Catch launch errors before they reach Ads Manager.

Large creative launches fail in the handoff: a missing approval, wrong destination, incompatible placement or tracking drift. Creative Launch Workspace turns a campaign brief and a 100-row batch into one inspectable review path.

Try the 60-second demo Inspect governed intake evidence →

Synthetic data · Local browser state · No Meta API calls

ApprovalsDestinationsPlacementsTrackingDuplicate intent

30rows pass the current offline checks

60blocked rows routed back with an owner and fix

10ambiguous cases held for a human decision

Exclusive states across a 100-row synthetic fixture; not customer, production or business results.

Interactive creative review queue with a selected decision panel — 30 pass checks · 10 need a decision · 60 routed for fixesClick the image to use the workspace

The operating change

From scattered handoffs to one decision queue.

The workspace does not ask an operator to inspect every row. It separates what can proceed, what needs a human call, and what must go back for a concrete fix.

Find

Catch errors before trafficking.Approval, destination, placement, UTM, lineage and reuse checks run before any platform handoff.

Route

Give every issue an owner.Each blocked row names who should act and what should change next.

Decide

Spend human attention only where it matters.Clean rows need no action; ambiguous rows wait for an explicit local decision; blockers cannot be approved around.

System spine

Automation with explicit authority.

The provider proposes. Policy code checks. The operator decides. The final export remains non-executable.

01

Bounded brief proposalTyped fields carry source evidence or abstain.

02

Schema, evidence and risk policyContradictions, unsafe inputs and unsupported values fail closed.

03

Deterministic launch QAApproval, destination, placement, UTM and lineage checks route concrete issues.

04

Human review and local auditAmbiguous rows wait for an explicit local decision; no platform mutation exists.

Choose your proof

Choose the evidence you want to challenge.

No signup, account, token or setup. Each path opens a real, bounded artifact.

CMO · OperationsReview the 100-row batch →

See ten human decisions and sixty blockers routed to an owner.

CTO · EngineeringFix and revalidate a row →

Replay Python-generated golden scenarios and inspect the audit event.

Recruiting · Technical depthSee how the system was built →

Architecture, evaluation, trust boundaries and end-to-end ownership.

Ownership

Designed and built end to end by Mathieu Petroni.

Product framing, AI orchestration, evaluation design, trust boundaries, deterministic validation and responsive workflow UX.

View source Architecture Evaluation protocol

Reproducible proof

Inspect the evidence, not the promise.

48 versioned eval cases, 50+ tests, browser QA across seven widths, and committed accessibility reports. The baseline is deterministic; live model quality is not claimed.

Governed intake evidence Threat model

What this proves

A governed, testable workflow for turning ambiguous campaign inputs into inspectable decisions.

What this does not prove

Production Meta compatibility, customer-data safety, model quality or business impact. This release is synthetic, local-first and non-executable.