Optionalage_v3.10 — whole days since the note's current on-disk mtime (freshness
signal for forgetting-aware recall; never negative). Computed by statting
the final hit paths, so it reflects the live file mtime, not the possibly-
lagging indexed mtime in FTS5/embed-db source_state. Omitted (along with
stale) only when the stat fails — e.g. the file was deleted between
fusion and response assembly (fail-soft, never throws).
Optionalchunk_v2.8.0 — content-source kind. Lets agents distinguish markdown notes from PDF chunks when both are indexed. Defaults to "md" for backward compatibility (legacy DBs and TF-IDF hits have no kind metadata).
Optionalline_Optionalline_Vault-relative path of the matching note (or path#chunk for granularity: "block").
Per-signal observability — which signals contributed at what rank/score.
Optionalreranker_v2.9.0 — cross-encoder reranker score in [0, 1] (sigmoid of the model's
relevance logit). Present only when the server was started with
--enable-reranker AND this hit was within the reranker's top-N
candidate set (default 50). Higher = more relevant. Compare across
results within the same response, NOT across queries (the absolute
value depends on the query).
Fused RRF score (sum of 1/(k+rank) terms across signals).
Snippet from whichever signal produced the best chunk hit.
Optionalstalev3.10 — true when age_days >= DEFAULT_STALE_DAYS (365). Lets an agent
flag a recalled fact as potentially out-of-date instead of presenting it
as current (the Memora stale-memory-reuse frontier). Read-only signal —
does NOT reorder results (opt-in recency re-ranking is a separate flag).
Stripped basename for display (.md or .pdf removed per kind).
One row of the fused searchHybrid response.
Exposes
per_signalfor full observability — agents can see which retrieval signal contributed (BM25 / TF-IDF / embeddings) and at what rank/score, which is critical for debugging recall regressions and for explaining results to end users.