Take command

Scorta · Inventory

See what it would have done.

Before you let it act on its own, replay the last ninety days of your real decisions against the policy you're considering. Not a case study from someone else's warehouse — your ledger, your calls, and exactly what it would have done without you.

See it on your stock →What it does, step by step
YOUR LAST 90 DAYS, AS THEY HAPPENEDconfirmReorder · MAPLE-ENG-12 · ~$17,328you approved itapprovePurchaseOrder · PO-G0N3VJyou approved itflagSupplierHold · Ridge Freightyou approved itreplayed against a policy you have not grantedIF confirmReorder HAD BEEN SET TO ACT + ATTEST2of those would have run without you1would still have stopped at your desk~$17,328capital that would have movedEvery AI vendor asks you to trust it first and find out afterwards.
26named operations, each with its own ceiling
8that can never be automated at all
3tiers: propose, act and attest, autonomous
60 minundo window, where undo is real
P10–P90bounds on every projection
12modules on one platform

These are things Scorta does — not results a customer measured.

01

The dial, per operation

Twenty-six jobs, twenty-six ceilings.

Autonomy isn't one switch. Reordering, approving a PO, transmitting it, quarantining a lot, applying a markdown, posting a variance — each is a named operation with its own setting. Move one up when you've seen it be right, and leave the rest where they are.

Three levels: propose and stop, act and then show its work, or run inside the scope you set. Nothing jumps a level on its own.

OPERATIONNOWCAN BE RAISED TOconfirmReorderproposeact + attestapprovePurchaseOrderproposeact + attestprioritizeFefoPickproposeautonomous + auditacknowledgeAlertproposeautonomous + auditNO UPGRADE EXISTS FOR THESEquarantineLotpropose— nothingdisposeExpiredLotpropose— nothingsetStockPolicypropose— nothingNot "off by default". There is no setting. Ask your risk committee which answer they'd rather have.

02

The shift board

Seats that work, and seats left empty.

Demand sensing and replenishment. Rebalancing. Supplier performance. Each is a seat with a tier, a remit and a running count of what it has proposed, what it executed, and what is sitting waiting for you.

And six operations have no seat at all — marked human by design on the same board. The product doesn't leave that implied; it prints it where you can see which is which.

ON SHIFTDemand Sensing & ReplenishmentT1proposed · executed · waiting on you — counted, per shiftInventory RebalancingT1moves stock between positions before a shortfall becomes an orderSupplier PerformanceT1watches on-time rates and average slip, by supplierUNSTAFFED — 6 operationshuman by design · no seat proposes here

03

Plans

A plan is only as good as its weakest input.

When three different seats notice related things, they resolve into one sequenced plan — hold the supplier who's slipping, then expedite the reorder that depends on them. You ratify the whole plan, or step by step, and a failed step rolls the rest back.

The plan's confidence is the lowest of its contributors, not the average. Arithmetic that makes a weak input impossible to hide behind two strong ones.

Slippage response — Ridge Freighthold the supplier, expedite the exposed reorder · ceiling T2AWAITING RATIFICATION1Place supplier on watchon time only 72% of the time · average slip 4.5 daysflagSupplierHold2Expedite the reorder that depends on themreorder before the slip becomes a stockoutconfirmReorderJoint confidence — the minimum of the contributors, never the average.Ratify as a unit and a failing step rolls the others back. Or take them one at a time.

04

The ledger

Undo, and the exact edge of it.

Some things can be taken back. A surplus transfer, a reserve extension, a risk flag — reversible for sixty minutes. The undo control appears only where the system would genuinely accept it, so it's never a button that fails.

A placed purchase order is never silently reversible, and the product says so rather than letting you find out. Every decision, human or agent, lands in the ledger with who, when, which tier and how much.

UNDO APPEARS HEREsurplus transfersreserve extends and allocationsrisk flags60 minAND NOT HEREa placed purchase ordernever silently reversible —cancel it properly, or not at allAND EVERYTHING LANDS HEREAPPROVEapprovePurchaseOrderT1~$17,328PO-G0N3VJTRANSMITtransmitPoT1immutably attestedWho, when, which tier, how much — and the order it produced.

Why this instead of what you have

Everyone selling you AI asks you to go first.

Turn it on, watch it for a while, see how you feel. In a warehouse that's a strange thing to be asked, because the mistakes have prices — a wrong reorder is working capital, a wrong disposal is stock you can't get back, and a wrong adjustment is a line an auditor will want explained.

The pilot proves nothing you can act on.

