Mind · Reasoning
StrategicReading
Navigate like a researcher, not linearly
Definition
Mechanism
Non-linear Information Retrieval Heuristic. Agent treats documents as random-access databases rather than linear streams. It builds a structural map first (Tree), identifies high-entropy sections (Abstract/Conclusion), and selectively loads only the relevant chunks. Utilizes ComputeBudget.
Preconditions
- Large document identified
- Specific information goal defined
Postconditions
- Relevant sections extracted
- Remainder of document remains unloaded
Invariants
- Budget Awareness: Read volume must be < 10% of total document size unless DeepRead approved.
- Structure First: Must execute `doc_get_tree` or `outline` before reading body content.
Risks
Failure modes
{{context}} Overflow
Loading entire documents saturates memory, displacing instructions.
Linear {{cognitive_bias}}
Reading irrelevant introductions while missing key results buried in appendices.
Fragmented {{context}}
Jumping too aggressively leads to missing critical connective logic.
Relationships
Builds on
References
Autonomous Goal-Directed Entity
Infrastructure
Structural processing error definition
Economic governor for cognitive spend
Infrastructure
Portable execution environment with inherited constraints
Infrastructure
Branching reasoning topology
Nothing in this library builds on this pattern yet.
Reference
Parameters
scan_depthInteger, [1, 10]- Pages/sections to scan before deciding relevance
strategyEnum, {Survey, DeepDive, Needle}- Default: Survey