Research · 1997–2000
Answering queries over data you cannot see.
Ten papers from Stanford on information integration — how to answer a question when the data sits in systems you do not control, cannot fully describe, and are not permitted to query freely. Written in Michael Genesereth’s Logic Group at Stanford, and cited more than 2,500 times.
Recursive Query Plans for Data Integration
Recursive query plans settle three open problems in data integration: maximal answers for arbitrary recursive queries, exploiting functional dependencies, and handling binding-pattern limitations.
Complexity of Answering Queries Using Materialized Views
How the complexity of answering queries using materialized views depends on the view definition language — and on whether views are assumed complete or partial.
Query Planning with Disjunctive Sources
Datalog is not expressive enough for sources holding disjunctive information. Disjunctive datalog with inequality is — and yields plans that extract everything available.
Answering Recursive Queries Using Views
Whether a datalog program can be rewritten to use only materialized views is undecidable — but a maximally contained rewriting always exists, and can be constructed.
Infomaster: An Information Integration System
Infomaster, a deployed information integration system that creates a virtual data warehouse over heterogeneous sources on the internet.
Recursive Plans for Information Gathering
Recursive information gathering plans give maximal answers where conjunctive plans cannot: under functional dependencies, and under binding-pattern restrictions.
Query Optimization Using Local Completeness
Local completeness — a weaker guarantee than completeness — lets an information broker cut the number of sources it has to query.
Query Planning in Infomaster
An overview of the query processing method behind Infomaster — the first system to handle arbitrary positive relational algebra queries and source descriptions.
Infomaster — An Information Integration Tool
The Infomaster system bridges schema and terminology differences between databases, with query planning driven by a specialised model elimination theorem prover.
Query Planning and Optimization in Information Integration
Stanford doctoral dissertation on query planning in information integration: why plans of fixed length are insufficient, and how recursive plans extract every available answer.
Why this still matters
The problem these papers address — answering a question from sources you do not control and cannot fully characterise, then being honest about which answers you can and cannot guarantee — turned out to be a governance problem before it was a database problem. It is the same question a regulated insurer now faces about a model it did not train. That thread runs through the current writing on AI governance.
Citation record: 2,533 citations and an h-index of 10 (Google Scholar, August 2026), across these ten papers and one unpublished Stanford technical report.