Version 1.0.0 · 2026-07-27

The Maha Provenance Standard

A claim-level provenance framework for self-published research artifacts. A written specification of how claims in Maha Strategies Research artifacts are identified, labelled, sourced, bounded, versioned, and corrected.

Internal methodology specification

Not a peer-reviewed standard, not issued by a standards body, and not adopted outside this project. It documents how one research project works, so that its outputs can be checked against a stated rule rather than taken on trust.

Scope

This document is descriptive, not aspirational. Every clause below records a practice already implemented in this project, and each one links to the artifact or endpoint where it is enforced. The clauses were written by reading the codebase, not the other way round.

The problem it addresses is narrow. Self-published research can be produced faster than it can be checked, and AI-assisted synthesis makes the gap wider: a fabricated citation and a real one are typographically identical, and a conjecture and a measurement read the same once paraphrased. The response here is to attach provenance and epistemic status to the individual claim rather than to the document, and to publish both in a form that survives extraction.

What it does not attempt: this is not a general research methodology, not a peer-review substitute, and not a proposal for anyone else to adopt. It is one project’s working rules, written down so that its outputs can be checked against a stated rule instead of taken on trust.

Clauses — 14

Grouped by prefix: P provenance, C claim structure, S separation of judgement from fact, V versioning and correction, M machine reuse. Each clause carries evidence links so the claim that it is followed can be checked.

  1. P1

    Every citation carries a provenance tag

    Rule
    A cited work must be marked with the basis on which it is cited: independently resolved against a primary index, cited in good faith but not re-resolved, or foundational and cited by journal reference. The tag states what was checked, not how good the work is.
    Why it exists
    A citation list is the easiest part of a document to fabricate and the least likely part to be checked. Tagging forces the question "was this actually resolved?" to be answered before publication rather than assumed afterwards.
  2. P2

    Verification is not endorsement

    Rule
    Marking a citation verified asserts only that the identifier resolves to a real work and that the work is correctly placed in the argument. It asserts nothing about whether that work is right.
    Why it exists
    Without this separation, a well-sourced document starts to read as a well-supported one. In a contested field most cited works disagree with each other, and a provenance system that blurred the two would misrepresent the literature it maps.
  3. P3

    A derived artifact inherits its source’s citation set

    Rule
    An artifact built on a verified source may cite only works already in that source’s verified set. Where a claim would benefit from a citation outside the set, the gap is stated on the claim instead of being filled.
    Why it exists
    The failure mode this prevents is subtle: a derived work adds a plausible, famous, entirely real citation that was never checked by anyone in the chain. Closing the set makes the boundary of what was verified survive derivation.
  4. C1

    Every claim carries an epistemic status

    Rule
    A claim must be labelled established result, active research, conjecture, or speculative interpretation. The status is part of the claim, not commentary on it, and must travel with the wording wherever the claim goes.
    Why it exists
    Prose flattens. A sentence about a conjecture and a sentence about a measurement read identically once quoted out of context. An explicit label is the only part of a claim that survives paraphrase intact.
  5. C2

    Claims have stable identifiers that are never reassigned

    Rule
    Each claim receives a permanent identifier. An identifier always refers to the same assertion; it is never reused for a different one, and renumbering is not permitted.
    Why it exists
    An external citation to a claim is worthless if the target can silently change underneath it. Fixing identifiers is what makes a claim citable at all.
  6. C3

    Exclusions are content, not disclaimer

    Rule
    Each claim states its scope and limitations, and each artifact states what it does not establish. These statements are published in the same machine-readable form as the claims themselves, not confined to a footer.
    Why it exists
    A disclaimer at the bottom of a page does not survive extraction. A limitations field attached to a claim record does. If the boundary matters, it has to be structured like the content.
  7. S1

    Curator annotation is kept separate from bibliographic fact

    Rule
    Fields taken from a source and fields added by the curator are stored separately and labelled distinctly in the published record. A reader must be able to tell which is which without reading the source code.
    Why it exists
    Editorial judgement is legitimate and useful; passing it off as a property of the source is not. Keeping the two apart in the data model makes the boundary visible rather than a matter of the curator’s memory.
  8. S2

    Derive, do not recall

    Rule
    Where a metadata value can be computed from a recorded identifier, it must be computed rather than remembered. The basis of the derivation is published alongside the value.
    Why it exists
    Recalled bibliographic detail is exactly where fabrication enters. A publication year decoded from an arXiv identifier can be re-derived by anyone; one written from memory cannot be distinguished from a plausible guess.
  9. V1

    One version number, and a separate review date

    Rule
    An artifact carries a single version that all its endpoints derive from, plus a review date recording when its claims were last checked. Packaging changes advance the version; only claim review advances the review date.
    Why it exists
    Two dates answer two different questions: has the artifact changed, and has anyone re-read the claims lately. Collapsing them into one hides which kind of change occurred.
  10. V2

