Model boundary
A model may draft, classify, retrieve, compare, test, or explain. Its output remains a candidate until another authority accepts it.
Serapis is designed so fluent output cannot quietly become authority. Models produce candidate work; bounded mechanics, identity, evidence, and named people determine whether an effect is eligible.
These principles guide the product. The exact controls, deployment profile, and evidence required for a customer environment remain part of the written engagement and customer approval process.
A model may draft, classify, retrieve, compare, test, or explain. Its output remains a candidate until another authority accepts it.
Versioned mechanics evaluate bounded predicates and return 0, 1, or UNKNOWN; persuasive language cannot override the result.
Missing, contradictory, stale, inapplicable, or unverifiable evidence holds visibly rather than being treated as a pass.
The requested action, evidence producer, evaluator, reviewer, and authorizer remain distinguishable roles with bounded scope.
The exact candidate, sources, rule revision, result, identity, decision, and effect travel as one inspectable record.
Customer-selected models, keys, data, rules, compute, and authority remain inside the approved profile and portable-exit agreement.
Standards guide evidence work; they are not product badges. Serapis can help structure a bounded evidence path, but it cannot certify a contractor, grant an authorization, or accept risk for the customer.
CMMC status belongs to a specific contractor information system and contractual requirement. Evidence, gaps, owners, assessment dates, and affirmations must remain bound to that scope.
Current CMMC evidence uses Revision 2 while NIST has published Revision 3. Catalogs and mappings must identify the exact revision instead of blending requirements.
Continuous evidence, traceable releases, control gates, and risk views can support an authorization workflow. ATO and cATO decisions remain with the designated Government authorization authorities.
A screen that works is evidence of a screen. It is not proof of a live connector, operational suitability, or continuing authority.
Marketing explains the design, team, and proposed path. It is not a security attestation, binding proposal, or proof of customer deployment.
Synthetic records can prove the visible workflow and receipt pattern. They have no customer-system or mission effect.
Customer-authorized access and acceptance criteria establish whether identity, data mapping, rules, usability, and deployment fit the selected environment.
Production use is deployment-specific. It requires the customer’s approved architecture, security review, identity and access design, data handling boundary, operational procedures, testing, and acceptance.
Serapis will not infer those approvals from a successful demonstration or from controls that exist in another environment.
The replacement test is architectural: can the customer preserve its facts, decisions, rules, and operating history if a model, cloud, database, or Serapis itself changes?
Where portability is limited by a selected dependency, that limitation should be visible before adoption—not discovered during exit.
The public marketing pages do not host product authentication, customer data, an application API, operational connectors, analytics pixels, advertising trackers, or a contact form.
Email links open the visitor’s own mail client. Hosting and security providers may process ordinary request data required to deliver and protect the site.
If a capability cannot be demonstrated or bounded yet, the right answer is UNKNOWN and a concrete path to learn—not a stronger adjective.