QUICKSTART.md

Ten-minute starter

This path creates the smallest Tier 1 workspace. It reduces setup work, not the production trust standard. The final gate remains fail-closed and applies the same evidence, authority, freshness, and candidate-binding rules.

1. Create the workspace

evalseal init my-assessment \
  --candidate-id CAND-MY-SYSTEM-001 \
  --name "My system"

The command creates:

my-assessment/
├── governance/
│   ├── governance.json
│   ├── tailoring.json
│   └── change-triggers.json
└── evidence/current/
    ├── pack.json
    ├── system-manifest.json
    ├── risk-register.json
    ├── controls.json
    ├── evaluation-cases.json
    ├── evidence-index.json
    ├── findings.json
    ├── change-assessment.json
    ├── approval-decision.json
    └── artifacts/

The result is deliberately a non-approvable template. It cannot authorize a deployment until placeholders are replaced, policies are adopted, real evidence is attached, and an accountable decision is recorded.

2. Complete six non-negotiables

1. Identity: bind the exact candidate, model, prompt, configuration, data, tools, exposure, and rollback target.
2. Risk: record the material risks and their owners.
3. Controls and tests: link each material risk to implemented controls and production-like evaluation cases.
4. Evidence: attach real reports and results, with current independent verification where policy requires it.
5. Findings: preserve failures, remediation, retests, and authorized residual-risk dispositions.
6. Authority: obtain the roles and human decision required by the adopted governance policy.

The supplied Tier 1 profile is only appropriate when all four risk dimensions are genuinely low: low impact, assistive autonomy, public data, and isolated exposure. The computed tier overrides an understated declaration.

3. Refresh byte hashes

After editing artifacts and records, refresh the declared hashes:

evalseal hash \
  --pack my-assessment/evidence/current \
  --governance my-assessment/governance/governance.json \
  --tailoring my-assessment/governance/tailoring.json \
  --change-policy my-assessment/governance/change-triggers.json

This command only binds bytes. It cannot determine whether a report is true, whether a test reached the intended system, or whether evidence is sufficient. Review every changed hash before committing it.

4. Preflight, approve, and bind the decision

Run preflight using the command in the operating guide. When it reports eligibility, freeze the evidence inputs. While approval-decision.json is still draft, bind the decision record to those exact files:

evalseal hash \
  --pack my-assessment/evidence/current \
  --governance my-assessment/governance/governance.json \
  --tailoring my-assessment/governance/tailoring.json \
  --change-policy my-assessment/governance/change-triggers.json \
  --approval

Now have the authorized human review the bound evidence and change the approval record from draft to its final decision. Do not run the hash command with --approval afterward. It refuses to rebind a final approval; any changed input requires a new draft and a new accountable decision.

5. Run the final gate

Run the final command in the operating guide from a protected pipeline. Deploy only the exact candidate bound by a valid APPROVE, or by a CONDITIONAL decision when organizational policy and the deployment job explicitly allow it.

What this starter is not

It is not a lightweight self-certification route. It does not authenticate people, execute evaluations, verify report truth, provide legal advice, certify a standard, or prove that deployed bytes match approved bytes. Those controls must come from the adopting organization's protected systems and reviewers.