Skip to main content

AI Guardrails

AI Guardrails is a suite of project-level checks that grade the health of your engineering artefacts and surface gaps before a release. Each check runs on demand and stores the most recent result for fast retrieval.

The suite includes:

  • Project Health Score — a single, fast SQL-based score covering coverage, traceability, and basic data quality
  • AI Quality Score — per-requirement and project-wide AI scoring of requirement clarity and completeness
  • FMEA Gap Analysis — AI scan for missing failure modes given the current requirement set
  • Release Readiness — gated checklist that combines the above signals into a release recommendation
  • AI Change Review — AI commentary on the impact of a requirement edit
  • AI Traceability Suggestions — suggested links between requirements / FMEA / tests that look related
Subscription Required
  • Project Health Score: available on all tiers (pure SQL, no AI calls)
  • AI Quality Score, FMEA Gap Analysis, Release Readiness, AI Change Review, AI Traceability Suggestions: Advanced tier and above
  • Risk Discovery, Compliance Gap Analysis: Power tier and above
  • Cross-project Risk Patterns: Enterprise tier

Project Health Score​

A fast snapshot of project hygiene that does not call any AI provider — useful for daily standups and dashboards.

It looks at:

  • Number of requirements with no parent and no children (orphans)
  • Number of requirements with no test coverage
  • Number of requirements with no linked FMEA
  • Number of FMEA failure modes with no requirement link
  • Number of requirements stuck in Draft for more than N days
  • Number of failure modes whose RPN exceeds the project's risk threshold without a mitigation action

The result is a numeric score (0–100) plus a per-dimension breakdown. Open it from the Analytics → AI Guardrails view.

AI Quality Score​

Asks the configured AI provider to grade each requirement on:

  • Clarity — is the requirement unambiguous?
  • Verifiability — can it be tested?
  • Atomicity — does it cover exactly one behaviour?
  • Necessity — is it justified by a parent requirement or a stakeholder need?

You can score one requirement at a time from the requirement detail panel, or batch-score the whole project from Analytics → AI Guardrails → Quality Summary.

The score is cached against the requirement's current text — re-scoring is only needed after the text changes.

FMEA Gap Analysis​

Scans your project for failure modes that are likely missing given the current requirement set. The AI is given:

  • Each requirement's title, type, and safety classification
  • Existing FMEA failure modes for context (so it does not duplicate)
  • Relevant Knowledge Base documents (if KB is populated)

Output is a ranked list of suggested failure modes with:

  • A description
  • The requirement(s) it relates to
  • A rationale
  • Suggested Severity / Occurrence / Detection starting points

You review and accept what is genuinely missing — accepted items are added to the appropriate FMEA analysis with traceability links pre-filled.

Release Readiness​

A gated checklist that combines several SQL gates (no orphan requirements, no FMEA without a requirement link, no high-RPN failure modes without action items) plus AI signals (quality scores, gap analysis) and produces a single Ready / Not Ready verdict for the project.

You can:

  1. Run it on demand
  2. View the cached most-recent result without re-running
  3. Generate a saved snapshot to attach to a release record

Use it before milestone reviews or formal handovers.

AI Change Review​

When you edit a requirement, NirmIQ can ask the AI to comment on the likely impact of the change — for example, which downstream FMEA failure modes may need updating, or which test cases may now be invalidated.

This is a review aid, not an automatic edit. Nothing is changed without your explicit action.

AI Traceability Suggestions​

Looks across the project's requirements, FMEA failure modes, and test cases, and proposes links the AI thinks are missing — for example, a failure mode that mentions a sensor but is not linked to the sensor-monitoring requirement.

Each suggestion has a confidence level. Accept what looks right; ignore the rest.

How to use the suite​

  1. Open Analytics → AI Guardrails inside any project
  2. The first tile is the Project Health Score (always available)
  3. The other tiles trigger the AI checks; results are cached for fast re-display
  4. Use the per-tile Re-run button after substantial changes to the project
  5. Snapshot the Release Readiness result before handing a project off

Privacy and cost​

  • AI checks send the relevant artefact text plus the KB context to your configured AI provider. The full project is not uploaded.
  • Each AI check counts against your monthly AI usage allowance for the tier (or against your AI add-on quota where applicable).
  • Results are stored encrypted at rest in your organisation's project record.