Repository Map

This repository contains the installable library, product documentation, examples, and the evidence record behind the public claims. Use this map when deciding what is product surface and what is retained for reproducibility.

Path	Role	Stability
recall/	Core Python library.	Supported API is listed in API.md.
recall_mcp/	MCP server and tool implementation.	Supported MCP tools are listed in API.md.
recall_interop/	Interop adapters used by benchmark harnesses.	Repository support, not packaged library API.
examples/	Runnable product examples.	Stable enough for users to copy and adapt.
corpus/	Small example corpus for local retrieval checks.	Example data only.
docs/	Product, operating, architecture, and evidence guides.	Product docs are the public entry path.
benchmarks/	Benchmark harnesses, protocols, and reproduction helpers.	Evidence support, not library API.
benchmarks/archive/	Older preregistrations and benchmark protocol records.	Audit archive.
results/	Published result summaries, compact artifacts, and claim baselines.	Evidence record.
docs/archive/	Historical changelog and program status records.	Audit archive.
scripts/	Reproduction and maintenance helpers.	Task specific.
tests/	Unit, integration, and regression tests.	Maintained gate.

Reader Paths

Goal	Start here
Try the product	README.md, then examples/README.md.
Connect an agent via MCP, or set up CLAUDE.md/memory files	USING_WITH_CLAUDE.md.
Integrate the library	API.md, PRODUCTION.md, and OPERATING_MODES.md.
Configure deployment	ENVIRONMENT.md, MIGRATIONS.md, and AUTH.md.
Audit public claims	EVIDENCE.md, ../results/README.md, and ../benchmarks/README.md.
Reproduce benchmarks	../benchmarks/README.md, then the referenced result artifact.

What Is Not Product Surface

Benchmark modules, result builders, preregistrations, archived program ledgers, and generated figures are retained so claims can be audited. They are intentionally visible, but they are not the supported application API.