Appearance
Search
search@1 is the Forge Pool primitive family for distributed retrieval, ranking, candidate generation, and search-time evaluation.
It supports search workloads where relevance is computed through a controlled execution contract rather than hidden behind a single opaque service boundary.
What search@1 is For
Typical uses include:
- candidate retrieval
- distributed ranking
- relevance scoring
- multi-stage search pipelines
- uncertainty-aware search evaluation
- retrieval plus post-score reduction
Example Identity
json
{
"op": {
"name": "search",
"version": 1,
"profile": "rank.v1"
}
}Design Position
Search in Forge Pool is not treated as a disconnected product silo. It is treated as a primitive family that can participate in the same Kernel execution model as simulation, graph, tensor, or media workloads.
That means search execution can inherit:
- consistent policy controls
- deterministic replay discipline where applicable
- artifact persistence
- integration with adapters and orchestration layers
Typical Output Surfaces
Search workloads commonly produce:
- ranked candidates
- score distributions
- per-stage metrics
- retrieval traces
- artifact references
- replay metadata
Why This Matters
As Forge Pool grows, search becomes more powerful when it is not isolated. It can combine with graph, ensemble, media, tensor, and memory surfaces as part of one execution fabric.
That is the strategic reason search@1 belongs inside the Kernel family model.