Bundled with Local Regression Studio v1.0.11. These files work offline and on static hosting such as GitHub Pages.

Approval and Operation Guide

Approval and Operation Guide

Application: Local Regression Studio v1.0.11

Purpose

v1.0.11 separates model development from routine prediction. A technical user trains and validates a model, a responsible reviewer records an approval decision, and an operator loads an approved package in prediction-only mode.

This is a governance aid. It does not replace an organization’s authorization, quality, security or regulatory processes.

1. Prepare the model for review

Before approval:

  1. Use a defensible training/validation/test split.
  2. Review data-quality findings.
  3. Compare suitable candidate models.
  4. Inspect residual, uncertainty and group diagnostics.
  5. Configure measurable acceptance criteria.
  6. Confirm that the active model’s acceptance result is appropriate.

Full Approved status requires the configured acceptance criteria to pass. Approved with conditions requires written conditions.

2. Record operational documentation

The approval workspace records:

  • Model name
  • Model owner
  • Reviewer or approver
  • Decision date
  • Next review date
  • Intended use
  • Prohibited or unsupported uses
  • Known limitations
  • Approval conditions
  • Decision notes

Be specific. For example, name the supported process, sites, populations, materials, instruments, input range and decision type.

3. Approval statuses

Draft

The model is still being developed or reviewed. It cannot be exported as an approved package.

Approved

The configured acceptance criteria passed, required documentation is complete, and the review date is current.

Approved with conditions

The model may be used only under the documented conditions. This status requires a non-empty conditions field.

Rejected

The model is not authorized for operation. Record the reason and required corrective action.

Retired

The model was previously used but is no longer authorized. Retain the record for traceability.

4. Operational input schema

The app derives ranges and categories from the fitted training preprocessor. These values are read-only. Add optional units and descriptions for operators.

Examples:

Feature Unit Description
temperature °C Reactor inlet temperature
pressure bar Absolute operating pressure
material — Approved material grade

A unit recorded in the package is documentation; the CSV format itself cannot prove that a numeric value was measured in that unit. Operators must follow the documented data procedure.

5. Validation report

Choose Download validation report (HTML). The report includes:

  • Model and dataset identity
  • Dataset fingerprint
  • Intended and prohibited uses
  • Approval state and history
  • Independent test metrics
  • Acceptance outcomes
  • Operational input schema
  • Limitations and conditions

Open the HTML file in a browser and use Print or save as PDF when a PDF record is required.

6. Approved prediction package

Choose Download approved prediction package only after the model is operationally eligible.

The package contains:

  • Fitted preprocessing and model
  • Target transformation
  • Approval record
  • Validation/evaluation summaries
  • Operational input schema
  • Up to three embedded self-test input/output vectors
  • SHA-256 integrity metadata

The original training CSV is not included.

7. Prediction-only mode

On an operator computer:

  1. Start the application.
  2. Choose Prediction-only mode.
  3. Load the .mlpredict.json package.
  4. Confirm the green package summary.
  5. Upload a compatible prediction CSV.
  6. Review applicability warnings.
  7. Download predicted values.

The app verifies integrity, approval status, review date and self-tests before enabling operational use.

8. Prediction output provenance

Downloaded prediction CSV files include:

  • package_id
  • approval_status
  • review_date
  • integrity_verified
  • Applicability status and reasons
  • Prediction and interval fields

Keep these columns with operational records.

9. Package review and retirement

Before the review date:

  • Compare recent measured outcomes with predictions.
  • Investigate distribution shift and new categories.
  • Reassess group performance and interval coverage.
  • Retrain or revalidate when the data-generating process changes.
  • Issue a new package after approval.

Do not edit an approved JSON package manually. Any edit causes integrity verification to fail.

10. Security limitations

SHA-256 confirms that the package has not changed since export. It is not a digital signature and does not prove the identity of the approver. Store packages in an access-controlled location and use organizational signing or document-management controls when required.

v1.0 deployment and governance

v1.0 preserves the documented modelling behaviour and adds separate deployment editions, versioned governance lifecycle records, monitoring and revalidation, model-change assessments, local recovery, stricter import/export validation, and optional verification of organizational signatures against deployment-pinned public keys. See the package-root DEPLOYMENT_GUIDE.md, GOVERNANCE_GUIDE.md, MONITORING_GUIDE.md, SECURITY.md, and MIGRATION_GUIDE.md.