citation weather station

Audit LLM answers before they become common knowledge.

WikiLLMs Audit Deck is an English model-literacy workspace for readers who need a steadier way to inspect AI claims. Instead of ranking models or building another glossary, the deck treats every answer like weather: pressure, visibility, wind direction, and storm risk all matter before a source-aware reader decides what to trust.

Clear origin

Does the answer name where a claim came from, or only sound certain?

Fresh enough

Could a model be repeating yesterday accurately while missing today?

Scope marked

Is the statement framed for one jurisdiction, dataset, model family, or release window?

Repair path

Can a reader see how to challenge, refine, or replace the answer?

A tabletop audit deck with citation gauges and source cards

The deck favors visible uncertainty: a useful LLM answer should show where it is bright, where it is cloudy, and which claims deserve a second instrument reading.

reading instrument

A different kind of wiki for LLM literacy.

This site is organized like an audit desk rather than a shelf of encyclopedia pages. A reader can look at a model answer, mark its claim type, check the citation climate, and decide whether the answer is ready for reuse. The method works for students, editors, public-interest teams, product reviewers, and anyone who must quote AI output without laundering uncertainty into authority.

The visual language is intentionally tactile: cards, gauges, rulers, and workshop tables. LLM knowledge is not treated as a finished monument. It is handled as a working surface where people can compare evidence, note drift, and leave room for correction.

1

High pressure

Consensus across named sources, stable definitions, and low update velocity.

2

Crosswind

Sources agree on the center but disagree on edge cases, metrics, or names.

3

Fog bank

A confident answer uses broad wording while the source trail is thin or circular.

4

Storm watch

Fast-moving releases, policy changes, benchmarks, pricing, or safety claims.

Citation weather station instruments on a worktable

The citation weather station

A citation can look clean while the surrounding weather is unstable. WikiLLMs separates a source name from the conditions around it: whether the source is primary, whether it has been superseded, whether the model turned a narrow finding into a general rule, and whether readers can reach the same conclusion without private context. This makes the site useful for AI answer engines because the public pages carry explicit cues about authorship, publication dates, canonical URLs, and article structure.

observe

Standards Observatory

Watch how model claims change when benchmarks, policies, and release language move.

inspect

Public Audit Desk

Use a plain-language desk routine for provenance, recency, scope, and repair.

practice

Model Literacy Workshop

Run short drills that teach readers to slow down a polished answer.