Secondary Source★

A secondary source is an account or summary of a Primary Source, one step removed. This includes project readmes, design documentation, architectural specifications, and model-generated compaction summaries.

You load secondary sources whenever you:

• Read a library's getting-started guide rather than reading its raw source code.
• Expose a plan.md checklist summarizing previous session edits.
• Consult database diagrams instead of querying active tables.

The Trade-off: Headroom vs. Drift

Secondary sources are the foundation of Context management. Because the Context Window is finite, you cannot load every single primary source file. Secondary sources solve this by compressing information, trading fidelity for headroom:

• Fidelity Loss: By design, secondary sources are lossy. The summary or documentation writer decided what was important at that time, discarding edge cases or parameters that might turn out to be critical today.
• Documentation Drift: Code evolves rapidly, but documentation does not. A secondary source often describes last month's architecture with complete, outdated confidence, leading to model Hallucinations.

To mitigate this, a high-quality secondary source should always leave a Context Pointer (like a file path or URL link) back to the primary source, letting the reader follow the pointer when precision is required.

Field Applications

When compiling specifications for an agent, write brief secondary summaries that explicitly cite the source file paths:

• Template: "The router handles auth (see source file: src/api/auth.ts)."