WFGY/ProblemMap/GlobalFixMap/Retrieval/checklists/reranker_sanity.md

Reranker Sanity Checklist

Scope: validate a hybrid pipeline where BM25 and ANN feed a deterministic reranker.

---

Pre checks

• Persist BM25 and ANN candidate lists before reranking.
• Fix a random seed and record model version for the reranker.
• Store per-candidate features needed by the reranker.
• Ensure identical tokenization between HyDE prompt and server.

Refs:
Rerankers · Query parsing split

---

Baseline health

• BM25 top k recall on the gold set ≥ 0.60 at k = 50.
• ANN top k recall on the gold set ≥ 0.60 at k = 50.
• Overlap Jaccard(BM25, ANN) logged for k = 50 to detect split.

If overlap < 0.20 and both recalls are good, a query split is likely.
Route using the split detector or run two paths then merge.

---

Reranker behavior

• Uplift over the better baseline ≥ 5 percent recall at k = 10.
• Ranking stable across 3 paraphrases. Kendall tau ≥ 0.60.
• Tie breaking rule fixed and documented.
• Deterministic output given the same candidates and seed.

---

Acceptance targets

• Coverage ≥ 0.70 on the gold set after reranking at k = 10 or 20.
• ΔS(question, top1) ≤ 0.45 for at least 70 percent of queries.
• λ_observe convergent across 3 paraphrases and 2 seeds.

---

Debug sequence

1. Log both candidate lists and measure overlap.
2. If split detected, enable the split route or increase k per branch.
3. Try a simpler deterministic reranker first.
4. Re-measure uplift and stability.

Refs:
Retrieval playbook · ΔS probes

---

🔗 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

Module	Description	Link
WFGY Core	Canonical framework entry point	View
Problem Map	Diagnostic map and navigation hub	View
Tension Universe Experiments	MVP experiment field	View
Recognition	Where WFGY is referenced or adopted	View
AI Guide	Anti-hallucination reading protocol for tools	View

If this repository helps, starring it improves discovery for other builders.