Evidence
LinkLore's claims aren't just marketing copy — they're backed by a peer-reviewable paper with a published DOI and real dogfooding data from projects that have used LinkLore in production for months.
Academic validation
LinkLore is the subject of an independent academic paper studying how AI agents use structured memory tools — including LinkLore itself as a case study. The paper's quantitative claims are backed by a public replication package: raw classification data, coding rules, and the scripts that reproduce every statistic.
"When the Agent Is the User: Agent-Driven Evolution of AI-Native Tools"
View paper & replication data on Zenodo ↗Dogfooding at scale
LinkLore has been used as the primary memory system across dozens of real projects — not a lab demo. These numbers come from the same corpus analyzed in the paper above.
Methodology
Every lore/doc entry in the corpus was classified by two independent AI raters under a written rule set, without knowing the paper's hypotheses. Agreement was measured with Cohen's kappa, and every disagreement was adjudicated with a recorded rationale — not silently resolved.
The 10.3% headline rate mixes LinkLore's own development (where 23.4% of entries are tool feedback) with fourteen unrelated projects (3.2%) — the conservative floor reported above is that lower, out-of-context number, not the inflated one.
The corpus is frozen and hashed: every entry's label, date, and a SHA-256 hash of its body are published, so any later disclosure can be checked against this freeze. Private business content is masked, but no quantitative claim in the paper depends on a masked field.
Read the full methodology & data ↗