How planned Macrovision would work

From a question to a preserved human decision.

The intended journey is question → experiment → actual evidence → material change → human review → preserved decision. The free Laboratory is available today. Customer data imports, monitoring, AI preparation and organisational approvals are planned.

See a Macrovision walkthrough Explore the free Laboratory

One complete journey

Each stage makes the next review easier to explain.

The paid workspace is planned to reduce repeated evidence gathering, exception preparation and organisational coordination. Customer-specific data flows and permissions would require explicit agreement before activation.

The first stage is available; the customer-specific stages describe the planned workspace.
Stage	Purpose	Required control
1. Test the question	Run an illustrative scenario in the free Laboratory.	Inspect assumptions, model version, result and limitations.
2. Confirm actual evidence	Validate a controlled import of cash, obligations, receipts and restrictions.	Confirm entity, currency, dates, time zone, mappings, duplicates and missing records.
3. Review material change	Compare approved evidence with the latest monitoring state.	Open an exception only when agreed review rules warrant attention; route routine changes to a digest.
4. Prepare the review	AI proposes source-linked inputs, grounded explanations and a draft document.	Keep changes reviewable; approved models supply the numbers.
5. Decide with authority	The responsible people review gaps, alternatives and rationale.	Apply effective roles, policy and segregation of duties.
6. Preserve and revisit	Freeze the approved evidence, actual AI output, model results and decision.	Retain subsequent reviews separately; later observations do not rewrite prior approval.

Specific AI jobs

AI prepares and explains. Approved models calculate. Your team decides.

AI would prepare reviewable fields, explain validated changes and draft the existing decision document. Extraction confidence describes a reading task; evidence completeness describes available support; neither is a probability that an obligation will be met.

Documents remain evidence and cannot grant permissions. Approved assumptions cannot be silently overwritten. The record must preserve the output and context seen by the reviewer. Forecasts of customer payment behaviour would require separate validation before any predictive claim.

Controls behind the experience

Each kind of authority has one owner.

The planned workspace uses explicit contracts to make evidence, economic tests, workflow responsibilities and connector permissions inspectable. Technical reproduction does not establish customer consent, business correctness or regulatory approval.

Contract architecture is demonstrated with synthetic fixtures and does not imply an available customer service.
Record	What it would make explicit
Evidence Contract	Admissible sources, freshness, knowledge cut-off, limitations and missing-evidence rules.
Decision Contract	The declared economic test, assumptions, policy thresholds and materiality.
Treasury Operating Contract	Responsible roles, steps, deadlines, exceptions, review and reopening.
Connector Contract	Named permitted operation, credential reference, retry, revocation and reconciliation.
Authority Contract	Who may ratify, activate, review, approve, override or withdraw.
Decision Packet	Frozen evidence, exact model identities, receipts, rationale and approval-time record.

Inspect the research-only v4 fixture

Future activation gates

Learn it. Test it. Ratify it. Run it. Replay it.

A source-linked candidate would need schema, graph, safety, replay and conformance checks before customer ratification. Shadow operation would run alongside existing processes; later activation could permit only named, bounded operations under current authority.

Planned writes: Draft documents, notifications, tasks, approved-folder archival and approval requests, only if expressly authorised.
Boundary: Payments, trades, custody and asset transfers remain excluded.
Change: Material evidence or policy changes would open a new review without altering the earlier approval basis.
Confidentiality: Defined purposes, permitted AI use, controlled access and contractual retention are prerequisites for customer data.

Discuss a planned recurring review without sending financial records or granting production access.

YieldGuard Ltd, company number 16914415, registered in England and Wales. Macrovision is planned and not accepting paid engagements. Contract and replay artefacts establish disclosed technical behaviour; they do not substitute for legal, regulatory, risk or approval obligations.