bickford · AI Workflow Evidence Pack

Start your evidence pack

About fifteen minutes of answers about one AI workflow, one decision-log export, and the pack lands on your delivery page within minutes: a readiness score, a pass / partial / gap matrix against ISO/IEC 42001, the NIST AI RMF and the EU AI Act, risk-ranked findings, a 30-day remediation plan, and a SHA-256 hash-chained ledger you can re-verify. See the sample report (PDF) · FAQ · Privacy and retention

Choose what to run

one workflow generated by software, unreviewed refund on request

This is a self-assessment generated from your answers and your logs. No person reviews it. It is not an audit opinion, a certification or legal advice. "Not evidenced" means your answers or logs did not show it, not that it is absent.

1. Your organization

The delivery link goes here. Nothing else is sent to it.

   

2. The AI system

3. EU AI Act scope

Decides whether EU rows are reported as obligations or as a reference benchmark. Say "unknown" when it is.

4. The workflow

The steps a decision goes through, in order. The report maps each step to the fields in your logs that prove it happened.

5. Decision logs

One or two exports, CSV or JSONL, up to 25 MB total. Export decision records only: IDs, timestamps, model version, output, score, reviewer, override, reason codes, and for a prompt-driven system the prompt ID, prompt version and prompt hash. Replace names and personal fields with references or hashes before exporting. Columns are read from the file in your browser so you can map them; the file itself is uploaded when you pay.

6. The review rule and adverse outcomes

7. Approved models and retention

Any model version found in the logs that is not listed here is a finding.

8. Prompts

Say yes if the system is driven by prompts: a language model or other generative model reads an instruction (the prompt) plus the case and produces the decision. Six prompt-governance checks then apply. For a scoring, classification or rules model they are reported as not applicable and left out of the score.

9. Practices

Answer from what you can show. "Unknown" and blanks are reported as Not evidenced, which is scored like a gap. Evidence is optional and is printed in the report next to your answer.

10. Remediation capacity

11. Pay and run

$750 for the full Evidence Pack, one workflow. Card via Stripe. Receipt from Stripe. Refund on request, no questions.

Bickford Technologies · Philadelphia, PA · derek@bickfordtechnologies.com