refactor: remove unshipped reflection service

reflexio/README.md

@@ -141,7 +141,6 @@ client (Python SDK)
   - `playbook/` - Playbook extraction, consolidation, and aggregation
   - `agent_success_evaluation/` - Success evaluation
   - `durable_learning/` - Claim/drain durable `learning_jobs` for deferred extraction when `REFLEXIO_DURABLE_LEARNING_QUEUE` is enabled
-  - `reflection/` - Post-horizon reflection extraction
   - `extraction/` - Resumable async extraction agent infrastructure
   - `shadow_comparison/` - Per-turn regular vs shadow verdict judge
   - `evaluation_overview/` - Evaluation-page aggregates and hero metrics

reflexio/lib/_profiles.py

@@ -624,8 +624,8 @@ def reconstruct_profile_change_log(
     * **removed(R)** — entity_ids of ``status_change`` lineage events with
       ``to_status == "superseded"`` and ``request_id == R``.  This is the
       exact signature emitted by ``supersede_profiles_by_ids`` (the dedup
-      soft-delete path).  It is distinct from reflection which emits
-      ``op="revise"``, so reflection events are never mis-counted as removals.
+      soft-delete path). Other revision events are never mis-counted as
+      removals.
 
     Groups are formed over the union of request_ids from both signals.
     Request_id ``""`` is skipped — it would merge unrelated runs.
@@ -656,8 +656,7 @@ def reconstruct_profile_change_log(
 
     # Dedup soft-delete signature: status_change to_status=="superseded".
     # Each such event records one profile removed in the dedup run ``request_id``.
-    # Distinct from reflection which emits op="revise" — so revise events are
-    # never counted as removals here.
+    # Revise events are never counted as removals here.
     removal_by_req: dict[str, list[str]] = defaultdict(list)
     sort_key: dict[str, tuple[int, int]] = {}  # request_id -> (created_at, event_id)
     for evt in all_events:

reflexio/lib/reflexio_lib.py

@@ -7,7 +7,6 @@
 from reflexio.lib._interactions import InteractionsMixin
 from reflexio.lib._operations import OperationsMixin
 from reflexio.lib._profiles import ProfilesMixin
-from reflexio.lib._reflection import ReflectionMixin
 from reflexio.lib._search import SearchMixin
 from reflexio.lib._user_playbook import UserPlaybookMixin
 
@@ -19,7 +18,6 @@ class Reflexio(
     UserPlaybookMixin,
     ConfigMixin,
     GenerationMixin,
-    ReflectionMixin,
     OperationsMixin,
     DashboardMixin,
     SearchMixin,

reflexio/models/api_schema/domain/entities.py

@@ -141,9 +141,8 @@ class Citation(BaseModel):
 
     Carried inline on an Assistant ``InteractionData`` row to mark
     which previously-injected playbook rule or user-profile row
-    materially shaped that response. The server uses these to drive
-    reflection (does the cited rule still look right after seeing how
-    it was applied?).
+    materially shaped that response. The server stores these for retrieval
+    attribution and evaluation.
 
     Attributes:
         kind (CitationKind): Which kind of cited item this references.
@@ -498,7 +497,7 @@ class LineageEvent(BaseModel):
         op (str): create|revise|merge|aggregate|archive|soft_delete|hard_delete|purge|status_change.
         prov_relation (str): W3C PROV relation (see spec §14).
         source_ids (list[str]): Records merged/superseded into entity_id.
-        actor (str): Who/what triggered it (consolidator|reflection|offline_optimizer|...).
+        actor (str): Who/what triggered it (consolidator|offline_optimizer|...).
         request_id (str): Triggering request — part of the idempotency key.
         reason (str): Free-text rationale (no PII).
         created_at (int): Unix epoch seconds (0 = unset; storage stamps it).

reflexio/models/config_schema.py

@@ -542,56 +542,6 @@ def _default_agent_success_config() -> AgentSuccessConfig:
     )
 
