aibvf-check

aibvf-check

SonarQube for AI initiatives. A CI/CD pre-flight gate that fails the build when a declared AI initiative would not survive a board review — using the deterministic AI BVF v1.0 engine.

The scoring belongs upstream of the slide deck. aibvf-check puts it in the one place a team already trusts to block bad work: the pipeline.

30-second use

Add a manifest at your repo root, .aibvf.json:

{
  "bvf_version": "1.0",
  "organization": { "industry": "financial", "revenue_eur": 400000000 },
  "policy": { "fail_on": ["Stop"], "max_governance_risk": 60, "min_decision_confidence": 50 },
  "initiatives": [
    { "id": "cx-genai-assistant", "function": "cx", "ai_tier": "gen2", "readiness": "traditional",
      "scores": { "strategic_alignment": 70, "financial_return": 55, "change_enablement": 55, "governance_risk": 45 },
      "signal_completeness": 0.8 }
  ]
}

Run it:

npx aibvf-check

It scores every initiative, prints a scorecard, and exits non-zero if any trips the policy — so CI goes red.

GitHub Action

- uses: Bahamas1717/ai-bvf@main
  with:
    manifest: .aibvf.json   # optional, this is the default

The step fails the workflow on any policy violation.

Policy

Field	Effect
fail_on	Classifications that fail the gate. Default ["Stop"]. Add "Fix" to be strict.
max_governance_risk	Optional hard ceiling (0–100). A higher governance_risk fails, regardless of verdict.
min_decision_confidence	Optional floor (0–100). A lower decision confidence fails.

signal_completeness (0–1, per initiative) forwards to the engine: estimated inputs honestly haircut confidence and surface a caveat, so a low-evidence initiative can't sail through on a confident-looking number.

Exit codes

Code	Meaning
0	All initiatives cleared the policy.
1	One or more policy violations — the gate failed.
2	Config error (manifest missing, malformed, or a bad enum value).

Options

aibvf-check [--manifest <path>] [--json]
  -m, --manifest <path>   Manifest file (default: .aibvf.json)
      --json              Machine-readable JSON instead of the table
  -h, --help              Help

Deterministic, no network, no telemetry. Same inputs → same verdict, every run. Built on @aibvf/core.

License

MIT.