diff --git a/reflexio/server/README.md b/reflexio/server/README.md
index 677be741..e1f1841a 100644
--- a/reflexio/server/README.md
+++ b/reflexio/server/README.md
@@ -424,13 +424,14 @@ Key files:
 - `agent_success_evaluation_constants.py`: Output schema (`AgentSuccessEvaluationOutput`)
 - `agent_success_evaluation_utils.py`: Message construction utilities
 - `scheduler.py`: `GroupEvaluationScheduler` singleton - min-heap priority queue with daemon thread, defers evaluation until 10 min after last request in session
+- `sampling.py`: deterministic per-session sampling for the session-success and retrieved-learning judge families; the publish scheduler samples once and passes explicit booleans to the runner
 - `runner.py`: `run_group_evaluation()` - fetches all requests/interactions for a session, builds `RequestInteractionDataModel` list, runs `service.py`, then the retrieved-learning phase; returns `GroupEvaluationOutcome` (per-family statuses)
 - `components/retrieved_learning_evaluator.py`: `RetrievedLearningEvaluator` - per-learning relevance/impact judges over `Interaction.retrieved_learnings`; results replace the session's `retrieved_learning_evaluation` snapshot atomically (generation + session-fingerprint fenced, see `services/storage/storage_base/retrieved_learning_state.py`)
 - `services/storage/storage_base/evaluation_state_keys.py`: single source of truth for the three evaluation `_operation_state` key formats (agent-success marker, grade-on-demand cache, retrieved-learning state) shared by producers and governance erasure
 
 **Flow**: Interactions → `agent_success_evaluation/scheduler.py` → `agent_success_evaluation/runner.py` → `agent_success_evaluation/service.py` → `agent_success_evaluation/components/evaluator.py` → `AgentSuccessEvaluationResult` → Storage
 
-**Session-Level Evaluation**: Evaluator treats one user's `request_interaction_data_models` in a session as a single conversation. Sampling rate checked once per session (not per-request). Results are keyed by `(user_id, session_id, evaluation_name)` so reused session IDs across users do not clobber each other.
+**Session-Level Evaluation**: Evaluator treats one user's `request_interaction_data_models` in a session as a single conversation. Sampling is checked once per session (not per-request): `sampling_rate` controls session-success, `evaluation_only_sampling_rate` can override evaluation-only publishes, and `retrieved_learning_sampling_rate` lets retrieved-learning judges run at a separate density while inheriting `sampling_rate` when unset. Results are keyed by `(user_id, session_id, evaluation_name)` so reused session IDs across users do not clobber each other.
 
 **Tool Context**: Reads `tool_can_use` from root `Config` level (shared with playbook extraction).
 
diff --git a/reflexio/server/services/README.md b/reflexio/server/services/README.md
index 002a90b4..e9c8c08f 100644
--- a/reflexio/server/services/README.md
+++ b/reflexio/server/services/README.md
@@ -35,7 +35,7 @@ strings before deleting old import paths in the same PR.
 |-----------|-------------|-----------|
 | `profile/` | `ProfileGenerationService` | `service.py`, `components/extractor.py`, `components/consolidator.py` |
 | `playbook/` | `PlaybookGenerationService` | `components/extractor.py`, `components/consolidator.py`, `components/aggregator.py` (cluster-fingerprint change detection) — has its own [README](playbook/README.md) |
-| `agent_success_evaluation/` | `AgentSuccessEvaluationService` | `service.py` (session-level service), `runner.py` (`run_group_evaluation`), `scheduler.py` (`GroupEvaluationScheduler`, 10-min defer), `regen_jobs.py`, `components/evaluator.py` |
+| `agent_success_evaluation/` | `AgentSuccessEvaluationService` | `service.py` (session-level service), `runner.py` (`run_group_evaluation`), `scheduler.py` (`GroupEvaluationScheduler`, 10-min defer), `sampling.py` (per-family session gates), `regen_jobs.py`, `components/evaluator.py` |
 
 ## Durable and Async Extraction
 
diff --git a/reflexio/server/services/agent_success_evaluation/README.md b/reflexio/server/services/agent_success_evaluation/README.md
index 7844868d..12ded390 100644
--- a/reflexio/server/services/agent_success_evaluation/README.md
+++ b/reflexio/server/services/agent_success_evaluation/README.md
@@ -7,6 +7,7 @@ Session-level agent success evaluation module.
 - `service.py`: `AgentSuccessEvaluationService`, the request-path service that runs configured evaluators and saves result rows.
 - `runner.py`: `run_group_evaluation(...)`, the background/manual workflow entry point that loads a session, runs the service, and marks operation state. After agent-success work it also runs the retrieved-learning evaluation (independent completion; generation + session-fingerprint fenced) and returns a `GroupEvaluationOutcome` carrying both statuses.
 - `scheduler.py`: `GroupEvaluationScheduler`, the deferred inactivity scheduler.
+- `sampling.py`: deterministic per-session sampling gates for the session-success and retrieved-learning judge families. `generation_service.py` is the only scheduled-publish caller; it samples once, admits the session when either family is selected, and passes booleans to the runner.
 - `components/evaluator.py`: `AgentSuccessEvaluator`, the LLM evaluator component.
 - `components/retrieved_learning_evaluator.py`: `RetrievedLearningEvaluator`, per-learning relevance/impact judges over `Interaction.retrieved_learnings`; results go to the `retrieved_learning_evaluation` table.
 - `regen_jobs.py`: regeneration job planning and execution; remains root-level because API/admin regenerate flows import it directly.
@@ -14,6 +15,12 @@ Session-level agent success evaluation module.
 - `agent_success_evaluation_constants.py`: prompt/model output constants.
 - `agent_success_evaluation_utils.py`: request DTO and prompt-message construction helpers.
 
+## Sampling Contract
+
+- `AgentSuccessConfig.sampling_rate` controls the session-success judge; `evaluation_only_sampling_rate` can override it for evaluation-only publishes.
+- `AgentSuccessConfig.retrieved_learning_sampling_rate` controls the retrieved-learning relevance/impact judges and inherits `sampling_rate` when unset.
+- Direct runner callers (regen jobs and on-demand grade routes) leave the sampling booleans at their defaults and run both families; sampling remains a scheduler/publish decision.
+
 ## Prompt IDs
 
 - Owns `agent_success_evaluation`, `retrieved_learning_relevance`, and `retrieved_learning_impact`.