We use cookies and similar technologies to improve your experience, analyze site traffic, and serve personalized content and ads. By clicking "Accept", you consent to our use of cookies. Click "Essential Only" to decline optional cookies.
See our Privacy Policy for more details.
Record the act of writing. Capture the effort, the revisions, the paste attestations. That behavioral record is the proof.
CPoE is a cryptographic protocol that records how text was written — not just what was written. An IETF Internet-Draft hosted by Linux Foundation Decentralized Trust Labs and protected by a pending patent application.
Built on established internet standards
AI-generated text is now indistinguishable from human writing. Existing approaches to this problem are fundamentally flawed.
AI detection asks: “Does this text look AI-generated?”
CPoE answers: “Here is cryptographic evidence that a human wrote it.”
From the first keystroke to a verifiable evidence packet. Each stage is independently auditable.
Each checkpoint includes a Sequential Work Function proof — a computation that provably took real wall-clock time. This prevents batch-fabrication of evidence: you cannot produce a 2-hour writing session in 2 seconds.
Checkpoints are chained using an append-only Merkle tree (Merkle Mountain Range). Each node commits to all previous nodes, so removing or modifying any checkpoint invalidates the entire chain.
All evidence packets are signed using Ed25519 (RFC 8032) via the COSE signing envelope. Keys can be software-generated (T1) or hardware-backed by Secure Enclave / TPM (T2+).
Inter-key timing, pause patterns, revision behavior, and editing cadence form a behavioral fingerprint. This data is hashed into checkpoints — the raw biometrics never leave the device.
AI simulations notoriously fail at pacing. They are Velocity-Aggressive, typing at superhuman speeds (d=-10 to -38), establishing hyper-velocity as a distinct, mathematical AI 'tell'.
Beyond timestamping and hashing, WritersProof analyzes the cognitive signature of writing. Eight independent classifiers operate across three layers — temporal, content, and structural — to distinguish original thought from transcription. Faking all signals consistently is exponentially harder than faking any one.
Cognitive writers pause 8-30x longer before new sentences (thinking). Transcribers pause 2-4x (reading next line).
Original composition produces multi-modal timing (fast motor bursts + medium flow + long thinking). Transcription is unimodal.
Cognitive writers type common letter pairs (th, he, in) much faster than rare ones. Transcribers show uniform speed.
Pre-word pause correlates with word frequency for cognitive writers (rare words = longer pauses). No correlation for transcribers.
Cognitive writers produce non-linear edit streams with multi-character deletions. Transcribers produce near-pure append streams.
Cognitive corrections are semantic (wrong word, restructuring). Transcription corrections are visual (similar characters, skipped words).
Cross-signal consistency check. Faking one signal (e.g., artificial pauses) without matching others exposes the deception.
Deviation from the writer's own historical cognitive profile. Detects imposters even when population norms are matched.
Combined verdict: Signals are weighted by confidence and combined into a single cognitive score (0-100) with an ENFSI-compliant likelihood ratio. The resulting forensic report includes exhibit-labeled evidence sections suitable for legal proceedings.
Evidence packets use CBOR (RFC 8949) for compact binary serialization and COSE (RFC 9052) for standardized signing.
The primary evidence container. Holds the checkpoint chain, behavioral digests, SWF proofs, and optional hardware attestation. Self-contained and independently verifiable.
The verification output. Produced by a verifier after validating a WritersProof packet. Contains the verified tier level, trust chain status, and any policy evaluation results.
Four levels of evidence strength. Each tier builds on the previous, adding stronger cryptographic guarantees and higher forgery costs.
Timestamps + Hashes
A writing process occurred on a specific device at a claimed time.
Personal records, informal proof of authorship.
Hardware Key Exists
A hardware-backed signing key exists on the device, binding evidence to physical hardware.
Academic submissions, professional writing, content marketplaces.
Secure Enclave / TPM
The signing key is protected by verified hardware (Secure Enclave, TPM). Attestation chain verified by third party.
Legal proceedings, insurance claims, regulatory compliance.
Platform Integrity
Full platform integrity verified — secure boot chain, no tampering with OS or capture software.
High-stakes litigation, government contracts, forensic evidence.
WritersLogic is an active contributor to the C2PA core specification, working on external assertions, hashed URI support for process evidence, HTML manifest embedding, and structured text embedding for C2PA manifests. CPoE evidence packets integrate natively with C2PA content credentials.
C2PA covers images, video, and audio provenance. CPoE fills the gap for text content, using the same external reference mechanism already defined in the C2PA specification.
CPoE evidence packets can be embedded within JUMBF boxes alongside C2PA manifests, or referenced externally via URI. Both patterns are supported by the specification.
A document with a C2PA manifest and CPoE evidence provides end-to-end provenance: who created it (C2PA identity), how it was created (CPoE effort evidence), and when (timestamp anchors).
CPoE is developed through established open standards processes and governed by recognized institutions.
The Internet-Draft draft-condrey-rats-pop is submitted to the WritersProof Protocol Remote ATtestation procedureS (RATS) Working Group for standardization.
The CPoE Lab is hosted under LF Decentralized Trust Labs, providing neutral governance, community infrastructure, and long-term stewardship.
LF Lab PageCBOR tag values for WritersProof (evidence packet) and CWAR (attestation result) are registered with IANA, ensuring global uniqueness and interoperability.
IANA CBOR TagsThe CPoE protocol is protected by a pending patent application. Consistent with WritersProof Protocol policy (RFC 8179), the patent holder has committed to reasonable and non-discriminatory (RAND) licensing terms for implementations of the standard.
CPoE is an open standard. Read it, implement it, improve it.
The full Internet-Draft specification, including wire formats, threat model, evidence tiers, and security considerations.
The reference implementation is open source (Rust). Build your own implementation, run the test vectors, and verify interoperability.
Open issues, propose changes, and discuss protocol design decisions. All feedback is welcome — from cryptographers, lawyers, and developers alike.
The CPoE Lab at LF Decentralized Trust welcomes contributors. Help build the ecosystem of tools, libraries, and integrations.