Docs/truthlens vision

workspace/knowledge/TruthLens-Vision.md

@@ -44,37 +44,6 @@ This prevents cascading errors across multi-agent workflows.
 
 ---
 
-## 🔎 Data Points TruthLens Must Cover
-
-- **Source Validation**
-  - Verify domain/API/dataset is allowlisted (tiered policy).
-  - Record provenance for every artifact: provider, template, purpose, timestamp.
-  - Enforce robots.txt, ToS, Bright Data acceptable use.
-
-- **Content Integrity**
-  - Deduplicate across feeds (SERPs, marketplaces, social, media).
-  - Tag facts vs. assumptions vs. speculation at item or line level.
-  - Block simulation/fictional language in prose.
-
-- **Compliance Enforcement**
-  - Default to UK-first (GBP, trading standards).
-  - Block PII and login-wall content.
-  - Require delivery model + revenue model in every spec.
-
-- **Execution Safeguards**
-  - Permit only whitelisted commands (`echo|ls|cat|head`) in approved runs.
-  - Reject unsafe shell operators and file writes.
-  - Log every `sf apply` with context + acceptance criteria.
-
-- **Analytical Validity**
-  - Sanity-check metrics (e.g., flag conversion rates >100%).
-  - Detect outliers in scraped datasets (price/volume).
-  - Validate rerank scores against thresholds.
-
-- **Lifecycle Tracking**
-  - Emit `{agent, input, output, ts, compliance_passed}` at each step.
-  - Cross-link artifacts to enable rollback on invalid inputs.
-
 ## 🚀 Long-Term Soulfield OS Positioning
 - **Short term** — SaaS model: hosted Claude Opus + TruthLens orchestration.  
 - **Long term** — Enterprise license: Soulfield OS + TruthLens + pluggable private LLMs.  
@@ -101,3 +70,4 @@ Each template = **research → strategy → execution → analytics**, gated by
 **Soulfield OS** is the orchestration hub.  
 **TruthLens** is the trust filter.  
 Together → “The AI Business OS That Doesn’t Hallucinate.”
+