NATO-1000 Model Cards
These cards follow the Hugging Face model-card convention of machine-readable YAML metadata followed by transparent documentation.1 The five entries below describe runnable adapters, not claimed foundation-model weights.
NATO1000-SENTINEL
model_id: NATO1000-SENTINEL
license: apache-2.0
library_name: nato1000-runtime
pipeline_tag: text-generation
base_model: null
backend_default: deterministic-demo
Summary. SENTINEL is the systems-engineering adapter for diagnostics, maintenance planning, configuration review, and operational checklists. Intended use is offline analysis and human-reviewed engineering support. Out of scope are autonomous infrastructure changes, destructive actions, security exploitation, and unsupervised production remediation. Implementation uses the shared ModelAdapter contract and deterministic SHA-256-derived output in the default backend. Parameters include temperature, top-p, maximum output tokens, timeout, policy label, network permission, and tool permission. Evaluation is contract-level: valid prompt handling, empty-input rejection, stable schema, logging, audit events, and health reporting. No accuracy claim is made without a real backend and benchmark set. Limitations include no live system telemetry, no trained weights, no causal diagnosis, and no guaranteed correctness. Risks include automation bias and false confidence; require change review and rollback.
NATO1000-ALCHEMY
model_id: NATO1000-ALCHEMY
license: apache-2.0
library_name: nato1000-runtime
pipeline_tag: text-generation
base_model: null
backend_default: deterministic-demo
Summary. ALCHEMY is the creative synthesis adapter for ideation, alternative generation, and structured brainstorming. Intended use is to produce candidate concepts for human selection. Out of scope are treating generated content as fact, impersonation, undisclosed publication, or automatic execution. Implementation shares the runtime contract and returns auditable results. Parameters are validated by RuntimeConfig; creativity is controlled through temperature and top-p, while permissions remain separate. Evaluation checks deterministic repeatability, schema validity, error handling, and audit completeness. No originality, safety, or quality guarantee is implied. Limitations include the deterministic demo backend and lack of domain-specific training. Risks include fabricated claims, style imitation, and unsafe suggestions; label output as generated and require review.
Nadose.Daboi
model_id: Nadose.Daboi
license: apache-2.0
library_name: nato1000-runtime
pipeline_tag: text-generation
base_model: Qwen-derived reference only; weights not included
backend_default: deterministic-demo
Summary. Nadose.Daboi is the general language and reasoning adapter slot for the TWINBRAIN design. The repository contains no Qwen weights and does not redistribute them. Any deployment using an upstream model must separately comply with that model's license and terms. Intended use is controlled text analysis and generation. Implementation is backend-neutral and produces a standard ModelResult. Evaluation currently covers runtime contracts only, not language quality. Limitations include no factuality guarantee, no current-world knowledge guarantee, and no trained checkpoint in this repository. Risks include hallucination, prompt injection, and misuse; use retrieval provenance, output validation, and human review.
TWINBRAIN-DELIBERATOR
model_id: TWINBRAIN-DELIBERATOR
license: apache-2.0
library_name: nato1000-runtime
pipeline_tag: text-generation
base_model: null
backend_default: deterministic-demo
Summary. DELIBERATOR is the multi-perspective comparison adapter. It is intended to organize competing proposals and expose assumptions for human decision-makers. It must not be treated as an authority, voting substitute, or autonomous governance mechanism. Implementation preserves model identity, parameters, run identifiers, and audit-chain events. Evaluation covers valid execution and fail-closed behavior; consensus quality has not been benchmarked. Limitations include no independent truth oracle and no evidence that multiple text perspectives improve decisions. Risks include majority bias, correlated errors, and false consensus; require source diversity and an explicit dissent record.
SOULDOCTRINE-TRUTHMATRIX
model_id: SOULDOCTRINE-TRUTHMATRIX
license: apache-2.0
library_name: nato1000-runtime
pipeline_tag: text-generation
base_model: null
backend_default: deterministic-demo
Summary. TRUTHMATRIX is the provenance and evidence-organization adapter. It does not establish truth merely by generating text. Intended use is to organize supplied evidence, hashes, source metadata, and verification status. Out of scope are legal, medical, financial, intelligence, or safety-critical certification without qualified review. Implementation integrates with the hash-linked audit ledger and passive ingestion pipeline. Evaluation covers hash-chain tamper detection, redaction, deduplication, and runtime schema checks. Limitations include no independent source verification and no cryptographic proof that a source is truthful. Risks include provenance laundering and misplaced trust; preserve original sources and record uncertainty.
Shared safety and licensing statement
All adapters are configurable, but configurability does not remove responsibility, access controls, or applicable law. The original code and documentation are Apache-2.0. Upstream models, datasets, and services retain their own terms. The default runtime is offline and does not execute model output as code.