Skip to content

goldset_curation.md

Latest commit

 

History

History
149 lines (104 loc) · 6.31 KB

goldset_curation.md

File metadata and controls

149 lines (104 loc) · 6.31 KB

Goldset Curation — Guardrails and Fix Patterns

🧭 Quick Return to Map

You are in a sub-page of Eval.
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.

A curated gold set is the foundation for evaluation stability. Without strict contracts on the gold data, all eval metrics become meaningless. This page defines how to build, audit, and maintain gold QA sets that align with WFGY acceptance targets.

Open these first

Acceptance targets for curated goldsets

  • Coverage ≥ 0.80 of target sections
  • ΔS(question, gold anchor) ≤ 0.35
  • λ state stable across 3 paraphrases and 2 seeds
  • No overlap: each gold item maps to exactly one snippet and section

Curation process

1. Select domains

  • Identify domains relevant to the pipeline (finance, law, product docs).
  • Ensure gold questions are drawn from actual user tasks.

2. Define anchors

  • Each QA item must cite a section ID and expected_doc.
  • Anchors must reference stable problem map sections, not ephemeral text.

Example:

{
  "id": "Q_0007",
  "question": "What causes hallucination re-entry after correction?",
  "answer_ref": "PM:patterns/pattern_hallucination_reentry",
  "expected_doc": "ProblemMap/patterns/pattern_hallucination_reentry.md",
  "section_id": "hallucination-reentry"
}

3. Add paraphrases

  • Minimum 3 per question.
  • Probe λ stability under phrasing variance.
{
  "id": "Q_0007_P1",
  "question": "Why do hallucinations return after being corrected once?"
}

4. Validate citations

  • Each gold item must include an exact citation offset.
  • If offsets drift, the goldset is invalid until refreshed.

5. Apply regression gate

  • No gold item should produce ΔS > 0.45 in baseline runs.
  • Violations are logged and flagged for refresh.

Common pitfalls and fixes

  • Gold overlaps across sections → Fix: merge or re-scope questions, ensure one-to-one mapping.

  • Anchors point to unstable docs → Fix: only link to long-lived WFGY ProblemMap pages.

  • Paraphrases flip λ → Fix: clamp with BBAM variance controls and revalidate.

  • Coverage below 0.80 → Fix: expand questions until goldset covers every critical node.

Quick workflow

  1. Draft 20–30 candidate QA items.
  2. Add 3 paraphrases each.
  3. Link every item to an anchor section.
  4. Run through eval_harness.md.
  5. Drop items that fail regression gate.
  6. Store final goldset in datasets/gold/.

🔗 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 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.
GitHub Repo stars