Query Modes¶

The canonical public modes are:

Mode	Purpose	Notes
`rag`	Grounded document question answering	Default online path
`llm_only`	Ungrounded generation	No document sources
`summarize`	Targeted act summary	Requires `celex`
`compare`	Act-to-act comparison	Requires `celex` and `compare_celex`

Legacy modes such as hybrid, semantic, lexical, graph, search, and qa are normalized to canonical modes for compatibility.

Note

The raw search modes semantic, lexical, and hybrid remain public on the /search endpoint. They are not separate canonical modes on /query.

Runtime decisions¶

The service can terminate with different response states:

        stateDiagram-v2
    [*] --> grounded
    [*] --> weakly_grounded
    [*] --> clarify
    [*] --> hard_block
    [*] --> llm_only

Outcome	Meaning
`grounded`	Evidence and citations are strong enough for a normal answer.
`weakly_grounded`	Evidence is partial or repaired, but still usable.
`clarify`	The question needs narrowing before a grounded answer is possible.
`hard_block`	The request is anchored to legal facts but the system has no evidence.
`llm_only`	The runtime intentionally bypasses retrieval for conversational requests.

Mode alias resolution: packages/lalandre_core/lalandre_core/utils/mode_aliases.py
Shared request and response models: packages/lalandre_rag/lalandre_rag/models/api.py
High-level orchestration: packages/lalandre_rag/lalandre_rag/service.py