Pre-Flight Twin
Pre-Flight Twin clones your live environment, simulates the implementation against it, and gates approval until every high-risk change is reviewed — with a signed audit anchor that goes stale the moment any input shifts.
The Twin is a read-only clone of your live ontology. Vault credentials are swapped for a tagged proxy; every write-shaped call is no-op shimmed.
Every prediction is scored by how it was reasoned — real provider preview API, SA graph reasoning, or operator assertion. Not a confidence number; a math one.
Seven inputs sign into a single audit anchor — checklist, ontology, guardrails, vault, findings, SA model, rule registry. Change any one, the approval goes stale.
Pre-Flight Twin pins your checklist, ontology snapshot, guardrails, and vault bindings into an immutable session — then clones the live baseline and simulates the plan against it. Materialization never writes to a provider.
Predictions group resource changes into additions, modifications, and removals — with field-level before / after diffs grounded in the live ontology. Project artifacts never appear as fake resources.
Predicted ingest of ~2.1 GB/day at current M365 audit volume crosses the Operating cost ≤ $200/mo guardrail.
Every finding ships with a structured suggestion — edit a task, add a guardrail, accept the risk. One click applies it back to the checklist with an idempotent audit log entry. No prose to parse.
Approval signs seven inputs into a single hash — checklist, ontology snapshot, guardrails, vault, findings, SA model, and rule registry. Change any of them and the approval goes stale automatically.
Coverage
Pre-Flight reasons over the providers Panaptico already maps — real provider preview APIs where they exist, structured graph inference where they don't. Each provider promotes from inference to preview as APIs come online.
Run a rehearsal. Review every predicted delta. Sign the approval and let the stale watcher catch what changes after.