diff --git a/plugins/nemo-insights/README.md b/plugins/nemo-insights/README.md
index 2ee4dcee60..baf2355011 100644
--- a/plugins/nemo-insights/README.md
+++ b/plugins/nemo-insights/README.md
@@ -58,8 +58,7 @@ those conventions. ATIF maps its required `agent.name` automatically.
 
 ### Where insights are written
 
-Insights always go to the platform, through the Insights plugin API. There is no
-mode that keeps them off it.
+Insights always go to the platform, through the Insights plugin API.
 
 Pass `--insights-file-output <path>` to also keep a local copy: the platform is
 written first and the file mirrors what it stored, platform ids included, so a
@@ -107,13 +106,50 @@ The plugin SDK is available as `client.insights`, including:
 
 ## Configuration
 
-Periodic analysis settings use the `NEMO_INSIGHTS_` environment prefix. For example:
+Periodic analysis is a *deployment* setting, not a per-run one. Like every other
+NeMo plugin it is a `NemoConfig`, so it can be set either in the `insights:`
+section of the platform config file or through the environment, and the
+environment wins. All settings live under `analyst`, with the
+`NEMO_INSIGHTS_` environment prefix.
+
+| Variable | Config key | Default | Meaning |
+|---|---|---|---|
+| `NEMO_INSIGHTS_ANALYST_ENABLED` | `analyst.enabled` | `true` | Whether the periodic analysis controller runs at all. |
+| `NEMO_INSIGHTS_ANALYST_FREQUENCY` | `analyst.frequency` | `daily` | Cadence for each opted-in agent: `daily` or `weekly`. |
+| `NEMO_INSIGHTS_ANALYST_TIMEZONE` | `analyst.timezone` | `UTC` | IANA name (e.g. `America/Denver`) the schedule is interpreted in. Converted to the server clock at evaluation time, so runs hold their local hour across DST. An unknown name fails validation. |
+| — (see below) | `analyst.run_at_hour` | `0` | Local hour-of-day, 0–23, that scheduled runs fire. |
+| — (see below) | `analyst.run_on_weekday` | `monday` | Day scheduled runs fire. Used only when frequency is `weekly`. |
+| — (see below) | `analyst.job_profile` | `default` | Jobs execution profile for scheduled analyst jobs. |
+| — (see below) | `analyst.base_url` | unset | Platform base URL passed to analyst jobs. When unset, jobs use their active platform context. |
+| — (see below) | `analyst.inference_api_key_secret_name` | unset | Platform secret whose value is exposed to analyst jobs as `INFERENCE_API_KEY`. Temporary until FP-202 moves analyst model execution to platform-registered models. |
 
 ```bash
-export NEMO_INSIGHTS_ANALYST_FREQUENCY=daily
+export NEMO_INSIGHTS_ANALYST_FREQUENCY=weekly
 export NEMO_INSIGHTS_ANALYST_TIMEZONE=America/Denver
 ```
 
+### Setting the underscored fields
+
+Only single-word fields — `enabled`, `frequency`, `timezone` — have a working
+flat environment variable. The plugin sets `env_nested_delimiter="_"`, so a
+name like `NEMO_INSIGHTS_ANALYST_RUN_AT_HOUR` is parsed as the nested path
+`analyst.run.at.hour`, which does not exist. **The variable is ignored
+silently: no error, and the default stays in effect.** Prefer the config file
+for these. To set them from the environment anyway, assign the whole `analyst`
+object as JSON — unlisted keys keep their defaults:
+
+```bash
+export NEMO_INSIGHTS_ANALYST='{"run_at_hour": 17, "run_on_weekday": "friday", "job_profile": "gpu"}'
+```
+
+### Analyst self-observability
+
+`NEMO_INSIGHTS_ANALYST_OBSERVABILITY` is read directly from the environment
+rather than through `NemoConfig`, and is off unless set to one of `1`, `true`,
+`yes`, or `on` (case-insensitive). When enabled, the analyst exports its own
+traces to Intake's workspace-scoped OTLP endpoint. The endpoint must be HTTPS
+unless it is loopback.
+
 ## Development
 
 ```bash
@@ -125,3 +161,25 @@ uv run ruff check plugins/nemo-insights
 
 The analyst-only testbed is in [`testbed/`](testbed/). It can replay pinned
 Intake traces or run Tau2 benchmarks before invoking `nemo agents analyst run`.
+
+## What consumes an Insight
+
+Insights is the analysis half of a two-plugin loop. The
+[NeMo Experimentalist](../nemo-experimentalist/README.md) plugin consumes what
+the analyst produces and uses it to improve the agent against Harbor-compatible
+train and validation datasets:
+
+```text
+nemo agents analyst run → Platform Insight ID (or --insights-file-output mirror)
+                       → nemo agents experimentalist doctor
+                       → nemo agents experimentalist run
+```
+
+The Experimentalist accepts either a Platform Insight ID or a local mirror
+file, so `--insights-file-output` is the option to reach for when you want a
+run that does not have to resolve an ID against Platform. Insights does not
+propose or evaluate agent changes itself; the Experimentalist does not analyze
+traces or host an Insight API.
+
+[Insight-driven optimization](../../docs/agents/insight-driven-optimization.mdx)
+walks the full loop end to end.