TL;DR: on repo-d, get_repository_stats reports 1235 total nodes and 2741 edges, but the kind breakdown only accounts for 43 of them (42 File + 1 Repository) and community_count is 0. So 1192 nodes carry no kind and the graph appears to have edges but no extracted symbols.
Environment
- memtrace 0.6.0, darwin-arm64
- engine in remote mode on 127.0.0.1:50051
- repo
repo-d, indexed
Repro
get_repository_stats(repo_id=repo-d)
-> {"total_nodes": 1235,
"nodes_by_kind": {"File": 42, "Repository": 1},
"total_edges": 2741,
"community_count": 0,
"last_episode_type": "replay_watermark"}
get_codebase_briefing on the same repo agrees something is off: it reports "0 symbols" while still listing 2741 structural edges.
Expected
One would expect each edge to connect two typed symbol nodes, so an edge count implies a symbol count. If the repo has 2741 edges, those edges would connect typed nodes (Function, Class, Method, ...), and the kind breakdown would reflect them. Communities form over a populated graph.
Actual
42 File nodes and 1 Repository node are clearly the only typed entries. 1192 nodes are untyped, 0 communities formed, and the symbol layer appears not to have completed for this repo, even though edges exist. A benign cause is possible (the repo may be docs-heavy, contain generated files, or use languages the extractor does not support), and I cannot rule that out from the outside.
Hypothesis
Symbol extraction or kind-tagging perhaps did not finish for repo-d, while edge creation and the File/Repository scaffold did. last_episode_type: replay_watermark (no embed_complete) is consistent with a half-finished index pass, though I cannot tell from the outside whether that is the actual cause.
Ask
Is this a known partial-index state, and should get_repository_stats surface an "indexing incomplete" signal when typed-node count is far below total_nodes? Right now the repo looks indexed (nonzero nodes and edges) while appearing symbol-less from these API results.
TL;DR: on
repo-d,get_repository_statsreports 1235 total nodes and 2741 edges, but the kind breakdown only accounts for 43 of them (42 File + 1 Repository) and community_count is 0. So 1192 nodes carry no kind and the graph appears to have edges but no extracted symbols.Environment
repo-d, indexedRepro
get_codebase_briefingon the same repo agrees something is off: it reports "0 symbols" while still listing 2741 structural edges.Expected
One would expect each edge to connect two typed symbol nodes, so an edge count implies a symbol count. If the repo has 2741 edges, those edges would connect typed nodes (Function, Class, Method, ...), and the kind breakdown would reflect them. Communities form over a populated graph.
Actual
42 File nodes and 1 Repository node are clearly the only typed entries. 1192 nodes are untyped, 0 communities formed, and the symbol layer appears not to have completed for this repo, even though edges exist. A benign cause is possible (the repo may be docs-heavy, contain generated files, or use languages the extractor does not support), and I cannot rule that out from the outside.
Hypothesis
Symbol extraction or kind-tagging perhaps did not finish for
repo-d, while edge creation and the File/Repository scaffold did.last_episode_type: replay_watermark(noembed_complete) is consistent with a half-finished index pass, though I cannot tell from the outside whether that is the actual cause.Ask
Is this a known partial-index state, and should
get_repository_statssurface an "indexing incomplete" signal when typed-node count is far below total_nodes? Right now the repo looks indexed (nonzero nodes and edges) while appearing symbol-less from these API results.