LoopXReproduce with real Agents

Interactive case · browser-local simulation

The request changed.
Did the result?

A received message can leave yesterday’s plan intact. Change one constraint, find the broken allocation, and carry a checked result back to the owner.

You operate the steps. No Agents run and no LoopX state is changed.

Follow one correction

01 · Owner → builder

Change the brief

Allocate six orders. Maximize value within shared stock, regional capacity and a 1,000-cent budget.

Keep in each stock group0
Minimum East allocation0

Current request v1

Try another request

Each change creates a new version. Prior checks stay visible but cannot accept the new input.

02 · Receipt

Receipt is one fact

Receiving a request does not update an artifact.

03 · Builder’s artifact

Inspect the actual allocation

Plan input v1
Editable quantities in order-id order; costs are cents per unit
OrderGroup / regionDemandCost / valueAllocate
ax / East4110 / 9
bx / West590 / 7
cy / East3140 / 12
dy / West4120 / 10
ez / East030 / 20
fz / West370 / 6
Value 84Cost 1000 / 1000Stock x 6 · y 5 · z 4Region caps East 5 · West 7

Loads a published example, not a model-generated answer. You can edit any quantity.

04 · Independent check → original conversation

A check belongs to exact inputs

Enumerate all 2,400 integer vectors; enforce every constraint, then the value and lexicographic tie-break.

Not checked yet. A delivery receipt does not verify this plan.

What this teaches—and what the source proves

The old plan uses 5 units of group y. Keeping 1 from stock 5 leaves only 4. A receipt cannot repair that inequality.

The original optimum is 84; the corrected optimum is 83 under different constraints. That decrease is not a performance comparison.

The reproducible source uses three managed Agents across fresh sessions, versioned requests, explicit file transfer, independent review and a real original-chat return pump. This page illustrates a smaller chain; it does not reenact those runs.

Same host, same Goal, controller-selected phases. It does not establish autonomous scheduling, cross-host support or long-horizon effectiveness/efficiency gains.