Three months of it suggesting things while a person approves everything anyway. You learn it's usually right, which you already suspected, and nothing changes.

Autonomy is one switch.

So it's off, because the reason it's off is one operation nobody will ever let it do, and that decision holds all twenty-five others hostage.

Nobody can answer the risk question.

"What stops it doing the dangerous thing?" A policy document, a promise, and a setting that someone with admin rights can change on a Friday.

Replay it against your own history before granting anything. Promote one operation at a time. And for the ones that should never run alone, there is no setting to find.

05

Assure

It prepares the record. You sign it.

Inventory adjustments are where shrinkage hides, so this screen has no agent seat on it at all. AEGIS can assemble the evidence. It can never write the signature — the who, the when and the why are a person's, by construction.

Adjustments group by reason code with net quantity and absolute quantity shown separately, so a hundred movements that cancel out can't disappear into a net of zero. And there's a violations table whose entire job is to stay empty.

REASONCOUNTNET QTYABS QTYVALUECycle-count adjustment1−2424−$912Expired1−1111−$704Net and absolute kept apart — offsetting movements cannot hide volume.AEGIS MAYassemble the recordgather evidence, propose, prepareONLY A HUMAN MAYsign itdecided by · decided at · reasonSEAT NONE · AEGIS PREPARES · A HUMAN SIGNS

06

Project

A median is never certainty.

Project any SKU forward thirty, sixty or ninety days over your real position ledger, and get back a range rather than a number — the tenth percentile, the median and the ninetieth, with the bounds always on screen.

Then ask it the harder question: given a service level you need and a carrying cost you'll tolerate, what reorder point, order-up-to level and safety stock actually sit on that frontier? It proposes policy, not just a forecast you still have to interpret.

TODAY60d90dP90P50P10ON-HAND, PROJECTED — BOUNDS ALWAYS SHOWN

07

Five reasons it wins

Nobody else lets you test the trust before you give it.

01

Replay it against your own history.

Take the last ninety days of decisions you actually made, run them against a policy you're considering, and see exactly which ones would have gone ahead without you — and how much capital that was.

02

Some things have no autonomous setting.

Quarantining a lot, disposing of expired stock, setting stock policy, signing an adjustment. Not off by default — there is no level to raise them to. Your risk committee can look at the screen.

03

An agent can never sign.

On adjustments there is no seat at all. AEGIS assembles the evidence; the signature that records who, when and why is a person's, and the screen tells you which permission you're missing if it isn't yours to give.

04

Undo tells you where it ends.

Sixty minutes on the reversible things, and the control appears only where it would genuinely work. A placed purchase order is never silently reversible, and you learn that from the screen, not from trying.

05

It gives you the range, not a number.

Every projection carries its tenth and ninetieth percentile alongside the median, because a median presented alone is a forecast pretending to be a fact.

08

What's in it

Six areas, one module.

Autonomy

6
  • Twenty-six operations, each with its own ceiling
  • Propose, act and attest, or autonomous
  • Eight with no autonomous level at all
  • Replay real decisions against a policy
  • What would have run, and the capital in it
  • Repeated overrides offered as defaults, once

The shift

5
  • Named seats with a remit and a tier
  • Proposed, executed and waiting, per seat
  • Six operations marked human by design
  • Per-seat precision, graded from the ledger
  • Threshold raises only through ratify

Plans

5
  • Several findings into one sequenced plan
  • Ratify as a unit, or step by step
  • A failed step rolls the others back
  • Confidence is the weakest contributor
  • Each step names the operation it calls

The ledger

5
  • Every decision immutably attested
  • Who, when, which tier, how much
  • The order it produced, linked
  • Sixty-minute undo on reversible actions
  • Shown only where undo really works

Assure

6
  • No agent seat on this stage at all
  • A signature is never an agent's
  • Who, when and why on every decision
  • The exact permission you're missing, named
  • Net and absolute quantity kept apart
  • A violations table meant to stay at zero

Project

5
  • Forward projection over the real ledger
  • Thirty, sixty or ninety days
  • Tenth, median and ninetieth, always shown
  • Reorder point and safety stock proposed
  • Search the service against carrying cost

Thirty-two of them. There are more, but these are the ones that decide whether you'd let it run.

Next step

See it on your stock.

Ninety minutes on your operation: your SKUs and positions, your suppliers and their slip, which operations you'd let it propose first, and which ones you'd never let it touch. You keep the configuration.

Request a briefing →Every discipline in The Hive

Scorta is the ELYON inventory module. Every capability here is live in the product today. No figure on this page is a customer outcome.