calibration operations

The exact fields each calibration.* operation takes and returns.

Every op is POST /v1/<op> over HTTP and the MCP tool loom_<op>. The identity fields (org, namespace, agent, user_id, session_id) are filled from the token and the MCP session when a transport can; for an HTTP call with no token on a single-user server, the install's org on every op but code.* and tracker.* (which keep their plane's), and on some ops its namespace and agent; see identities and namespaces. The operations index explains role, kind and the MCP all listing. Unknown input fields are served and named in ignored_args.

calibration.get

Resolve the active calibration for (org, namespace): per-namespace, then cluster-wide, then scalar fallback, with the source tier reported.

POST /v1/calibration.get · MCP loom_calibration_get · role viewer · read · not listed under the MCP all toolset

Input field	Type	Default	Meaning
org	str	required	The tenant (first element of the identity tuple).
namespace	str	required	The namespace the op reads or writes.

Output field	Type	Meaning
source	str	"namespace" / "cluster" / "scalar"
embedder_model	str	The embedder the thresholds were derived for; with scalar, the configured model.
noise_floor	float	Cosine below which a search hit is dropped as noise (distractor P90 + 0.02 when probed).
rerank_skip_above	float	Top-hit cosine at or above which the LLM rerank is skipped (question-to-fact P75, never below hyde_fallback_threshold).
hyde_fallback_threshold	float	The question-to-fact P25 gate, never below query_rewrite_threshold; loom_search reads it as its precision floor.
query_rewrite_threshold	float	The question-to-fact P10 gate, held at least 0.01 above noise_floor.
abstain_threshold	float	Top-hit cosine below which a search abstains (noise_floor + 0.05 when probed).
conflict_threshold	float	Cosine at or above which a write against an active peer is flagged conflicted (paraphrase P50 when probed).
duplicate_threshold	float	Cosine at or above which a write with a canonical-form mismatch is marked duplicate (paraphrase P75 when probed).
similar_p10	float	Probe percentiles; default 0.0 when source is 'scalar' (no probe ran).
similar_p25	float	25th percentile cosine of the paraphrase pairs.
similar_p50	float	Median cosine of the paraphrase pairs.
similar_p75	float	75th percentile cosine of the paraphrase pairs.
distractor_p50	float	Median cosine of the unrelated (distractor) pairs.
distractor_p90	float	90th percentile cosine of the distractor pairs, the anchor of noise_floor.
similar_pairs_n	int	Paraphrase pairs the probe embedded; 0 when source is scalar.
distractor_pairs_n	int	Distractor pairs the probe embedded; 0 when source is scalar.