 
-class ReflectionConfig(BaseModel):
-    """Configuration for the sliding-window reflection step.
-
-    Reflection runs inside ``GenerationService.run`` as its own
-    sliding-window step (window = global ``window_size``, stride = global
-    ``stride_size``, bookmark via ``OperationStateManager``). When
-    the gate opens and at least one Assistant interaction in the window
-    cites a current user playbook / user profile row, the LLM is asked
-    whether any cited rows should be replaced. When ``enabled`` is
-    False the step short-circuits.
-
-    Args:
-        enabled (bool): Master switch. When False, no LLM call is made.
-        model (str | None): Optional model name override. Falls back to
-            ``LLMConfig.generation_model_name`` and then the site
-            default for ``ModelRole.GENERATION`` when None.
-        post_horizon_size (int): Minimum interactions after a citation before
-            reflection judges it with full confidence. Citations near the recent
-            edge of the window with fewer than this many follow-up turns get a
-            'last_chance' judgment with the prompt biased toward no_change.
-            Set to 0 to disable the filter (legacy behavior).
-        max_revisions_per_pass (int): Cap on the number of revision decisions
-            applied in a single reflection pass (regularization). Once the cap
-            is hit, remaining revision-intent decisions are skipped and counted
-            in ``ReflectionResult.capped_count``.
-    """
-
-    enabled: bool = True
-    model: str | None = None
-    post_horizon_size: int = Field(
-        default=3,
-        description=(
-            "Minimum interactions after a citation before reflection judges "
-            "it with full confidence. Citations near the recent edge of the "
-            "window with fewer than this many follow-up turns get a "
-            "'last_chance' judgment with the prompt biased toward no_change. "
-            "Set to 0 to disable the filter (legacy behavior)."
-        ),
-        ge=0,
-    )
-    max_revisions_per_pass: int = Field(
-        default=8,
-        gt=0,
-        description=(
-            "Cap on revision decisions applied per reflection pass "
-            "(regularization; excess are skipped)."
-        ),
-    )
-
-
 class RetrievalFloorConfig(BaseModel):
     """Read-path relevance floor: drop search results below a per-arm cross-encoder score.
 
@@ -877,8 +827,6 @@ class Config(BaseModel):
     api_key_config: APIKeyConfig | None = None
     # LLM model configuration overrides
     llm_config: LLMConfig | None = None
-    # Post-publish reflection service configuration
-    reflection_config: ReflectionConfig = Field(default_factory=ReflectionConfig)
     # Read-path relevance floor (per-arm cross-encoder score cutoff)
     retrieval_floor: RetrievalFloorConfig = Field(default_factory=RetrievalFloorConfig)
     # Optional GEPA-backed playbook content optimizer
@@ -949,7 +897,6 @@ def _migrate_field_names(cls, data: Any) -> Any:
             for key in (
                 "window_size",
                 "stride_size",
-                "reflection_config",
                 "playbook_optimizer_config",
                 "lineage_gc",
                 "expiry_reclamation",

reflexio/server/README.md

@@ -18,7 +18,7 @@ Description: FastAPI backend server that processes user interactions to generate
   - [Playbook Extraction](#playbook-extraction)
   - [Agent Success Evaluation](#agent-success-evaluation)
   - [Durable Learning Queue](#durable-learning-queue)
-  - [Reflection and Async Extraction](#reflection-and-async-extraction)
+  - [Async Extraction](#async-extraction)
   - [Shadow Comparison and Evaluation Overview](#shadow-comparison-and-evaluation-overview)
   - [Playbook Optimizer and Braintrust](#playbook-optimizer-and-braintrust)
   - [Lineage](#lineage)
@@ -212,12 +212,12 @@ python -m reflexio.server.scripts.manage_invitation_codes list --show-used
 **Service Boundary**: The service layer owns LLM orchestration, extraction, evaluation, optimization, search preparation, storage access, and long-running operation state. API endpoints should validate/authenticate requests, build `RequestContext`, and delegate into `Reflexio` or focused service helpers rather than embedding business logic.
 
 **Encapsulated Components**:
-- **Publish pipeline**: `generation_service.py` coordinates interaction persistence, profile generation, playbook generation, reflection, and deferred evaluation scheduling.
+- **Publish pipeline**: `generation_service.py` coordinates interaction persistence, profile generation, playbook generation, and deferred evaluation scheduling.
 - **Profile memory**: `profile/` extracts, deduplicates, and applies user profile updates.
 - **Playbook memory**: `playbook/` extracts user playbooks, consolidates them against existing rows, aggregates them into agent playbooks, and tracks aggregation change logs.
 - **Evaluation**: `agent_success_evaluation/service.py`, `agent_success_evaluation/runner.py`, `agent_success_evaluation/scheduler.py`, `agent_success_evaluation/components/evaluator.py`, `shadow_comparison/`, and `evaluation_overview/` handle session grading, per-turn shadow verdicts, regeneration jobs, and dashboard-facing rollups.
 - **Durable learning queue**: `durable_learning/scheduler.py` and `durable_learning/worker.py` drain `learning_jobs` after deferred publishes and report coverage through `GET /api/learning_status`.
-- **Async clarification**: `extraction/` and `reflection/` manage resumable agent runs, pending tool calls, prior-answer search, and long-horizon reflection updates.
+- **Async clarification**: `extraction/` manages resumable agent runs, pending tool calls, and prior-answer search.
 - **Search preparation**: `pre_retrieval/` and `unified_search_service.py` handle query reformulation, document expansion, embeddings, and cross-entity search orchestration.
 - **Optimization/integrations**: `playbook_optimizer/` and `braintrust/` run candidate playbook optimization, rollout support, and Braintrust export/sync.
 - **Lineage**: `lineage/` resolves active records across superseded chains and schedules tombstone garbage collection for profile/playbook storage.
@@ -446,13 +446,11 @@ Key files:
 
