Hybrid Search Engineering — Part 2: Filtering, Evaluation, Latency Budgets & Operations — 42 Templates & Checklists for Filtering & Structured Search, Per-Class Evaluation, Latency Budgets, the Reference Architecture & Operating the Hybrid

## The ranking was perfect. The facet count just leaked a sealed matter.

No document body leaked. The access-control trim ran on the final results, exactly as designed. But a title had surfaced in an autocomplete suggestion — computed from an untrimmed candidate cache — and a facet count on the same page announced exactly how many privileged documents matched a query inside a matter the searcher had no right to know existed. **Hybrid Search Engineering — Part 2** is the working second half of the discipline for systems that already fuse: 42 copy-paste templates and checklists distilled from Chapters 7–12 plus Appendices A–B of the dedicated hybrid-search book — promote filters to a retrieval arm inside the eligible universe, engineer the filtering layer where facets, permissions, and pushdown share one machinery, prove quality per class in a court that can block a release, budget latency at the tail, assemble the system under contracts, and operate it against drift.

Every template carries the book's tested positions built in: eligibility applied before ranking, never subtracted after it; every structured signal classified gate, evidence, or presentation; selectivity routing in the book's three bands (brute force under roughly 2%, in-search filtering to roughly 40%, oversampled post-filtering above it) with the k/s oversampling arithmetic shown; facet counts computed index-side over the eligible universe with other-filters semantics; a golden set sampled from logs, graded 0–3, pooled, and agreement-audited; a five-number metric panel reported per class, never only in aggregate; a stage ledger with the book's own per-stage ranges; a failure-semantics table in which eligibility failure means refuse; and an operating calendar — weekly, monthly, quarterly, per-change — that keeps every contract alive. Where the book gives an example — its 250 ms illustrative budget, its facet display counts, its sentinel floors — the template labels it as the book's example for you to tune. Bracketed fields are yours to fill; nothing here invents a threshold for your system.

### What's inside — 42 templates & checklists in 7 sections

**1. Filters Are Retrieval — Structured Search & the Eligible Universe (6)** — Eligible-Universe Principle Worksheet · Gate / Evidence / Presentation Classification Sheet · Metadata Schema Audit · Postgres Single-Statement Hybrid Specification · Substrate Decision Sheet · Structured-Arm Failure Audit

**2. Facets, Permissions & Pushdown — The Filtering Layer (6)** — Three-Places-to-Filter Physics Sheet · Selectivity Routing Worksheet · Facet Computation Specification · Access-Control-as-Retrieval Checklist · Disclosure-Channel Red-Team Battery · Deep Paging, Export & Suggestion Layer Sheet

**3. Prove It Per Class — Judgments, Metrics & Ablations (6)** — Golden Set Construction Protocol · Judging Guidelines & Agreement Audit · Metric Panel Definition Sheet · Ablation Battery Record · Significance & Pre-Registration Sheet · Online Validation & Regression Gate Sheet

**4. Latency Budgets — 250 Milliseconds, Budgeted at the Tail (6)** — Latency Budget Worksheet · Stage Ledger · Fan-Out, Deadlines & Straggler Policy Sheet · Four-Layer Cache Plan · Degradation Ladder · Monday Load Test, Capacity & Tail Forensics Sheet

**5. The Reference Architecture — One System, Three Engines (6)** — Three-Planes Architecture Specification · Indexing Plane Contract · Query Plane Assembly & Provenance Record · Failure-Semantics Table · Tenancy Partition Decision Sheet · Observability, Control-Plane Memory, Vendor & Recovery Sheet

**6. Run It Like You Mean It — Operations, Drift & the Honest Exits (6)** — Five-Drift Detector Register · Failure Atlas Incident Index · Maturity Model Self-Assessment · New-Species Assessment Sheet · Honest-Exit Review Sheet · Operating Calendar

**7. Part 2 Gates & Working Artefacts (6)** — Appendix A Production Readiness Checklist — the book's 16-line pre-launch and quarterly gate · Appendix B Quick Reference Card — the book's 10 rules of thumb on one page · Failure-Analysis Round Docket · Drift Sentinel Worksheet · Contract Ownership Roster & Annual Review Sheet · Part 2 Production Gate — Chapters 7–12 Sign-Off — the book's six chapter checklists consolidated into one evidence-backed gate (47 lines)

### Who this is for

Search and platform engineers running a hybrid system in production, relevance leads who own its quality, and architects deciding where the hybrid should live and how its tenants stay separate. The source book assumes you already run — or are about to run — more than one retriever; Part 1 of this series covers the arms themselves. If your constrained queries return half-empty pages, your facet counts describe a corpus your users cannot reach, or your headline quality number has never been sliced by query class, start at Template 1.

### What this pack is not

- **Not Part 1.** BM25 and the keyword arm, ANN recall contracts, rank fusion, query understanding, and reranking cascades are the book's Chapters 1–6. They live in the Part 1 companion pack; nothing here restates them, and this pack is complete and useful without them. (One flagged exception: Template 38 reproduces the book's Appendix B quick-reference card, which itself restates four Part 1 rules of thumb — carried because Appendix B is this pack's assigned source, and the template says so.)

- **No runnable code or tooling.** The book's code miniatures are simplified teaching shapes. The pack ships the templates that specify, audit, and operate these components — not a working search system, evaluation harness, or drift sentinel.

- **No vendor setup guides.** Postgres/pgvector, OpenSearch, Elasticsearch, and vector stores are treated as substrates with the book's stated methods and ceilings. No product is installed, configured step-by-step, or benchmarked here.

- **No compliance crosswalk or certification.** Nothing in these chapters maps hybrid search to a regulatory framework, and the Part 2 gate in Template 42 is a Chapters 7–12 production gate — not a compliance certification.

- **No guarantee of quality, latency, or security outcomes.** The book's 250 ms figure is an illustrative budget, labeled as such; eligibility enforcement is a property to engineer and red-team, not a promise. These templates manage, measure, and bound the system — they do not certify it.

- **Not the RAG or vector-database packs.** The RAG Architecture pack (Product #2) and the Vector Databases pack (Product #9) come from different books and treat hybrid search from their own angles. This pack is the dedicated hybrid-search book's second half, at book depth; the packs do not restate each other.

### Companion products

**Hybrid Search Engineering — Part 1: BM25, ANN Recall & Rank Fusion Templates** is this pack's sibling, built from Chapters 1–6 of the same book: a companion for the arms, fusion, and query layer this pack deliberately does not restate — never a requirement for this pack to work. This pack stands alone.

*Note: the $99 Architecture Templates Bundle currently contains Products #1–#3 only. This pack launches standalone; no bundle inclusion is promised for this product.*

Built by the author of **Hybrid Search Engineering: Vector + Keyword + Structured Retrieval, Fused Properly**, from the production discipline behind enterprise hybrid search systems.

### About the author

Naveen Kumar Josyula is the author of 35 AI books on Amazon KDP and the CEO and founder of AISoltrics Technologies Inc. He is the author of the source book, *Hybrid Search Engineering: Vector + Keyword + Structured Retrieval, Fused Properly*, from which every template in this pack is distilled.

### Support promise

If anything in this file is wrong or unclear, email support@aisoltrics.com before you judge it — we fix verified issues within one business day and send you the corrected file free. (You can also reach the founder's office at aadhya@aisoltrics.com.) Updates to this product are free to past buyers via your library. Sold by AISoltrics Technologies Inc.