 0.60.
4. λ must stay convergent across 3 paraphrases.

🔗 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

Layer Page What it’s for
Proof WFGY Recognition Map External citations, integrations, and ecosystem proof
Engine WFGY 1.0 Original PDF based tension engine
Engine WFGY 2.0 Production tension kernel and math engine for RAG and agents
Engine WFGY 3.0 TXT based Singularity tension engine, 131 S class set
Map Problem Map 1.0 Flagship 16 problem RAG failure checklist and fix map
Map Problem Map 2.0 RAG focused recovery pipeline
Map Problem Map 3.0 Global Debug Card, image as a debug protocol layer
Map Semantic Clinic Symptom to family to exact fix
Map Grandma’s Clinic Plain language stories mapped to Problem Map 1.0
Onboarding Starter Village Guided tour for newcomers
App TXT OS TXT semantic OS, fast boot
App Blah Blah Blah Abstract and paradox Q and A built on TXT OS
App Blur Blur Blur Text to image with semantic control
App Blow Blow Blow Reasoning game engine and memory demo

If this repository helped, starring it improves discovery so more builders can find the docs and tools. GitHub Repo stars