Provenance.

What is Provenance?

In TeXposit, provenance is a cryptographically signed, third-party-verifiable record of who wrote what. It answers: which words came from each human author, which from the AI assistant?

Unlike AI-detection tools that guess based on statistical patterns (known to produce false positives, especially against non-native speakers), TeXposit provenance is deterministic. Attribution is computed from the actual edit history of the project, not from analysis of the finished text. The result is a signed report any third party can verify in seconds.

How It Works

Provenance tracking is always on. You never need to configure it. Every action and change in TeXposit creates a record identifying the origin.

Blockchain anchoring

From the start, each project is submitted to OpenTimestamps (OTS), which anchors it to a blockchain. Once sealed, the seal proves the project existed before a specific time. All subsequent changes are cryptographically chained to this seal, so the edit history cannot be tampered with.

Edit history as commits

Every save, whether you typed it, the AI assistant inserted it, or it came from the initial project template, is stored as a git commit. Each commit carries the author's identity and a source tag.

Word-level diff attribution

At report generation time, TeXposit walks the commit log and computes the difference between each consecutive version of every file. This gives a concrete, per-author word count that reflects writing effort rather than final presence in the document.

AI textual correlation

Separately, the prose each human wrote is compared against the project's AI chat history using word-weighted sentence-embedding similarity. This measures how much of the human-written text echoes what the assistant proposed in chat. That's distinct from words the AI inserted into the document directly.

Signing & verification

When you generate a report, the document is re-compiled and its SHA-256 hash, together with the report's core identifiers, is signed with HMAC-SHA256. Anyone can later confirm the report belongs to that exact PDF and has not been altered since.

Example Report

Real PDF generated by the same ReportLab pipeline the app uses.

What a Report Contains

Every report is organised into five sections.

Project

Author & email: The project owner's name and email address, flagged verified or unverified.
Created: The date and time the project was first created.
Entry file: The main file compiled to produce the document (e.g. main.tex).
Description: The project's short description, if one is set.

AI Textual Correlation

Correlation %: The share of human-authored text that semantically matches the project's AI chat log, computed with word-weighted sentence-embedding cosine similarity.
Near-verbatim %: The portion of that overlap that is near-verbatim rather than merely similar in meaning.
Method note: No universal acceptability threshold exists; interpretation is governed by venue or institution policy.

Integrity

Report ID: A unique 10-character identifier for this report (e.g. M7KX2AQPNR).
Project ID: The permanent UUID of the TeXposit project.
First commit: The git hash and date of the project's earliest commit, anchoring its origin.
OTS seal: The OpenTimestamps blockchain seal status (sealed, pending, or unavailable).
HMAC-SHA256: A cryptographic signature over the report's core fields; verify it to confirm nothing was altered after generation.
PDF SHA-256: Hash of the compiled document, used to confirm a submitted paper matches the certified version.
Verify QR: A QR code linking to the verification page, pre-filled with this report.

Contributors

Per-contributor row: Every human author and the AI assistant, each with a distinctly coloured and shaped marker, name, email, and role.
Words · Edits · Share: Words added and number of edits per contributor, and their share of the total. Tallied from added text in each diff, not final presence in the document.

Edit Timeline

Per-file activity: One row per file; each edit is a marker placed on a time axis, coloured and shaped by contributor and sized by how many words it added.
Legend: Maps every marker colour and shape back to its contributor.

Source View: Per-Character Attribution

The Provenance panel has two tabs: Reports (the signed PDF) and Source View. Source View opens the current file in a read-only editor where every character is colour-coded by its author. Hover any segment to see who wrote it and when. Use the history dropdown to step back through any past revision.

Blue is AI-generated text. Each human author gets a distinct colour. Grey is template or genesis text, and purple marks text pasted from an unknown source.

Source View·main.tex

TemplateBob B.Daniel S.AI (TeXposit)

Loading editor…

For Authors: Students, Researchers & Employees

Whether you are a student submitting coursework, a researcher submitting to a journal, or an employee writing a business report, provenance protects your intellectual contribution and gives you verifiable evidence of your authorship.

Open the project you want to generate a report for. In the editor, select the Edit dropdown and click Provenance. The Reports tab shows existing reports on the left with a preview on the right; click Generate Report to create a new one (this also produces a compiled document PDF cryptographically tied to the report). The Source View tab shows the document itself, colour-coded by author.

When submitting, attach the provenance report PDF or share the report ID alongside your document.

What Provenance Does for You

It proves your authorship. If your submission is questioned by a professor, peer reviewer, or employer, the provenance report provides cryptographic, timestamped evidence from your actual edit history.

It lets you disclose AI use precisely. Many institutions now require AI-use disclosure. A provenance report replaces “I used AI” with concrete figures: the exact share of words the AI assistant contributed, alongside a textual-correlation score showing how much of your own writing overlaps the AI chat log.

It protects credit in collaborations. In co-authored documents, the contributor table shows each person's exact word and edit count: a transparent basis for authorship order or contribution statements.

It helps you comply with emerging mandates. Funders (NIH, Wellcome Trust), publishers (EU AI Act Article 50), and universities increasingly require quantified AI-use declarations. A TeXposit report is a compliance-ready artefact.

For Authorities: Publishers, Peer Reviewers, Educators & Employers

If you evaluate submitted work as a conference chair, journal editor, department head, HR manager, or compliance officer, TeXposit provenance gives you a deterministic, third-party-verifiable authorship breakdown derived from the project's actual edit history, not from statistical analysis of the finished text.

How to Verify

Go to app.texposit.com/provenance. The page has two tabs: Report and Match. You can verify in three ways, depending on what the author provided:

By Report ID

Under the Report tab, enter the 10-character Report ID printed on the document or report PDF. Returns the full contributor breakdown, integrity status, the report and compiled document, and a per-character Source View.

Upload report PDF

Under the Report tab, drop the provenance report PDF. The Report ID and Project ID are read from its embedded metadata and the HMAC signature is verified.

Match

Under the Match tab, upload the compiled paper and the report PDF together. Their SHA-256 hashes are compared in your browser (neither file is sent to the server) to confirm the paper is the exact version the report certifies.

What Provenance Does for Your Organisation

For academic venue organisers, it automates AI-use triage during submission review: the verification API returns structured JSON you can integrate into your editorial system.

For university staff and faculty, it verifies student coursework and thesis submissions against your academic integrity policy, giving you a defensible, evidence-based finding unlike detector tools that can be legally challenged.

For company executives and HR, it confirms that business reports, proposals, and IP filings carry adequate human authorship, particularly for patent filings where a human creative contribution must be demonstrated.

For compliance and legal officers, it documents AI use in published content to satisfy emerging regulations: EU AI Act Article 50, California SB 942, Hiroshima AI Process.

Write your next research project with AI transparency.

Unique certification id Blockchain immutability Verification API

Verify a Document Contact for Institutional Access