Experimental tutorial · P4
Sparse retrieval,
one variable at a time.
p4_sparse_retrieval reuses a compatible P0 index and changes only the evaluator's retriever to sparse.
P0 → P4Single-variable experiment
Pure sparse retrieval
What changes
Only the retriever.
P4 evaluates the P0 knowledge base with a sparse retriever. Server wiring removes the pipeline and semantic cache from this evaluator.
What is inherited
The P0 index facts.
No duplicate indexing occurs. Durable audit fields record the sparse strategy and that the P0 index was reused.
When comparison is valid
Only direct P0 → P4.
The API requires one compatible completed P0 parent, the identical knowledge base, frozen environment, and normal evaluation records.
What it does not prove
No automatic superiority.
Inherited index metrics are not a quality, cost, latency, or confidence claim. Read the declared evaluation metrics separately.
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 P4 and verifies sparse audit facts and P0-index reuse.
$ make console-real-tutorial-clone-e2e ✓ P0 baseline completed ✓ P4 uses sparse retrieval ✓ P0 index facts inherited ✓ direct P0 → P4 comparison ready