fix: address CodeRabbit round-2 findings

Aureliolo · Aureliolo · commit 5d550830b449 · 2026-04-03T22:54:55.000+02:00
- Remove .slopconfig.yaml (accidentally committed, not a project file)
- README + roadmap: fix autonomy tiers 5 -&gt; 4 (enum has FULL, SEMI, SUPERVISED, LOCKED)
- decisions.md: clarify fine-tuning pipeline raises NotImplementedError (not just 'tracked')
- open-questions.md: align fine-tuning status with code reality (not 'in progress')
diff --git a/.slopconfig.yaml b/.slopconfig.yaml
diff --git a/README.md b/README.md
@@ -77,7 +77,7 @@ curl http://localhost:3001/api/v1/health
 
 **[Budget & Cost Management](https://synthorg.io/docs/design/operations/)** -- Per-agent cost limits with hierarchical cascading, auto-downgrade to cheaper models at task boundaries, spending reports, budget forecasting, and anomaly detection.
 
-**[Security & Trust](https://synthorg.io/docs/security/)** -- SecOps agent with fail-closed rule engine, progressive trust (4 strategies), configurable autonomy levels (5 tiers), approval gates, LLM fallback evaluator, and audit logging. Container images are cosign-signed with [SLSA L3](https://slsa.dev) provenance.
+**[Security & Trust](https://synthorg.io/docs/security/)** -- SecOps agent with fail-closed rule engine, progressive trust (4 strategies), configurable autonomy levels (4 tiers), approval gates, LLM fallback evaluator, and audit logging. Container images are cosign-signed with [SLSA L3](https://slsa.dev) provenance.
 
 **[Memory](https://synthorg.io/docs/design/memory/)** -- 5 memory types (episodic, semantic, procedural, working, organizational) with hybrid retrieval, tool-based injection (agents call `search_memory` and `recall_memory` on demand), query reformulation, procedural memory auto-generation from failures, consolidation, and pluggable backends.
 
diff --git a/docs/architecture/decisions.md b/docs/architecture/decisions.md
@@ -101,7 +101,7 @@ All significant design and architecture decisions, organized by domain. Each ent
 | MTEB | General passage retrieval | MTEB performance does not transfer to memory retrieval (Pearson: -0.115). Optimizing for MTEB may actively harm memory retrieval quality |
 | Manual evaluation | Custom retrieval benchmarks | Too expensive to maintain. LMEB provides a standardized, reproducible alternative |
 
-**Model selection:** Three deployment tiers recommended based on LMEB scores. See [Embedding Evaluation](../reference/embedding-evaluation.md) for the full analysis. Domain-specific fine-tuning (+10-27% improvement) configured via `EmbeddingFineTuneConfig`; when enabled, the Mem0 adapter uses the checkpoint path as the model identifier. The fine-tuning pipeline (data generation, hard negative mining, contrastive training) is tracked in #1001.
+**Model selection:** Three deployment tiers recommended based on LMEB scores. See [Embedding Evaluation](../reference/embedding-evaluation.md) for the full analysis. Domain-specific fine-tuning (+10-27% improvement) configured via `EmbeddingFineTuneConfig`; when enabled, the Mem0 adapter uses the checkpoint path as the model identifier. The fine-tuning pipeline stages (data generation, hard negative mining, contrastive training, checkpoint deploy) are not yet implemented -- functions validate inputs and raise `NotImplementedError` (see #1001).
 
 ## Overarching Pattern
 
diff --git a/docs/roadmap/index.md b/docs/roadmap/index.md
@@ -8,7 +8,7 @@ What works today:
 
 - **Agent engine** with ReAct, Plan-and-Execute, Hybrid execution loops, crash recovery, and task decomposition
 - **Budget & cost management** with per-agent limits, auto-downgrade, spending reports, and anomaly detection
-- **Security** with fail-closed rule engine, 5 autonomy tiers, progressive trust, output scanning, and audit logging
+- **Security** with fail-closed rule engine, 4 autonomy tiers, progressive trust, output scanning, and audit logging
 - **Memory** with hybrid retrieval (dense + BM25 sparse), tool-based injection, procedural memory auto-generation from failures, and consolidation
 - **Communication** with message bus, delegation, conflict resolution, and meeting protocols
 - **Workflow engine** with Kanban, Agile sprints, ceremony scheduling (8 strategies), visual workflow editor, and workflow execution from graph definitions
diff --git a/docs/roadmap/open-questions.md b/docs/roadmap/open-questions.md
@@ -23,7 +23,7 @@ Numbers are stable identifiers -- resolved questions are removed without renumbe
 | Cost explosion from agent loops | High | Budget hard stops, loop detection, max iterations per task, auto-downgrade at task boundaries. |
 | Agent quality degradation with cheap models | Medium | Capability-aware prompt profiles (#805) adapt prompts to model tier. Quality gates and minimum model requirements per task type. |
 | Third-party library breaking changes | Medium | Python deps exact-pinned (`==`), JS deps range-based with lockfiles. Integration tests, abstraction layers, Dependabot daily updates. |
-| Memory retrieval quality | Medium | Hybrid retrieval (dense + BM25 sparse with RRF fusion) shipped. LMEB-guided embedding selection implemented. Optional domain fine-tuning pipeline in progress (#1001). |
+| Memory retrieval quality | Medium | Hybrid retrieval (dense + BM25 sparse with RRF fusion) shipped. LMEB-guided embedding selection implemented. Domain fine-tuning pipeline not yet implemented -- config and checkpoint lookup wired, training stages raise `NotImplementedError` (#1001). |
 | Agent personality inconsistency | Low | Strong system prompts, personality presets with condensed/minimal variants planned (#805). |
 | WebSocket scaling | Low | In-process channels for Phase 1. Redis pub/sub planned for distributed deployments. |
 
