JAMESLI.WORK

Case file 05 / Open Source

Memtruth

Evidence and document runtime

An open-source SDK and runtime for evidence, parsing, corpus contracts, chunking, projection, diagnostics, provenance, and citation-aware AI applications.

CLASSIFICATION
Open Source
PUBLIC STATUS
REVIEWED
CASE
05 / 06

The problem

AI applications often turn documents into untraceable text, making it difficult to inspect where a field came from, what parser produced it, or whether a result can be audited.

My role

I shaped and contributed evidence, provenance, citation, parser, CLI, and release contracts across the public repository and its runtime integrations.

Capabilities

  1. 01

    Document parsing contracts

  2. 02

    Per-field citation and provenance structures

  3. 03

    Corpus and chunking contracts

  4. 04

    Projection and diagnostics

  5. 05

    Audit-oriented CLI behavior

Product decisions

  1. 01

    Keep evidence explicit instead of burying source context in prompts.

  2. 02

    Use schema-versioned public contracts for outputs that other systems consume.

  3. 03

    Do not imply that a named trust document is automatically audit-grade.

Evidence

  • 01

    Public repository includes explicit citation and provenance contracts.

  • 02

    Tests cover evidence-aware extraction and readable audit output.

  • 03

    Public releases preserve clone-safe onboarding and language-specific SDK structure.

Honest limits

  • !

    Evidence and provenance support audit work but do not replace domain review.

  • !

    Historical project naming is not used as the primary public identity.

Documents need memory with receipts

AI systems frequently detach extracted fields from the documents that support them. Memtruth keeps parsing, evidence, citations, provenance, corpus contracts, projection, and diagnostics available as explicit public structures.

Contract first

Outputs that other systems consume are schema-versioned. Citation and provenance records are modeled directly. CLI behavior is designed for both readable human inspection and stable machine output.

Honest naming

Evidence and provenance support audit work, but they do not make every output audit-grade by default. Domain review, parser quality, source quality, and the specific acceptance profile still matter.