Glossary Updates16 new terms added to the glossaries · 12 November 2026, 21:23 CET
AI TechDocKnowledge
  • English (US)
  • English (UK)
  • Deutsch
All context cards

Context card

Judging a curated collection of sources

What makes a curated collection of responses or sources usable as evidence?

By knowledge.aitechdoc.world

Reviewed

Review log and changes

The short answer

A curated collection is a selected set of sources assembled by named people under stated rules, not a census of what exists. It becomes usable as evidence when its selection rules, its editorial standpoint, its checks against silent failure and its change history are published. Item counts alone say nothing about completeness or about the weight of the material.

For: Analysts, technical writers and project managers assessing sources

Key points

  • A curated collection is a view on a debate, selected for a purpose.
  • Selection rules — sources monitored, screening, search terms, access limits — decide what becomes visible.
  • A known-item check distinguishes "nothing new found" from "the search broke".
  • Published corrections and change history make incompleteness inspectable.
  • Item counts are counts of material, not of independent positions.

The context

A collection is a view, not a census

Any curated collection is a view on a debate: elements selected for a purpose and an audience, with the rest left out. That is legitimate and unavoidable; the problem starts when a list of items is read as the full picture. Counts make this worse, because a figure such as "58 articles" counts pieces of material, not independent positions, authors or formal submissions. Several items can restate one argument, and one item can carry four.

Selection rules decide what becomes visible

Usable collections publish how they search: which accounts or feeds are monitored, how often, how candidates are screened, which search terms and perspectives are used, and what is unreachable because of paywalls or platform limits. Where a model does a first relevance pass on titles, that belongs in the method statement too. A disclosed standpoint — for example a deliberate focus on one community — is a different thing from bias in a model's outputs: it is a declared editorial position that readers can correct for, provided it is stated.

Controls that make incompleteness inspectable

Two controls do most of the work. A known-item check requires every sweep to recover items already known to exist; if it misses them, the sweep is treated as broken rather than as proof that nothing new appeared. And corrections are published: when a monitoring list is rebuilt and entries are lost, the gap, the number affected and the retrospective re-search are documented. A public copy in a source code repository adds a change history showing what was added when and what it replaced.

None of this establishes completeness; it makes some forms of incompleteness visible. The final requirement is named accountability: who curates, who edits, who answers for corrections. Readers should also expect the collection to keep item types apart — published arguments, personal accounts, model-generated self-reports and peer-reviewed findings can all be preserved without being treated as interchangeable evidence.

Questions readers ask next

How is this different from a systematic review?
A systematic review follows a pre-registered protocol and aims at reproducible coverage of a defined literature. A curated directory usually aims at timeliness and context, so its value depends on disclosed rules rather than on reproducibility.
Can I cite such a directory in a report?
Yes, if you cite it as a curated secondary source with the date and version you consulted, and cite the primary documents separately for any factual claim.
What should I check first?
The method statement, the date of the last update, and whether corrections are documented. A collection that records its own failures is usually easier to assess than one that looks complete.

Sources

  1. openai/model_spec (repository) — OpenAI on GitHub, 6 October 2026
  2. ISO/IEC 42001 explained — ISO, 6 October 2026

Review log and changes

Every context card is checked against its sources before it is published, and again whenever it changes; the date under the byline is the last review. Corrections (something was wrong) and additions (something was missing) are logged below with date and time (Berlin time). Typos, formatting and link fixes are not listed.

Reviewed

No corrections or additions since publication.