The central claim is straightforward.

Authorship can be established by observing the writing process in real time and binding that process to the resulting document with verifiable evidence.

This is not detection. It is instrumentation.

Real Time Signal Capture

Puddin operates during composition, not after the fact.

As a user writes, the system captures behavioral signals that describe how the document is being produced. These signals are not linguistic features of the text. They are properties of the writing process itself.

Examples include:

• Temporal structure such as keystroke timing, pauses, and bursts

• Revision behavior such as insertions, deletions, and restructuring

• Input modality such as typed input versus pasted content

These signals form a continuous stream tied to the act of creation. They are recorded as the document evolves, not reconstructed later.

The result is a process level record of authorship activity.

Distinguishing Authored Versus Inserted Content

A critical component is the ability to differentiate between content that is genuinely composed and content that is introduced externally.

Puddin incorporates a machine learning model trained on behavioral patterns of writing. The model does not analyze the semantic content of the text. It evaluates the structure of the input process.

Self authored text exhibits characteristic patterns. It is generated incrementally, with natural variation in timing and revision. Copied or externally generated text typically appears as large insertions with distinct temporal signatures.

The model classifies segments of the document based on these signals. This allows the system to identify which portions were composed by the user and which were inserted.

This is not probabilistic authorship inference from text. It is classification of observed behavior during creation.

From Signals to Evidence

Raw behavioral signals are not sufficient on their own. They must be transformed into a form that can be trusted and verified.

Puddin constructs an authorship record that summarizes the captured signals and binds them to the document at specific points in time.

This record includes:

• A representation of the document state

• A summary of the associated creation signals

• A timestamp indicating when the state was observed

Crucially, this record is generated during writing. It reflects actual events rather than reconstructed hypotheses.

What “Cryptographic Proof” Means in Practice

The term “cryptographic proof” is often used loosely. Here it has a specific meaning.

The authorship record is passed through a cryptographic function that produces a unique, tamper evident fingerprint. This fingerprint is mathematically bound to both the document state and the captured signals.

Any modification to the document or the underlying record results in a different fingerprint.

This enables two properties:

Integrity. The document and its authorship record cannot be altered without detection.
Verifiability. A third party can recompute the fingerprint and confirm that it matches the original.

The system does not rely on trust in a central authority. It relies on standard cryptographic primitives that can be independently validated.

What Is and Is Not Stored

A common concern is whether capturing behavioral signals introduces privacy risk.

Puddin is designed to minimize data exposure while preserving verifiability.

What is stored:

• Derived representations of behavioral signals

• Cryptographic fingerprints of document states

• Metadata necessary for verification

What is not stored:

• Full keystroke logs in raw form

• Continuous recordings of user activity outside the document context

• Sensitive content beyond what is required to bind the document to its proof

The system captures enough information to establish authorship without creating a comprehensive surveillance record.

Third Party Verification

The value of authorship evidence depends on whether it can be verified independently.

Puddin enables a verification flow that does not require access to the original environment in which the document was created.

A verifier receives:

• The document

• The associated authorship record

• The cryptographic fingerprint

Using these inputs, the verifier can:

Recompute the fingerprint from the document and record
Confirm that it matches the original value
Validate that the document has not been altered since the proof was generated

In addition, the behavioral classification allows the verifier to assess whether the document was composed by the claimed author or assembled through external insertion.

This process is deterministic. It does not depend on model interpretation of the final text.

Why This Approach Holds

The system works because it avoids inference.

It does not attempt to guess authorship from linguistic patterns. It records the conditions under which the document was created and binds that record to the artifact.

This introduces three properties that detection systems lack:

Direct observation of the writing process
Tamper evident linkage between process and output
Independent verification without reliance on trust

Conclusion

Authorship evidence must be generated at the moment of creation.

Puddin achieves this by capturing behavioral signals in real time, classifying how content is produced, and binding that information to the document using cryptographic methods.

The result is not a probability. It is a verifiable record of authorship.