# What a Humanly writing certificate can show

A Humanly writing certificate packages evidence about a recorded writing session. Depending on the configured detectors and sharing settings, it can show writing setup, source composition, activity, replay and behavior analysis. Its integrity seal helps verify protected certificate fields. The certificate is not an authorship verdict, proof of originality or a guarantee that no outside AI was used.

## Read each part for the question it answers

| Evidence | What it describes | What it does not establish |
| --- | --- | --- |
| Writing setup | The rules and options recorded with the writing evidence | That every action outside the workspace followed those rules |
| Final-text composition | Recorded sources of text remaining in the document | The intellectual origin of every idea |
| Process input volume | Recorded input during drafting, including input later removed | The composition of the final document or the writer's effort |
| Activity and replay | Available recorded events and the document's development | A complete recording of other tabs, applications or devices |
| Behavior analysis | Configured rule-based signals and typing-model results | An automatic finding of misconduct |
| Integrity seal | Whether the protected certificate fields match their signed evidence | That the underlying writing is truthful, original or unaided |

The source categories distinguish typed, pasted and AI-assisted text. “Typed” records a way text entered the workspace. Typing can transcribe an external source, so typed text is not proof of idea origin. A pasted quotation may be appropriate, and an AI-assisted passage may be explicitly permitted.

## Fictional example: interpreting the same record

**The following example is fictional. It uses no real document, user, certificate or verification token.**

Suppose a practice response contains 1,000 attributable characters at the end: 700 typed, 200 pasted and 100 AI-assisted. Suppose the session also records 1,600 characters of input before deletions and revisions. The final proportions are 70%, 20% and 10%; the larger input total describes the process, not an extra 600 characters in the final response.

If the task allowed a cited quotation and AI polishing, these proportions could be consistent with the task. They do not prove compliance by themselves. A reader would still examine the quotation, task rules and relevant activity. Nor does the 70% typed portion prove that 70% of the ideas originated without help. Those are different claims from the observed input categories.

## What does verification mean?

The certificate's signature protects selected fields against undetected changes after sealing. Verification checks that signed evidence against Humanly's public verification key. A valid signature says something about the integrity and issuer of those protected records. It does not independently verify the truth of a statement in the document or establish that the named person personally originated its content.

Respect the certificate's stated coverage: a derived summary or a field outside the signed set should not be described as independently protected by that signature. Also check which document and submission the certificate refers to before using it as evidence for a particular piece of writing.

## How should behavior signals be used?

Abnormal Behavior uses event rules; Auto Typing Detector uses a model of writing behavior. A detector can be disabled, unavailable or inconclusive. Missing evidence does not mean “no problem,” and a flag does not prove misconduct. Workspace visibility events indicate that the workspace was hidden or visible again; they do not identify the destination website or activity.

## Who can see the evidence?

The certificate owner can choose public full-text and edit-history display options and an access code. A recipient can only inspect the evidence exposed by the shared view. The owner's view and a task owner's authorized submission view are distinct from that public display.

Read the [privacy policy](https://cloud.writehumanly.net/privacy) before sharing sensitive writing. The [Humanly paper, version 1](https://arxiv.org/html/2607.21758v1), describes certificate integrity and limitations in its certificate and ethics sections. See [AI policy](/reference/ai-policy), [the comparison with AI detection](/reference/provenance-vs-ai-detection) and [the research guide](/reference/research) for further interpretation and published study context.

---

Reviewed: 2026-09-24.

Applies to: Humanly Cloud release baseline 75eba744, checked against source revision d5818c71 on 2026-09-24. Product descriptions cover capabilities shared by that release and source, not a new deployment. Available controls and evidence depend on the writing setup. Research findings refer to arXiv:2607.21758v1.

Canonical: https://cloud.writehumanly.net/reference/evidence

[Markdown version](/reference/evidence/index.md) · [All reference topics](/reference)