Fun and Philosophy

Reflective, opinionated, and conceptually provocative writing about search — pieces that ask what is search, really? rather than how do we tune the model? Includes metaphors, thought experiments, ethical framings, and challenges to conventional assumptions.


Core Philosophical Questions

What is search for?

Most systems are built as inventory systems: find documents that match the query. Search - Intent Not Inventory argues this is the wrong frame — users search to accomplish goals, not to find documents. This reframes the entire design problem.

Daniel TunkelangSearch: Intent, Not Inventory · blog: queryunderstanding.com

Related: Search Intent, Query Understanding, Conversational and Agentic Search

What does “relevant” even mean?

What is a Relevant Search Result unpacks why “relevant = about the topic” fails in practice. Relevance is task-centered, comparative (relative to alternatives), and dynamic (changes over time). A result is relevant if it helps the user verb the item — accomplish their goal.

Doug TurnbullWhat is a ‘Relevant’ Search Result?

Related: Search Evaluation, Judgment Lists, NDCG

Who does search serve?

Economics of Search sits at the intersection of business rules, user needs, and platform incentives. Margin boosting, promoted listings, and seller support all legitimately affect ranking — but how far is too far?

Related: Search Governance, Results Merchandising, Results Boosting, E-commerce Search

What are we actually measuring?

Mutually Assured Distraction makes a sharp philosophical point: traditional IR metrics (nDCG, MAP, MRR) assume irrelevant documents are neutral. In agentic systems they are harmful. The UDCG metric assigns negative utility to distractors — a fundamentally different theory of what “good retrieval” means.

Lester SolbakkenMutually Assured Distraction · company: hornet.dev

Related: Search Evaluation, Agentic Search, Retrieval Pipeline


Metaphors That Illuminate

Thermodynamics

How Etsy Uses Thermodynamics for Search applies Shannon entropy to result diversity. Low entropy = narrow, converged results. High entropy = diverse, exploratory results. For broad queries like “geeky”, high entropy is the goal. The metaphor reframes diversity as a measurable physical quantity, not a soft editorial preference.

Etsy Engineering — How Etsy Uses Thermodynamics to Help You Search for “Geeky”

Related: Diversity Metrics, Search Result Diversity, MMR, Query Types

The Goldilocks Problem

Searching for Goldilocks frames diversity through the Wundt Curve: too little diversity wastes rank positions on near-duplicates; too much pushes relevant items down. Finding the optimum is NP-hard — MMR is the standard greedy approximation.

Daniel TunkelangSearching for Goldilocks

Related: Diversity Metrics, Search Result Diversity, Thoughts on Search Result Diversity

Mutually Assured Destruction

Mutually Assured Distraction borrows from Cold War deterrence theory: as retrievers and reasoners both improve, the system becomes less reliable — sophistication amplifies distractor-induced failure.

“You cannot reason your way out of bad context, and you cannot compute your way out of distraction.”

Related: Agentic Search, LLM as Judge, Retrieval Pipeline

The Priority Stack

Putting Search Ranking in Perspective argues that ranking is the last lever to pull, not the first. Query understanding → binary relevance → query-independent signals → personalization → query-dependent signals. Most teams invest in ranking too early.

Daniel TunkelangPutting Search Ranking in Perspective

Related: Learning to Rank, Query Understanding, Personalization, Position Bias


Ethical and Social Dimensions

Diversity and Filter Bubbles

Surfacing diverse results is not just a quality metric — it is an editorial choice about what users get to see. Thoughts on Search Result Diversity sits at the intersection of technical trade-offs and value judgments.

Daniel TunkelangThoughts on Search Result Diversity

Related: Diversity Metrics, MMR, Personalization, Search Governance

Rethinking the Whole System

Rethinking Spotify Search shows what happens when a team questions every assumption about what search should do — not just how to tune it.

Spotify Engineering — Rethinking Spotify Search


Key Voices

PersonBlog / SiteThemes
Daniel Tunkelangqueryunderstanding.comRelevance, diversity, intent, evaluation
Doug TurnbullSoftwaredoug blogRelevance definition, BM25, practical search
Lester Solbakkenhornet.devAgentic retrieval failure modes, UDCG