ORDER BOOK

Commandplant manager — shop health & the orders to issue

Controlmove a lever — the book re-solves at each position

Floorscheduler — sequence, what slips, the bottleneck

Sensingmodel fidelity — isolate the bottleneck, where we're blind, the one measurement to take

Accountscustomer-facing — promises, the date you can hold, price

Capitalowner / board — over-committed? buy? trustworthy?

Evidenceowner / board — solver speed and promise calibration, on real data

Station ↗operator — one queue (not this screen)

Situationwhere we are · how we got here · why

Alertsred — acute, act now

Orderseach ties to a lever — issue it, or flex it in Control; the impact is shown before you commit

Control surfacemove a lever — the book re-solves at each position (every position is a real re-solve)

Order boardpress ▷ — each step is a candidate accept date; the book re-solves

WOCUSTOMERPART QTYON-TIME · EXP $

Plant mapresource × time over the horizon — where the schedule packs and the loss sits

Floor layout · hotspotsthe same machines in space — shaded by load; the Gantt's bottleneck, on the floor

Robustnesswhich promises are brittle to run-time swing

Repaira resource breaks — the minimal-disruption patch vs a full reshuffle

Isolate the bottleneckthe binding constraint, known two independent ways — free off the held plan, and confirmed by a price re-solve

Where we're blindthe observation floor — above it promises are calibrated; below it the model goes blind and says so

The one measurement to takethe cheapest sensor that expands what we can trust — observability as a costed, prioritized roadmap

Promisesthe date you can stand behind, per customer

Negotiationmove the date, who pays

Rush pricewhat qualified capacity is worth

Capacitymarginal $ of one more unit, per resource

Riskthe book's lateness exposure, in finance terms

Dynamicshow the plant runs under stress

Trustdo stated confidences hold

Solverwarm re-solve vs cold, on standard + the real instance

§6 integer layersolve_milp and price_feed_integer over the session — the integer optimum and the $/cell the LP returns as 0

Real data — not a demothe certified quote + calibration on the SOURCE-01 production log

Intent recoveryrecovered dispatch intent from held-out choices, and where it isn't identified

engine-certified · viz/board.py → board.json. promise intervals = simulated completion spread of the committed plan; each play-frame = a real accept+re-certify; price & capacity value = re-solve per resource; situation, hotness, alerts = read of the same plan. no synthetic events.