 **Pattern**: `POST /api/publish_interaction` returns immediately when `wait_for_response=false`; callers use the returned `request_id` with `GET /api/learning_status`. Queue workers run the LLM compute **outside** any writer transaction; only the persist half + the fenced `complete_learning_job()` run inside `storage.commit_scope()`, and must raise/rollback if `complete_learning_job()` returns 0 because another worker stole the lease.
 
-### Reflection and Async Extraction
+### Async Extraction
 
-**Directories**: `services/reflection/`, `services/extraction/`
+**Directory**: `services/extraction/`
 
 Key files:
-- `reflection/service.py`: Post-horizon reflection orchestration for synthesizing longer-range memory artifacts
-- `reflection/components/extractor.py`: LLM extractor used by the reflection service
 - `extraction/resumable_agent.py`: Resumable extraction agent runtime
 - `extraction/resume_scheduler.py` and `extraction/resume_worker.py`: Background scheduling/worker loop for paused extraction runs
 - `extraction/pending_tool_call_dispatch.py` and `extraction/prior_answer_search.py`: Tool surface and prior-answer context for async extraction agents

reflexio/server/billing_meter.py

@@ -231,8 +231,8 @@ def emit_learnings_generated(
 ) -> None:
     """Resolve ``platform_llm`` from config and emit the Learning value facet.
 
-    Convenience wrapper for the non-extraction learning-mutation paths (reflection,
-    resumable-extraction finalization, aggregation, offline-tuner auto-apply). It
+    Convenience wrapper for non-extraction learning-mutation paths such as
+    resumable-extraction finalization, aggregation, and offline-tuner auto-apply. It
     owns the ``configurator.get_config()`` + ``platform_llm_from_config`` lookup so
     each call site stays a thin one-liner, and — critically — is **guarded**: the
     product path must never fail because metering failed, so config resolution and
@@ -244,7 +244,7 @@ def emit_learnings_generated(
         org_id: Organisation identifier.
         configurator: Object exposing ``get_config()`` for platform-LLM resolution.
         count: Number of learnings durably produced by this path.
-        source: Metering source/path label (e.g. ``"reflection"``).
+        source: Metering source/path label (e.g. ``"offline_optimizer"``).
         pipeline: Optional pipeline tag (e.g. ``"playbook"``).
         user_id: Optional user ID tied to the generated learning.
         request_id: Optional request correlation ID.
@@ -302,8 +302,8 @@ def emit_learnings_generated_records(
     Entity-backed counterpart to :func:`emit_learnings_generated`, currently
     adopted by two of the non-extraction learning-mutation paths —
     resumable-extraction finalization and aggregation — the callers with
-    durable per-record ids in scope. Extraction, reflection, and offline-tuner
-    auto-apply do not have a safe 1:1 id per unit of count (see
+    durable per-record ids in scope. Extraction and offline-tuner auto-apply do
+    not have a safe 1:1 id per unit of count (see
     :func:`record_learnings_generated_records`) and use the count-based
     :func:`emit_learnings_generated` fallback instead. Same guard semantics:
     config resolution and emission are wrapped and any exception is logged
@@ -314,7 +314,7 @@ def emit_learnings_generated_records(
         org_id: Organisation identifier.
         configurator: Object exposing ``get_config()`` for platform-LLM resolution.
         learning_ids: Ids of the learnings durably produced by this path.
-        source: Metering source/path label (e.g. ``"reflection"``).
+        source: Metering source/path label (e.g. ``"aggregation"``).
         pipeline: Optional pipeline tag (e.g. ``"playbook"``).
         user_id: Optional user ID tied to the generated learning.
         request_id: Optional request correlation ID.