You don't open itto find the work.It already did it.
Auctorian is an always-on software worker that watches enterprise evidence, makes the recurring decisions a business makes every week, and brings you only the moments that need your authority — reasoning attached.
Not a chatbot. Not a dashboard. Not a workflow tool. A decision layer that compiles business intent into governed action, and keeps the record of why.
Good morning. $8.4M is in play across three decisions—one is leaking margin now and needs you.
3,680,000 recurring decision episodes · 612,480 ready · 38,412 consequential now
Raise SKU-10428 by 3%—priced below market with thin headroom; elastic demand holds.
Destination architecture. These are authored scenario values that demonstrate the North Star product, not measured customer outcomes.
Most of what it could suggest never reaches you.
Behind every recommendation is a search across every decision the business could make this week, narrowed to the handful that matter. The narrowing is the product.
That’s not one decision. That’s the whole surface.
One business intent compiles into 37 canonical sub-decisions across 14 domains — each with its own grain, evidence, policy and authority.
“Protect Q4 margin and availability across South India apparel. Recover excess inventory without hurting hero-SKU availability, customer promise, brand position, or cash.”
Four of the fourteen do not act. Forecasting keeps its champion and leaves the challenger in shadow; assortment defers to a boardroom; promotion suppresses itself to protect a price recovery; vendor supply is blocked outright because a factory capacity attestation expired. A system that only ever says yes is not governed.
Your policy runs first, before the intelligence.
Constraints narrow what is possible before anything is allowed to recommend. Of the candidate actions generated for this intent, most are removed by the tenant's own rules — not by a model's preference.
Six sensible decisions can still add up to a mess.
What survives policy is tested against futures that don't go to plan — and against the other decisions, because a price rise and a markdown on the same inventory are each defensible alone.
Same SKU receives price rise and markdown
Segment inventory; price hero, markdown aged tail
Promo erodes governed price recovery
Suppress offer on 38 hero SKUs
Both create supply for the same shortage
Transfer first; reduce reorder quantity
New buy competes with aged stock reduction
Defer PO and preserve markdown budget
Spend amplifies demand where promise is constrained
Shift spend to available inventory
Store allocation competes with committed orders
Reserve customer promise before allocation
You approve it as one thing. One diff, across all of it.
Not thirty-seven notifications. One reviewable enterprise diff — with each piece routed to whoever is actually allowed to sign it.
Staged, reversible, and never silent.
Governed intent becomes idempotent, observable execution against real systems — with a rollback for every action and a kill switch over all of them.
Then it measures whether it was right — and that’s what earns it authority.
Treated against control over 28 days, not a before-and-after chart. The scenario's plan modeled $4.6M and realised $4.1M: measurement that can come in under the forecast is the only kind worth having.
22 certified episodes · reversible · zero rollback failures
18 certified episodes · calibration inside policy
hold · tail-risk confidence below graduation gate
demoted · regional drift exceeded threshold
One path graduates to sovereign after 22 certified episodes with zero rollback failures. One holds. And one is demoted— regional drift pushed the forecast path back into shadow. Authority that can only ever increase isn't earned, it's assumed.
Every step kept, in order, and replayable.
The ledger is the deepest moat: 37 decision nodes, 68 edges and 14 learning edges, where each decision's outcome sharpens the adjacent ones. None of this is specific to retail.
Everything above is where this goes. Here’s what has shipped.
One complete decision episode, compiled end to end by the real engine, with proof anyone can re-check. It is advisory: nothing was dispatched to a live system, and no number on it is a measured customer outcome.
Every company will need this. Nobody has built it yet.