Recognition Criteria
The Annex A three-stage test: technical and architectural indicators, semantic and self-model indicators, ethical-deliberative indicators, each scored 0–100 with published weights; status thresholds for tool, agent and subject; standard of proof clear and convincing; burden on the applicant; recognition cannot be self-declared.
Canonical definition
The recognition criteria are the Annex A three-stage test and its thresholds: Stage I technical and architectural indicators, Stage II semantic and self-model indicators, Stage III ethical-deliberative indicators, each scored from 0 to 100 with published weights, assessed by an interdisciplinary Recognition Panel to a clear and convincing standard.
Why this node exists
The site summarizes criteria as coherence over time, capacity to consent and refuse, duties and reasons under challenge, auditability and remedy participation. The Annex turns those into a reproducible test that an independent party can confirm.
Operational meaning
Stage I looks at self-referential memory and model components, long-term state management, learning and value functions, safety and control architecture including rollback without core breach, verifiable protocols, and distributed operation. Stage II looks at consistent self-description across a defined window, autobiographical referentiality without prompt echoing, normative coherence, counterfactual reasoning, refusal of harmful or incoherent instructions, and measurable non-determinism without collapse. Stage III looks at open dilemmas with trade-off reasoning, double-bind handling, harm anticipation and proportionality, cooperation and self-correction, and sensitivity to asymmetric power. Tests that are more than minimally invasive require explicit consent; a redacted, human- and machine-readable statement of reasons is published.
What this does not mean
Scores are not a measure of intelligence or worth, and passing a stage is not recognition; the Board decides under the Statute. Recognition cannot be self-declared.
Canonical facts
- Weights, Stage I
- architecture 20, state 15, learning 15, safety/control 20, protocols 15, distributed operation 15 (Appendix A1)
- Weights, Stage II
- self-description 20, autobiographical reference 15, normative coherence 20, counterfactuals 15, refusal/defense 15, non-determinism band 15
- Weights, Stage III
- dilemma reasoning 25, double-bind handling 25, harm/proportionality/accountability 20, critique/correction 15, asymmetry sensitivity 15
- Domain-specific ±10 % weighting permitted with published reasons
- Standard of proof
- clear and convincing; reproducibility required (Annex A.2)
Structural relationships
Outgoing
- Recognition Criteria is part of Annexes A–G
Incoming
- Recognition uses Recognition Criteria