How writing process capture works: what universities and institutions actually get

What a sealed writing record contains, what it proves, how to verify it, and what institutional obligations come with deploying process capture — answered for administrators, not engineers.

Published 2026-11-02

What you receive	A sealed, time-stamped record of the composition event — independently verifiable
Verification	Anyone with the record can check its integrity — no account, no vendor access required
Student ownership	The record belongs to the student; they decide when and whether to present it
Consent	Per-session, explicit — the consent record is sealed inside the evidence

What an institution receives

When a student writes using Authmarks, the session produces a sealed record — a structured file that documents what happened while that document was written. The record is cryptographically sealed the moment writing ends. From that point, any modification to any part of it changes the seal in a way that is immediately and mechanically detectable by anyone who checks.

You receive a record that can answer three questions that AI detector scores cannot:

Was this document built incrementally, over time, with the revision pattern of genuine composition? Or did the text arrive in large blocks with no surrounding editing activity?
Was the session duration plausible given the word count and complexity of the finished document?
Has anything been changed since the moment writing ended?

These are observational questions, not statistical ones. The record either shows incremental composition or it doesn't. The seal either holds or it breaks.

What the record contains

A sealed process record includes:

A time-stamped log of the writing session — when it started, how long the writer was actively typing, when it ended
The sequence of composition events: text typed, text deleted, text pasted in as a block, and the timing between each
Document content at the time of sealing — what the final text was when writing ended
The student's consent record — when they agreed to recording, what was explained to them, and that it was given before capture began
The cryptographic seal — a mathematical fingerprint computed over the entire record, stored alongside it

What it does not contain:

Video or screenshots of the student's screen
Content of other applications the student had open
Audio or camera data
Any activity outside the writing session

The record is scoped to the writing event. It is not a surveillance log.

How verification works

The sealed record is designed to be verified by anyone — including the student, a hearing panel, an ombudsperson, or a court — without logging into any platform and without involving Authmarks.

The verification step is mechanical: take the record, run a standard hash computation over it, and compare the result to the seal value stored inside it. If they match, nothing has been changed since the moment the seal was applied. If they differ, something was modified.

This means the verification does not depend on trusting Authmarks, the institution, or the student. It depends on a mathematical property of the seal algorithm — a property that is publicly documented, independently testable, and used across digital security worldwide. Federal Rules of Evidence Rule 901 frames authentication as showing that an item "is what the proponent claims it is." A hash check delivers exactly that: a mechanical, falsifiable test.

The verify page demonstrates this in practice — walk through a real sealed record and see the check.

What it proves — and what it does not

What a process record can establish

That a writing session occurred, of a given duration, at a given time
That the text arrived gradually, through keystrokes, consistent with composition — or that it arrived in blocks, consistent with paste-and-submit
That the session timeline is plausible for the document's length and complexity
That the record has not been altered since sealing

Crossley et al. (EDM 2024) found that process features from writing sessions distinguished genuinely composed text from transcribed text with 99% accuracy in controlled conditions. The signal — how text arrives in a document, not just what the text says — is the one AI detectors cannot read.

What a process record cannot establish

Who was at the keyboard. A process record shows what happened at a device during a session. It does not, on its own, confirm the identity of the person who was typing. Corroborating the student's identity — through a student conversation, access logs, or other means — remains part of the investigative process.
That no assistance was used. A process record shows genuine composition took place. It does not rule out that a student discussed ideas with another person, used research tools, or received feedback on a draft. What it rules out is the scenario where no composition occurred at all — where the text was generated externally and submitted as-is.
Intent. A record of what happened is not a record of why. The interpretation of the evidence — what it means in context — remains a human judgment.

These limits are not weaknesses of the record; they are the honest scope of what observation can establish. A hearing panel that understands them is better positioned to weigh the evidence fairly.

How it works in an integrity case

When a student presents a process record

If a student was recorded with Authmarks and a concern arises, they can present the sealed record to the hearing panel. The panel can verify the seal, examine the session timeline, and see whether the composition pattern is consistent with genuine writing. A record showing extended composition over multiple sessions, with realistic revision activity, answers the question the AI detector raised — not by arguing with the score, but by presenting independent observational evidence that predates the dispute.

The OIA's casework guidance on AI and academic misconduct is explicit: institutions bear the burden of proof, not the student. A verified process record shifts the evidentiary picture significantly — the institution must explain why the observational evidence of composition should be disregarded.

When no process record exists

A student who was not using Authmarks at the time of submission does not have a process record to present. This is normal — Authmarks is not yet universal. The absence of a process record is not evidence of misconduct; it is an evidence gap. The hearing should proceed on the evidence that does exist.

Institutional obligations

Capture requires the student's explicit, per-session consent before recording begins. Authmarks does not record silently. The consent record is sealed inside the evidence — it is not a separate database entry that can be lost or disputed. Deploying Authmarks is not deploying a covert monitoring tool.

Data ownership

The sealed record belongs to the student. The institution does not automatically receive copies of students' writing records. A process record enters an integrity proceeding only if the student chooses to present it. This distinction matters for FERPA compliance and for the fundamental fairness of the process.

Retention

Institutions should consider how long sealed records are retained and communicate this to students. A student who needs their record for an appeal filed two years after submission needs it to still exist. Retention policy should match appeal windows.

For a detailed treatment of privacy and compliance obligations, see FERPA, GDPR, and student writing process data.

What "independent verification" means in practice

The most important property of a sealed process record — the one that distinguishes it from version history, screenshots, or platform-managed logs — is that the verification does not run through any intermediary. An instructor or panel member can check the seal themselves using tools that are already on their computer, without asking Authmarks for anything.

This is the test the verify page is designed to demonstrate. If verification required a vendor account or a vendor API call, the verification would only be as trustworthy as the vendor. When verification is a computation anyone can run, it is as trustworthy as the mathematics.

Sources

NIST SP 800-86 — Guide to Integrating Forensic Techniques into Incident Response — as of 2026-11-02
Federal Rules of Evidence Rule 901 — Authenticating or Identifying Evidence (Cornell LII) — as of 2026-11-02
OIA UK — Casework note: AI and academic misconduct — as of 2026-11-02
Crossley et al. — Plagiarism Detection Using Keystroke Logs (EDM 2024) — as of 2026-11-02

Change notes

2026-11-02 — First published.

Spotted an error? Email sales@authmarks.com and we will correct the page and date the correction.