v0.1 · DRAFT FOR PUBLIC COMMENT · MAINTAINED BY MAHA STRATEGIES LLC · SPEC TEXT CC BY 4.0 · DOI: 10.5281/ZENODO.21241308
MPS is a claim-level tagging system for nonfiction produced with or without AI assistance. It makes the epistemic status of every substantive claim in a document explicit, auditable, and machine-readable.
It exists because AI-assisted writing fails in a characteristic way: fluent, confident, well-formatted fabrication. Document-level disclosure — “AI was used in this work” — tells the reader nothing about which sentences to trust. MPS operates at the claim level.
New to the standard? Read the concise MPS explainer for its scope, limits, citation format, and canonical sources.
RUN A FREE PREFLIGHT →RUN A PRIVATE PREFLIGHT — $49 →PURCHASE API AUDIT ACCESS →
MPS applies to substantive claims: statements of fact, attribution, quantity, causation, or expert consensus that a reader might reasonably rely on. It does not apply to opinion clearly framed as opinion, rhetorical questions, or structural prose.
Every substantive claim receives exactly one tag.
Confirmed by the author against a primary source, direct computation, or first-hand observation.
Test: Did a human check the primary source or reproduce the result?
Attributed to an identified, citable secondary source the author has read but not independently verified.
Test: Can the reader follow a citation to a real, identified document?
Accurately reports the limits of knowledge: open questions, untested conjectures, contested findings — where the claim's content is the uncertainty itself.
Test: Is the claim honest about what is not known?
Analogy, thought experiment, composite example, or structural metaphor. Carries explanatory weight only; asserts nothing about the world.
Test: Would the argument survive if this were literally false?
Asserted without confirmation: recalled from memory, AI-generated and unchecked, or awaiting verification. A flag of honesty, not a license.
Test: Is this claim still owed work?
Inline form (human-readable documents): a trailing tag per claim or claim cluster — Claim text. [TAG] — as practiced across the Maha Strategies book series.
Structured form (audit records): the JSON record below. The live auditor at /audit exports this format.
{
"mps_version": "0.1",
"document": "string (title or URI)",
"audited": "ISO-8601 date",
"claims": [
{
"id": "c001",
"excerpt": "verbatim claim text",
"tag": "VERIFIED | SOURCED | BOUNDARY | ILLUSTRATIVE | UNVERIFIED",
"rationale": "why this tag",
"source": "citation or null",
"action": "none | verify | cite | reword | remove"
}
],
"summary": { "counts_by_tag": {}, "compliance": "pass | conditional | fail" }
}Site-level declaration: a document or site may declare its provenance regime in llms.txt or structured-data metadata: provenance-standard: MPS/0.1.
The MPS Registry publishes versioned, machine-readable claim records for this standard. Each public record includes its evidence context, review metadata, and a content hash.
MPS-Declared — the document states it follows MPS and tags its claims.
MPS-Audited — an independent party has produced a structured audit record (§4).
MPS-Certified (reserved) — audited, with all UNVERIFIED resolved and a published audit trail.
MPS is complementary to, and stricter than, document-level AI-use disclosure. A compliant document additionally discloses how AI was used (drafting, research, editing) in front matter. MPS does not prohibit AI assistance; it prohibits unlabeled uncertainty.
Developed 2025–2026 across the Maha Strategies book series and research program, including an audit of AI fabrication failure modes and a pre-registered 15,000-query study of model reliability. The standard is published under its own discipline.
Maha Strategies conducts full manuscript audits against this standard — every claim resolved source-by-source, delivered as a structured MPS/0.1 record. To request an audit or discuss adopting MPS for your publication: mahastrategies.com/contact or mayone@mahastrategies.com.
Start with a free public preflight for a short passage, or use the private MPS Preflight for a longer document extract and a retained private report before commissioning a source-by-source human review.
MAHA STRATEGIES LLC · MAHASTRATEGIES.COM
FEEDBACK & MANUSCRIPT AUDITS: VIA MAHASTRATEGIES.COM · VERSIONING: SEMANTIC