AOCS Technical Basis and Validation Note

Public evidence basis, score semantics, controls and limitations for the active AOCS v1.0 reference profile.

Publication date
22 July 2026
Effective date
22 July 2026
Status
Final companion note for the active Authio reference profile

AOCS Technical Basis and Validation Note v1.0

Publication date: 22 July 2026
Effective date: 22 July 2026
Status: Final companion note for the active Authio reference profile

1. Purpose and scope

This note publishes the non-proprietary technical basis for AOCS-AUTHIO-FORENSIC-1.0, the initial active Authio reference profile for AOCS v1.0. It describes internal evidence, score semantics, controls and limitations. It is not external blind validation, external certification or an endorsement by any third party.

This note records the public evidence basis for the initial Authio reference profile within its stated scope. Future changes to a technical profile are handled through profile versioning and do not alter the final status of AOCS v1.0.

2. Reference implementation

The active reference profile uses Authio Forensic. It is a proprietary, frozen and versioned Authio detection system with deterministic full-mix resolution. It produces an authoritative full-mix HUMAN, HYBRID or AI verdict and a Canonical Full-Mix AI Score. Proprietary implementation details remain confidential.

3. Canonical score contract

The Canonical Full-Mix AI Score is a finite, unrounded value on [0,1] emitted by the active profile. Classification uses the canonical value, never a rounded display value. Higher values indicate stronger AI-oriented evidence.

For the active reference profile, the calibration target is AI presence. Human-labelled material is treated as the negative class, while AI-labelled or Hybrid-labelled material is treated as the positive class. This target does not represent mixture share or creative contribution.

Calibration is assessed within the declared evaluation population by comparing the Canonical Full-Mix AI Score with those reference labels.

The score is not a percentage of mixture share, duration, stems, instruments, labor, authorship, ownership or creative contribution.

Within an already valid Hybrid verdict, 0.50 is the normative orientation boundary defined by AOCS. It is not a legal threshold.

4. Inputs, validity and abstention

Eligible input is one complete supported sound-recording asset with immutable identity, analysis version and profile version retained.

The profile fails closed or abstains when input is corrupted, unreadable, unsupported, insufficient or incomplete. It also fails closed when required profile evidence is missing, an output is non-finite, the profile does not match or an integrity check fails. It MUST NOT fabricate a Human, Hybrid or AI class after such a failure. A materially changed asset requires a new analysis.

5. Internal evaluation program

The active reference profile is supported by a versioned internal evaluation program covering Human, AI and Hybrid material across multiple generative systems and production conditions. This is the declared evaluation population for the active reference profile.

The program evaluates classification behavior, calibration against the declared target, false-positive controls for Human-labelled material, Hybrid preservation, abstention behavior and deterministic reproduction under the declared profile conditions.

The evidence is internal. It is not an external blind holdout, external certification or third-party endorsement, and no performance outside the stated evaluation scope is claimed.

6. Hybrid and boundary evidence

The internal evaluation program includes Hybrid material on both sides of the 0.50 boundary. The public evidence does not claim invariance across transformations or execution environments outside the declared profile conditions. Materially changed assets require reanalysis under the applicable profile and version.

7. Known limitations and safety interpretation

Performance outside the stated evaluation scope is not claimed.

The evaluation program includes Human false-positive controls and AI false-negative analysis. Detailed quantitative records remain internal to Authio / Forward Digital. No claim of zero risk or universal performance is made.

A technical classification is not provenance and cannot establish the complete creation history. AOCS therefore keeps detection, disclosure, provenance review and legal or commercial policy separate.

Material Adverse Actions require documented review and appeal under the standard. Invalid, abstained, incomplete and technically indeterminate results receive Review Required status without a fabricated technical class.

8. Reproducibility and change control

The evaluation is reproducible under the recorded profile and integrity conditions.

Reproduction requires exact asset identity, profile and implementation versions and integrity controls. A later reanalysis creates a new immutable record and never silently rewrites the prior result.

Any material change to detection behavior, calibration, score semantics, precision, input eligibility or abstention behavior requires impact assessment and either a new profile version or an updated versioned validation note. Scores across profiles or versions are not comparable unless separately demonstrated.

9. Public and proprietary boundaries

Public evidence includes AOCS class semantics, evaluation scope, score semantics, supported claims, limitations, reproducibility controls and change-control requirements. Detailed quantitative evaluation records are maintained internally by Authio / Forward Digital. The Authio Forensic implementation, private evaluation assets and operational materials remain proprietary and are excluded from the public package.