Production planning starts with a drawing that runs the factory.

• ✓ Drag three work centres onto the canvas, connect them, attach two materials to the first stage and a sub-assembly to the third, and press Release. That is the whole flow, and it is genuinely under two minutes on a part you know.
• ✓ You place work centres, not machines — a bank of four interchangeable presses, a trim cell, an inspection bench with no machine at all. The palette is grouped by area, because that is what a work centre is, and each one carries its machine count so “any of these four” is a real instruction rather than an empty field.
• ✓ Every stage shows its operation number, its work centre and area, and its times — setup 20m · 45s/ea — so the picture is readable across the room and not just by the person who drew it.
• ✓ Cycles are refused as you draw them: an edge that would loop the flow back on itself greys out mid-drag, and if one already exists the message names the path — 10 → 20 → 30 → 10 — instead of saying “invalid routing”.
• ✓ It works on a shop-floor tablet: long-press to move a stage, long-press a port to connect one, and a finger that was scrolling never leaves a node somewhere you did not put it.
• ✓ Auto-layout tidies what it drew and leaves alone what you moved — a planner’s arrangement is never discarded to make the algorithm’s picture neater.

• ✓ Draft, Pending approval, Released, Superseded. A released version cannot be edited at all — the screen has no unlock button, because an editable-looking form that cannot save is worse than a disabled one.
• ✓ Revise deep-copies a released version to a new draft, canvas positions and all, and releasing it supersedes the old one in the same transaction. Nothing is overwritten and nothing is lost.
• ✓ Where approval is switched on, the approver cannot be the author — submitting freezes the document exactly as releasing does, so the approver signs off the thing they read rather than the thing it became while they were reading it.
• ✓ Versions carry effectivity dates. A works order raised last month still resolves the version that was in force last month, so a two-year-old job stays reproducible after two process changes — the same discipline the specification approval trail applies to inspections.
• ✓ Who submitted it, who released it, who rejected it and why — recorded on every version whether or not approval is switched on, so turning the control on next year does not leave a hole in the history behind it.
• ✓ Two planners editing one process is caught rather than merged: the second save is refused with “this process was changed by someone else — reload and try again”, and neither person’s work quietly disappears.

The designer says how a part is made. The schedule board says when it runs. The designer has no time axis, no calendar and no view of machine availability — a process drawing that knew about capacity would be a scheduler with a worse interface.

• ✓ ERP-agnostic by design: a CSV or Excel file with named columns, up to 5,000 rows — no ERP project, no middleware, no waiting for your vendor’s integration quote.
• ✓ Column names are matched forgivingly — Order Number, order_number and ORDERNUMBER all land in the same place — and a column we do not recognise is a warning, never a failed import.
• ✓ Preview before commit writes nothing at all. You see counts to create, update and skip, and an exception report naming the row, the column, the value and an error code you can send to whoever owns the file.
• ✓ Dates are never guessed. ISO and SAP formats are read directly; anything ambiguous is raised as an exception for you to resolve, not silently read as American.
• ✓ Re-importing an unchanged file is a no-op — the same rows are skipped, not duplicated — so a nightly drop can run without anyone watching it.
• ✓ Orders that have already started are never modified, and nothing is ever deleted: an order that drops out of the file stays exactly as it was.

• ✓ Two-part bars — setup then run — on one lane per machine, with the run segment filling live from real cycle counts, so you can see progress without asking anyone.
• ✓ Schedule backwards from a due date and SmartFactory works out when the job has to start, using the working calendar; if that start is already in the past, it says so rather than pretending.
• ✓ Pin the jobs that cannot move. Everything else is arranged around them; a pinned job is never quietly shifted to make the picture tidier.
• ✓ Eight named conflict codes, colour-coded on the bar: pushed by a pinned job, pinned overlap, late, late-forecast, starts in the past, no cycle time, no capacity, machine inactive. Every warning has a name you can look up.
• ✓ The due-date caret turns red on forecast overrun before the job is late: a job running 20% slow flips to Late-forecast with no page refresh, while the schedule itself still says on time.
• ✓ An at-risk widget ranks the five jobs with the worst negative slack — the shortlist for the morning meeting, in the order the morning meeting needs it.
• ✓ Forecasts use a rolling average of recent completed cycles, and mark themselves low-confidence while they are still working off too few samples.

• ✓ Every running job carries a completion forecast built from its own recent cycle times, banded On Track, Behind or Critical.
• ✓ Schedule-late and forecast-late are reported separately, so “the plan says we are fine” and “the floor says we are not” never get averaged into one comfortable number.
• ✓ Schedule-adherence reporting compares what you planned with what happened, per machine and per period, with the variance signed so late is always positive.
• ✓ Material consumption and labour time are booked against jobs from the operator station — which is where cost per part starts.

SmartFactory computes durations, dependencies and conflicts for the sequence a planner chooses. The system does not re-order your factory behind your back.