---
title: calibration operations
---

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

<!-- Generated by `python scripts/codegen/docs_reference.py` from the op registry; edit the models, not this page. -->

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](/docs/reference/http-api#request) its namespace and agent; see [identities and namespaces](/docs/concepts/identities-and-namespaces). The [operations index](/docs/reference/operations) explains role, kind and the MCP `all` listing. Unknown input fields are served and named in `ignored_args`.

<!-- prettier-ignore-start -->

## `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`. |

<!-- prettier-ignore-end -->
