Experimental tutorial · P7
Graph retrieval,
one variable at a time.
p7_graph_retrieval builds a separate graph-enabled index and enables only server-owned graph expansion during evaluation.
P0 → P7Single-variable experiment
Server-owned graph expansion
What changes
Graph build and expansion.
P7 creates its own index, builds lightweight graph relations, then combines regular hybrid retrieval with graph expansion.
What is disabled
No hidden defaults.
Evaluator v5 removes pipeline, cache, router, graph and RAPTOR ingestion defaults, and keeps rewrite, HyDE, rerank and multi-query off.
What remains fixed
P0 evaluation inputs.
Parser shape, dataset, profile and Top-K remain frozen. The candidate does not reuse the P0 knowledge base.
When comparison is valid
Only direct P0 → P7.
The API requires a compatible evaluator-v5 P0 parent, identical evaluation inputs, and a distinct graph candidate index.
Reproduce it
Run the controlled walkthrough.
The real browser regression starts PostgreSQL, Qdrant, migrations, API, Console, and a controlled immutable Pack. It runs P0 then P7 and verifies graph audit facts, independent indexing and a comparable result.
$ make console-real-tutorial-clone-e2e ✓ P0 baseline completed with hybrid retrieval ✓ P7 builds a separate graph index ✓ P7 graph retrieval enabled ✓ direct P0 → P7 comparison ready