Good questions.
Useful answers.

numericspace is a proposed AI enactment controller: a layer for selecting permitted enactment paths, rechecking the current policy at supported action boundaries, and recording the decisions and costs associated with the workflow.

The website is an interactive concept preview. The Platform page explains the design and its enforcement boundaries.

You can run three simulated scenarios, inspect and export sample evidence, compare proposed plans, calculate hypothetical value, and prepare a pilot brief. You can also explicitly submit the completed inquiry for owner review. The enactment demo does not connect to live models, cloud accounts, or private workers.

The proposed role is enactment control across selected models and workers. A pilot would identify which local environments and external endpoints can be supported. Underlying model access, hosting, and compute remain separately scoped.

They are future research directions. The initial pilot focuses on a defined workflow, policy checkpoints, routing decisions, and enactment evidence. Research can later investigate how eligible resources are allocated and how incentives affect their behavior.

Rules can change after a workflow starts. The proposed controller checks the applicable current policy before the next supported dispatch, retry, transfer, or result release. A record links the action to the policy version used for that decision.

A policy change can prevent a future action governed by the controller. It cannot reverse a completed action or recall data already transferred. Cancellation of external work depends on provider support and requires confirmation.

The sample provider has accepted the request, but its completion and cancellation are unconfirmed. The demo keeps the reserved cost exposure visible, holds the next protected action, and avoids assuming a duplicate request is safe.

This is an illustrative failure condition to evaluate during a pilot.

The sample evidence describes enactment decisions. It is not a certification, legal conclusion, or independent assurance report. A production evaluation would need to define coverage, integrity, retention, access controls, and the requirements that the evidence is meant to support.

Prices are proposed planning figures in USD. The website does not take payment or create a subscription. A commercial agreement would separately define availability, scope, fees, acceptance criteria, and support.

The planning scope is one workflow and up to two enactment endpoints over 4–6 weeks. The objective is to establish a baseline, exercise an agreed policy change, and review the resulting evidence. Final deliverables and exclusions would be agreed before kickoff.

Underlying model and compute charges are separate. Ongoing subscriptions and any implementation beyond an agreed scope are also separate. The calculator subtracts the selected subscription fee; its output is not a complete project-cost model.

No annual discount is assumed. The annual equivalent for Team and Business is twelve times the proposed monthly price. The toggle supports comparison and does not create a billing commitment.

Reviewing the brief prepares a local planning draft. Submission is a separate action: review the complete information, agree to the consent statement, and choose “Submit pilot inquiry.” A confirmed submission stores the inquiry for review; it does not book a meeting, take payment, or accept a service agreement.

A pilot draft is stored in this browser only when you choose to save it. Resume and delete controls are explicit. A saved draft may remain accessible to another person using the same browser profile. Deleting the draft or resetting the planner does not delete a server-submitted inquiry, downloaded files, or copied text.

Define the permitted workflow and policy transition, then agree on observable acceptance criteria. Examples include blocking the next prohibited action, linking decisions to policy versions, identifying uncertain outcomes, and separating confirmed charges from outstanding cost exposure.

The downloadable policy JSON is an illustrative record for discussion. It is not a published API contract, validated configuration format, or deployment package. A production interface would need its own versioned specification and validation.

The server stores your contact and organization details, workflow, policy-change scenario, success criteria, endpoints, environment, plan, budget, and full submitted pilot brief. It records a reference, submission time, and the consent version and server timestamp.

The owner can add private review notes and a status. These are separate from the original submitted brief.

The application protects its private inbox, review actions, and owner exports with authenticated ChatGPT identity and a verified owner allowlist. The owner can review, print, or download the submitted record and update internal notes and status.

The website does not publish your inquiry as a customer example or research result.

No automated email notifications or confirmation emails are configured. After the server accepts your submission, the page shows its reference and confirmed brief. Keep that record. The owner may use your provided contact details to follow up about the inquiry, but submission does not promise a response time or a booking.

Submitted content, the full brief, and review notes remain in the application database until the owner deletes the inquiry. Keep your reference when discussing a correction or deletion with the owner; there is no public self-service deletion control yet.

Deletion does not remove existing exports. A request-key-only anti-duplicate record expires after 30 days and is cleaned up on subsequent submissions. The privacy notice also explains the separate, expiring rate-limit records.