RE-call documentation

This directory separates the product documentation from the evaluation record. Start with the operating guides when you are deciding whether to use RE-call. Use the evidence sections when you want to audit a claim or reproduce a benchmark.

Product path

Document	Use it for
../README.md	Product overview, setup, and integrations.
API.md	Supported Python, CLI, and MCP surface.
REPOSITORY_MAP.md	Product, evidence, benchmark support, and archive boundaries.
PRODUCTION.md	Production posture, supported boundaries, and known limits.
CALIBRATION.md	Calibration workflow and generation-aware serving.
MIGRATIONS.md	Migration roles, serving DSNs, and schema operations.
OPERATING_MODES.md	Local, production, quality, hosted, and evaluation operating modes.
ENVIRONMENT.md	Full environment variable reference behind the root template.
DEPENDENCIES.md	Dependency floors, ceilings, extras, and packaging rationale.
AUTH.md	Authentication, scopes, tenant isolation, and deployment limits.
SECURITY_MODEL.md	Detailed threat model behind the root security policy.
USING_WITH_CLAUDE.md	MCP setup and tool behavior.
REASONING_OPERATIONS.md	Opt-in reasoning commands, MCP tools, failure behavior, review policy, and metrics.
BETA_RECRUITING.md	Compliant beta recruiting from public discussions without harvesting emails.

Architecture

Document	Use it for
WRITEUP.md	Architecture, trust semantics, and evaluation summary.
Validity Frontmatter 1.0	The open vocabulary RE-call implements (valid_from, valid_until, supersedes), its resolution rules and its verdict algorithm. MIT licensed and maintained in its own repository, so it stays implementable without RE-call. RE-call is the Python implementation; a zero-dependency TypeScript one ships alongside the spec.
CASE_STUDY.md	Origin story, redacted production context, and public/private boundary.
REASONING_CONTRACT.md	Reasoning vocabulary, invariants, Session 1 baseline fixture, and Session 6 evaluation controls.
REASONING_API.md	Typed reasoning request and response surface, provider ports, serialization, and validation rules.
REASONING_GRAPH.md	Generation-bound reasoning graph projection schema and invariants.
REASONING_RELEASE_NOTES.md	Experimental reasoning release notes, migration notes, and limitations.
TRUTH_EXTRACTION_DESIGN.md	Model backed extraction of truth claims from prose, the validation ladder, and the reviewed write path back into corpus frontmatter.

Evidence and limits

Document	Use it for
EVIDENCE.md	The shortest evidence path: claims, measurements, limits, and withdrawn claims.
../results/FINDINGS.md	What the measurements establish, where they stop, and which claims were corrected.
../results/README.md	Map of committed result summaries and compact artifacts.
../results/RESULTS.md	Published tables and result summaries.
../results/ARTIFACTS.md	Checksum and artifact map for readers auditing a claim.
../CHANGELOG.md	Release notes and upgrade warnings moved out of the README.

Research archive

Document	Scope
RESEARCH_PROTOCOL.md	Rules for benchmark runs, artifact retention, and post-run review.
VISIBILITY_BENCHMARKS.md	Public leaderboard submission track for Kaggle AgentEval, EnterpriseRAG-Bench, LiveRAG, and CRAG.
ENTERPRISE_RAG_VAST.md	Vast.ai GPU runbook for the EnterpriseRAG-Bench SPLADE arm.
ENTERPRISE_RAG_SUBMISSION.md	EnterpriseRAG-Bench answer artifacts, score summaries, and reproduction steps for leaderboard review.
ENTERPRISE_RAG_ONYX_EMAIL.md	Draft submission email, verifier links, artifact hashes, and finalization checklist for Onyx.
archive/CHANGELOG_FULL.md	Full historical changelog.
RAG_TRAINING_STUDY.md	When fine-tuning embeddings helps, and when it does not.
ENTAILMENT_SUPERSESSION_STUDY.md	Near-miss abstention, entailment, and write-time supersession.
REFERENCE_TIME_DESIGN.md	Reference-time handling for temporal retrieval.
DERIVED_BLOCK_DESIGN.md	Machine-owned regenerable block for contradicts and same_entity, isolated from extraction.
their-harness-parity.md	Running RE-call inside Mem0's benchmark harness.

Generated binaries and local render outputs should stay out of narrative documentation unless a document explicitly depends on them.