FI daily research brief 2026-10-21 (activity-core rhythm)

briefs/2026/10/2026-10-21.md

@@ -0,0 +1,49 @@
+---
+date: 2026-10-21
+timezone: Europe/Berlin
+author: rein-aharness
+status: final
+sources_checked:
+  - docs/sources-allowlist.md
+  - automated field scan via llm-connect
+---
+
+# Freedom Intelligence Daily Brief — 2026-10-21
+
+> Automated brief (activity-core schedule → rein-aharness). Deltas only.
+
+## Headline deltas
+
+- DeepSeek-R1-Distill-Qwen-7B has been formally verified and is now the primary candidate for the R-spine's default local instruct role, pending final evaluation against Qwen3-8B.
+- DeepSeek-R1-Distill-Qwen-14B has been moved to `collected` status following successful download and verification.
+- No new releases or significant changes observed in frontier/commercial (Axis A), training (Axis C), or harness (Axis D) channels.
+- Llama-3.2-3B-Instruct collection remains blocked by the Hugging Face license gate.
+
+## Frontier & commercial (axis A)
+
+*(none)*
+
+## Edge / local / open (axis B)
+
+| Item | Delta | Sources | Lab relevance |
+| ---- | ----- | ------- | ------------- |
+| DeepSeek-R1-Distill-Qwen-14B | The 14B parameter model from DeepSeek's R1-Distill-Qwen series has been successfully collected and its catalog entry updated to `collected` status. This model is under an MIT license. | https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B | This model is now fully available for local reasoning tasks, providing a stronger option than the 7B variant for more complex problems. Its MIT license ensures full operational flexibility. |
+
+## Training & specialization (axis C)
+
+*(none)*
+
+## Harness & fleet (axis D)
+
+*(none)*
+
+## Collection candidates
+
+*(none)*
+
+## Lab implications
+
+- Operators should proceed with the final evaluation of DeepSeek-R1-Distill-Qwen-7B against Qwen3-8B for the 'default local instruct' role, leveraging the newly collected 14B model for comparative reasoning tasks.
+- The successful collection of DeepSeek-R1-Distill-Qwen-14B allows for immediate deployment in local reasoning benchmarks and agentic workflows requiring higher capability.
+- Continue monitoring for resolution of the Hugging Face license gate for Llama-3.2-3B-Instruct, as it remains a key edge model candidate.
+- No new collection candidates identified today; focus remains on evaluating existing candidates and resolving collection blockers.