0-based chunk number within the source file.
v2.8.0 — content-source kind ("md" | "pdf").
1-based end line of the chunk (inclusive).
1-based start line of the chunk in the source file.
Vault-relative path of the source file (markdown or PDF).
Cosine score in [-1, 1], rounded to 4 decimals. Sort key.
~240-char excerpt from the matching chunk.
.md/.pdf-stripped basename for display.
One chunk-level hit from embeddingsSearch.
Unlike SemanticHit, embedding hits are chunk-scoped (not note- scoped) —
chunk_index/line_start/line_endlet the agent jump to the exact paragraph that matched.