What causes "Invalid result.ind: {result.ind}" in run-llama/llama_index?

Use a function-calling LLM with the default PydanticSingleSelector for reliable 0/1 output

run-llama/llama_index · error · ValueError

Invalid result.ind: {result.ind}

Error message

Invalid result.ind: {result.ind}

What it means

SQLJoinQueryEngine._query uses a single selector with two choices: ind 0 routes to SQL+other join synthesis, ind 1 routes to the other query engine. Any other index is a programming/selector error and raises ValueError(f"Invalid result.ind: {result.ind}").

Source

Thrown at llama-index-core/llama_index/core/query_engine/sql_join_query_engine.py:347

        """Query and get response."""
        # TODO: see if this can be consolidated with logic in RouterQueryEngine
        metadatas = [self._sql_query_tool.metadata, self._other_query_tool.metadata]
        result = self._selector.select(metadatas, query_bundle)
        # pick sql query
        if result.ind == 0:
            if self._verbose:
                print_text(f"Querying SQL database: {result.reason}\n", color="blue")
            logger.info(f"> Querying SQL database: {result.reason}")
            return self._query_sql_other(query_bundle)
        elif result.ind == 1:
            if self._verbose:
                print_text(
                    f"Querying other query engine: {result.reason}\n", color="blue"
                )
            logger.info(f"> Querying other query engine: {result.reason}")
            return self._other_query_tool.query_engine.query(query_bundle)
        else:
            raise ValueError(f"Invalid result.ind: {result.ind}")

    async def _aquery(self, query_bundle: QueryBundle) -> RESPONSE_TYPE:
        # TODO: make async
        return self._query(query_bundle)

View on GitHub (pinned to afd0fef371)

Solutions

Use a function-calling LLM with the default PydanticSingleSelector for reliable 0/1 output
Do not override the selector with extra choices; SQLJoinQueryEngine passes is_multi=False and expects exactly two options
Catch ValueError around query() and retry once, or route manually to one engine

Example fix

// before
join_engine = SQLJoinQueryEngine(
    sql_tool, other_tool,
    selector=LLMSingleSelector.from_defaults(llm=weak_llm),
)

// after
join_engine = SQLJoinQueryEngine(sql_tool, other_tool)  # default selector from Settings.llm
# ensure Settings.llm is a function-calling model, e.g. OpenAI(model="gpt-4o-mini")

Defensive patterns

Strategy: retry

Try / catch

try:
    resp = join_engine.query(q)
except ValueError as e:
    if "Invalid result.ind" in str(e):
        resp = join_engine.query(q)  # retry: selector output is nondeterministic
    else:
        raise

Prevention

Use the default selector with a function-calling LLM
Do not customize selector choices for SQLJoinQueryEngine — exactly two options are expected
Wrap routing queries with one retry to absorb rare invalid indices

When it happens

Trigger: The selector LLM returns an index outside {0, 1} (e.g. 2+) when routing in SQLJoinQueryEngine, typically with LLMSingleSelector on a weak model or a custom selector misconfigured with extra choices.

Common situations: Non-deterministic LLM output failing the two-choice format; custom selector prompt that lists more than two options; small local models.

Related errors

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/ec49c3d0eaadcd4c. Report an issue: GitHub.