deep-finance-isolation-forest-fraud(automation) · Curator Automations

NAME

deep-finance-isolation-forest-fraud — "This order — $1,200 from a buyer averaging $40, shipping to a new country, card flagged for prior chargebacks — review before fulfilling."

SYNOPSIS

id deep-finance-isolation-forest-fraud tier deep-purple section §22. Finance — AI/subsystem + deep-purple cluster (37) trigger event (per order) shape SINGLE

DESCRIPTION

Throw every order onto a random-decision-tree obstacle course; the orders that get isolated fastest are the weird ones. No labeled fraud data needed — the algorithm just notices outliers.

"This order — $1,200 from a buyer averaging $40, shipping to a new country, card flagged for prior chargebacks — review before fulfilling."

MECHANISM

not-yet-implemented — no cart file at curator-web/src/scheme/carts/…/deep-finance-isolation-forest-fraud.sks

EXAMPLES

Sakura uses this on every incoming order, quietly. Isolation forests are good at "this point looks unlike the others" without you defining what fraud means. "$1,200 from a buyer who usually spends $40, new country, card has prior chargebacks — review before fulfilling."

TIER

deep purple

deep reasoning · batched · dollars per run

FITNESS

Not yet

This method only gives a real answer above a sample threshold. Running it on a small shop produces a confidence band the size of the universe — useless. When the shop crosses the threshold, this lights up.

HISTORY

canonical source: curator/docs/AUTOMATIONS-CANONICAL.md · §22-finance-aisubsystem-deep-purple-cluster-37
last regenerated: 2026-06-05 20:31 UTC (mechanical — edits land in the MD or sidecars, not here)

KNOWN BUGS

(none recorded — flag in AUTOMATIONS-CANONICAL.md or open an issue against the cart file)