Subsystem · SRC
Sovereign Reasoning Core
Multi-step reasoning chain execution, strategy selection, depth control, and causal inference for autonomous and semi-autonomous agent operation.
Active — v3.1.4
Active Chains
14
↑ 3 dispatched in last 5m
Avg Reasoning Depth
6.4
→ Within target (max 10)
Strategy Success Rate
91.4%
↑ +3.2% vs last 24h
Depth Limit Hits
2
→ Depth cap: 10 steps
Chain: CHAIN-0047
octane-reason-v10 · Strategy: CoT-Extended
Step 4/7
✓
Problem Decomposition
Break the primary objective into atomic sub-tasks
✓
Evidence Retrieval
Query CBE belief store and external knowledge
✓
Hypothesis Formation
Generate candidate solutions from evidence
4
Causal Validation
Validate causal consistency of top hypotheses against CBE state
→ Checking hypothesis H1: causal chain depth 6…
→ Cross-referencing belief:causal from CBE…
→ Delta analysis running…
→ Cross-referencing belief:causal from CBE…
→ Delta analysis running…
5
Scoring & Ranking
Score candidates by confidence, cost, reversibility
6
Selection & Synthesis
Select optimal strategy and synthesize output
7
Dispatch to ELX
Send final reasoning output to ELX for execution
Step 4 of 7 · 57% · Est. remaining: ~1.2s
All Active Chains
| Chain ID | Strategy | Depth | Progress | Status |
|---|---|---|---|---|
| CHAIN-0047 | CoT-Extended | 4/7 | Active | |
| CHAIN-0046 | Tree-of-Thought | 2/10 | Active | |
| CHAIN-0045 | MCTS | 7/10 | Converging | |
| CHAIN-0044 | Reflexion | 3/5 | Active | |
| CHAIN-0043 | CoT-Extended | 5/5 | Complete | |
| CHAIN-0042 | Direct | 1/1 | Complete | |
| CHAIN-0041 | Tree-of-Thought | 10/10 | Depth Limit |
Chain-of-Thought Extended
Default
Linear step-by-step reasoning with extended intermediate steps and self-consistency checking. Best for structured, well-defined problems.
Max Depth: 10
Branching: 1
Success: 91.4%
Tree-of-Thought
High Depth
Explores multiple reasoning branches in parallel, evaluating and pruning at each node. Best for ambiguous or multi-solution problems.
Max Depth: 10
Branching: 3
Success: 88.7%
Monte Carlo Tree Search
Exploratory
Uses simulation-based search to explore solution spaces probabilistically. Best for planning and optimization under uncertainty.
Max Depth: 8
Simulations: 128
Success: 86.2%
Reflexion
Self-Correcting
Iterative self-evaluation loop where the agent critiques and refines its own reasoning output. Best for tasks requiring high accuracy.
Max Depth: 5
Iterations: 3
Success: 93.1%
Direct Inference
Fast
Single-shot inference with no intermediate steps. Best for simple, well-scoped tasks where latency is critical.
Max Depth: 1
Branching: —
Latency: 18ms
Custom Strategy
Depth Limits & Budgets
Global Depth Cap
Hard stop at max depth regardless of strategy
10
Token Budget per Chain
Maximum tokens allocated to a single reasoning chain
32K
Auto-Summarize at Depth Limit
Compress chain to summary when depth cap is hit
Depth Limit Alerts
Notify when chain reaches 80% of depth cap
Parallel Chain Limit
Max concurrent reasoning chains
16
Reasoning Depth Distribution
Depth Utilisation
Deep (≥7)55%
Mid (3–6)30%
Shallow15%
Avg depth per strategy
CoT-Extended5.8
Tree-of-Thought8.1
MCTS7.4
Reflexion3.2
Reasoning Performance (30d)
Chain Completion Rate
Completed successfully
91.4%
Hit depth limit
5.2%
Timed out
2.1%
Aborted by operator
1.3%
Avg tokens/chain
12,420
Avg time/chain
842ms
Chains today
1,840
CBE round-trips
4,210