Where Colophon and TWFF Stand Today

The extension

Colophon records how a piece of writing came together, every edit, paste, and AI interaction, as a tamper-evident TWFF file the writer keeps control of, not a verdict handed down after the fact. It runs inside Google Docs, watches Gemini's "Help me write" suggestions specifically, and never uploads anything until the writer chooses to export or submit.

Install it here: Colophon on the Chrome Web Store.

We're also working toward our first real classroom case study. A teacher creates a course and gets a join code in seconds, no district procurement required to try it on one assignment. Students consent before anything is recorded. A researcher gets pseudonymized, cohort-level data, never named rows, once a real study is actually running. If you're running a class and want to help us get there, or you're a researcher who wants to be kept in mind for access once the data exists, start at colophon.firl.nl/waitlist.

The paper

The working paper behind TWFF (Tracked Writing File Format) is now live on SSRN. It makes the case for recording how a piece of writing came together instead of guessing at it after the fact. AI-text detectors ask whether a document was written by AI. That question has real, documented problems: high false-positive rates against non-native English writers, and no auditable trail behind the answer. TWFF records the writing process itself, every edit, paste, and AI interaction, timestamped and tamper-evident, so the evidence is something you can actually look at.

Read the paper: papers.ssrn.com/sol3/papers.cfm?abstract_id=7272718.

What's actually next

Getting a first real pilot section through the whole pipeline end to end: applied, activated, a student consents, writes, submits, and a teacher actually sees the file in their dashboard. That's the real blocker right now, not a missing feature. Alongside it: Google Classroom and Canvas integration, so students stop uploading files by hand, and aligning the AI-content icons shown in the viewer with the EU's AI-content-labelling recommendations.

The format specification stays open and versioned at github.com/Functional-Intelligence-Research-Lab/twff. Colophon, the reference implementation, is at colophon.firl.nl.