    Corrections are disclosed, not silently replaced

    Rule
    A withdrawn or corrected result is described in the record rather than quietly overwritten. The version history states what changed and why.
    Why it exists
    The value of a self-published corpus rests entirely on whether its corrections are visible. A project that silently fixes its mistakes provides no basis for trusting the parts that were never flagged.
  11. V3

    The artifact states its own review status

    Rule
    Every artifact declares whether it is peer reviewed. Self-published status is stated on the page, in the citation exports, and in the machine-readable record. Repository publication is never represented as review.
    Why it exists
    A stable URL, a DOI, and a citation export together look a great deal like a journal article. Saying plainly that it is not one is the minimum a self-publisher owes a reader.
  12. M1

    Machine-readable parity

    Rule
    Anything a human reader is shown — claims, statuses, sources, limitations, exclusions, versions — is published in machine-readable form carrying the same labels. The structured record must not be a weaker version of the page.
    Why it exists
    Most downstream reuse is now machine reuse. If the caveats live only in prose, they are dropped exactly when they matter most.
  13. M2

    Packaged artifacts carry their own use boundaries

    Rule
    An artifact packaged for reuse ships explicit instructions on how it may be represented, including that it confers no authority on a system that consumes it. The instructions are part of the package, not of the website.
    Why it exists
    A downloaded copy leaves the site and its framing behind. Whatever constrains reuse has to be inside the file.
  14. M3

    The instrument is disclosed

    Rule
    Where an AI system contributed to synthesis, verification, or drafting, it is named along with the human direction applied, and its observed failure modes are recorded rather than smoothed over.
    Why it exists
    A reader assessing an AI-assisted document needs to know which parts were machine-generated and what went wrong during production. Concealing the instrument removes the reader’s main basis for calibrating trust.

Conformance levels

L1Labelled
The artifact declares its own status, version, and review status, and does not represent itself as peer reviewed.Adds clauses: V1, V3
L2Sourced
Everything in L1, plus provenance-tagged citations, disclosed corrections, and a stated boundary on what the artifact does not establish.Adds clauses: P1, P2, V2, C3
L3Claim-level
Everything in L2, plus stable per-claim identifiers, per-claim epistemic status and limitations, a closed inherited citation set, separated curator annotation, and machine-readable parity.Adds clauses: P3, C1, C2, S1, S2, M1

Self-assessed conformance in this repository

  • L3The de Sitter / String Swampland Atlas

    Twelve claims with stable identifiers, statuses, and limitations; a closed citation set inherited from its source paper; curator annotation separated from bibliographic fields; full machine-readable parity. Its context pack additionally satisfies M2. One open gap against M3 is recorded below.

  • L2The de Sitter Problem in the String Swampland (working paper)

    Provenance-tagged citations, a verification ledger, disclosed corrections, stated status and version, and instrument disclosure. It does not assign per-claim identifiers, so it does not reach L3 — the atlas built on it supplies that layer.

Conditional clauses

These sit outside the levels because they apply only when their condition holds. An artifact is neither penalised for failing a clause that does not apply to it, nor credited with one it never had to meet.

  • M2 — applies when The artifact is packaged for download or for reuse away from this site.
  • M3 — applies when An AI system contributed to the artifact’s synthesis, verification, or drafting.

Open gaps against this standard

Places where this project does not currently meet its own rules. Published deliberately: a self-assessed standard reporting no shortcomings would not be worth reading, and clause V2 requires that they be disclosed rather than smoothed over.

  • M3 not metThe de Sitter / String Swampland Atlas

    The atlas page discloses the instrument used to produce its source paper, but carries no note naming the AI assistance used in drafting the atlas layer itself — its prose, concept definitions, and claim wording. The working paper it derives from does carry that disclosure. Until the atlas states its own, its M3 conformance is inherited rather than direct.

Conformance is self-assessed. No external body has audited these assignments. The evidence links on each clause exist precisely so that a reader can check the assessment rather than accept it.

Status, limits, and citation

  • This document is descriptive. It records how this project already works so that its outputs can be checked against a stated rule instead of taken on trust.
  • It is not issued by a standards body, has not been peer reviewed, and is not claimed to be in use anywhere outside Maha Strategies Research.
  • Conformance is self-assessed. The evidence links on every clause exist so a reader can check the assessment rather than accept it.
  • The standard has no DOI. Cite it by URL and version.
  • Where a clause and the codebase disagree, the codebase is the fact and the clause is the error.

Cite as

Rajan, M. M. (2026). The Maha Provenance Standard (Version 1.0.0) [Methodology specification]. Maha Strategies Research. https://research.mahastrategies.com/standards/maha-provenance-standard

This standard has no DOI. Cite it by URL and version. Machine-readable record at metadata.json.