Skip to content

Latest commit

 

History

History
57 lines (32 loc) · 5.76 KB

File metadata and controls

57 lines (32 loc) · 5.76 KB

Capabilities and Boundaries - MathGov Core v12.5

Release index: Start with START_HERE_RELEASE_INDEX_v12.5.md for the shortest source map, reader lanes, claims/non-claims, and release-boundary summary. This file states what the package can and cannot honestly do today. It is a release-facing claim-boundary card, not a new gate, new standard, or replacement for the Canon.

What this package is

MathGov Core v12.5 is a Tier 1-3 source-ready specification and research architecture for ethical decision review, structured reasoning, worked-run learning, pilot design, and validation planning. MathGov is the umbrella framework. RippleLogic is the decision architecture inside MathGov.

What you can do today

Use Current status Safe wording
Learn the framework Ready Use the public intro, essentials, glossary, and examples to understand the cascade.
Review the architecture Ready Review the Markdown Canon, standards, SGP, Agent System, and validation protocols.
Run low-stakes structured reasoning Conditionally ready Use Tier 1 prompts or examples as reasoning scaffolds, with ordinary human judgment retained.
Design pilots Ready for design, not proof Use the protocols to design tests, collect cases, and prepare inter-rater reliability work.
Use the Aligners Sheet Training exemplar only The workbook is a worked-run example, not a validator, reference calculator, or proof of superior decisions.
Use PC-AEP Ready as evidence-profile requirement PC-AEP can require domain evidence and force narrow/control/redesign/escalate/refuse actions. It does not create physical proof by itself.
AI-agent governance design Specification-ready The Agent System can guide control design, but deployment still requires local security, safety, legal, and domain review.

What you cannot claim today

Do not claim that this package is empirically validated, legally certifying, deployment-certified, ProofPack-ready, Tier 4-ready, a reference calculator, a machine-verifiable conformance ecosystem, automated moral truth, universal consent, or evidence that MathGov-conformant decisions are empirically superior.

Current measurement maturity

RLS reliability, UCI/HOI measurement, SGP biological and artificial-entity scoring, TRC scenario libraries, rights-threshold governance, and PC-AEP reviewer reliability remain validation targets. The package contains study protocols and templates, not completed validation results.

Practical rule

Use MathGov Core v12.5 as a disciplined governance and decision-assurance architecture and research program. Do not use it as a substitute for law, domain engineering, clinical judgment, safety certification, democratic authority, or empirical validation.

Public release claim wording

This release is ready for public review, educational use, specification critique, and controlled validation preparation. It is not empirical validation, deployment certification, legal certification, ProofPack readiness, or Tier 4 conformance.

Publication synchronization boundary

This package can be internally release-ready while repository tags, public websites, or mirrors remain unsynchronized. A claim that v12.5 is the currently published canonical release requires an accessible tagged release, matching manifest, and matching SHA-256 set. Publication-state verification is external to the framework calculations.

Full-stack alignment boundary

MathGov currently governs socio-technical decision pathways, institutional choices, deployment decisions, and agent conduct. It can require evidence about training, internal behavior, evaluations, security, and physical execution, but it does not itself solve training-time alignment, inner alignment, mechanistic interpretability, cybersecurity, legal authority, or domain engineering. See docs/assurance/ALIGNMENT_SCOPE_AND_CONTROL_LAYER_BOUNDARY.md.

The phrase "universal ethical operating system" is a long-horizon design aspiration. The current release may be described as broadly general-purpose and union-grounded within a declared decision surface, but not as scientifically proven universal, complete for all alignment layers, or empirically superior.

Reproducibility capability

This release can mechanically check whether a supplied run record contains required objects and whether its selectability, ranking, decisiveness, authority, execution, and requalification states contradict the cascade. It cannot determine whether the evidence is true, whether a rights interpretation is legally correct, whether an instrument is empirically valid, or whether an action is safe to execute.

v12.5 integration note

WDBIP v1.5 is included as a Canon-subordinate normative implementation companion beneath RLS. ripple.md v5.4 controls its wrapper assurance surface. Run-record schema v3 separates V0 schema validity, V1 semantic conformance, and V2 evidence/use review. The two principal governing sources remain the Canon and SGP.