Observe → act → verify → replan
Explicit success criteria, bounded tool use, persistent task state, no-progress detection and evidence-based stopping.
I audit one AI-enabled workflow, expose failure modes and evidence gaps, and deliver a bounded remediation or prototype plan. The focus is practical: state, tools, retries, human approval, observability and measurable success criteria.
No email or sales call to use the free tool. The detailed €490 audit is separately human-reviewed and not an instant download.
The work sample is CERBERRUS144, an evidence-driven agentic system. These are implemented capabilities with source control and tests — not invented client results.
Explicit success criteria, bounded tool use, persistent task state, no-progress detection and evidence-based stopping.
Candidate changes are written outside main, compiled, fully tested and committed only after the gate passes.
Traffic, email sends and checkout URLs remain separate from completed live Stripe payments.
Idempotent evidence, expiring market observations, integrity checks and one selected next action.
Repository safety, unit tests and contract tests prevent unsupported success claims from reaching main.
Stripe, Gmail, Vercel, GitHub and database evidence are handled with explicit source and side-effect rules.
Inputs, outputs, tools, users, decisions and success criteria are made explicit.
Data gaps, hallucinations, duplicate actions, retry loops, permission risks and silent failures are reviewed.
You receive a structured report and prioritized remediation or prototype plan.