Live Demo

Past-Due Order Triage

Stop past-due production orders from over-ordering your raw materials.

→ An old order nobody closed is quietly telling the system to buy more material.

The problem

An open in-house production order keeps generating dependent demand for the raw materials it was built to consume. Past due and never going to run, that demand is false — but MRP has already procured the components to cover it, so they land and pile up as overstock. Nothing flags which stale orders are quietly inflating your raw-material inventory.

What it does

A triage worklist for past-due in-house production orders — the ones still signaling false demand for the raw materials MRP keeps procuring. It ages every open order, weights each by size and staleness, and sorts it into one clear action — unfirm, close, verify a partial, reschedule, or leave — with the where-used BOM showing which purchased components each stale order is over-ordering, and a snapshot diff that tells the orders actively confirming apart from the dead ones inflating your stock.

Initializing · Past-Due Order Triage — real code, loading in your browser

About this tool

Past-Due Order Triage: what it solves

An in-house production order doesn’t just wait in MRP — while it stays open, it keeps generating dependent demand for the raw materials it was built to consume. When the order is past due and never going to run, that demand is false: MRP has already triggered procurement to cover it, so the externally-sourced components arrive, get stored, and pile up as overstock no one ordered on purpose. Past-Due Order Triage finds those stale orders, ages each one, weights it by quantity and how long it’s been late — a phantom score — and sorts it into exactly one action: unfirm a firmed planned order, close a dead one, verify a partial remnant, reschedule to an honest date, or leave it. The where-used BOM shows which purchased components each order is over-ordering, and a snapshot diff tells the orders actively confirming apart from the dead ones quietly inflating raw-material stock. It’s the planning-signal integrity check MRP exception reports were never built to run.

An open production order keeps buying material for work that will never happen

This is the part that surprises people who have not worked a planning floor.

A production order does not just sit there waiting. While it stays open it keeps telling MRP that it needs the raw materials it was built to consume. That is dependent demand, and MRP treats it as real.

So the components get ordered. They arrive. They get put away. And the order they were bought for is six weeks past due and is never going to run.

Nobody made a purchasing mistake. The system did exactly what it was told by an order nobody remembered to close, and the result is raw-material stock nobody chose to hold.

Age alone does not rank them

Every past-due order is late. That tells you nothing about which one to deal with first.

The tool weights each by how much it is asking for and how long it has been asking. A large order a month late is doing more damage than a small one late since spring, and a small order that has been open for a year is a different kind of problem again.

The weighting gives each stale order a score, and the score is what sorts the list.

Each one comes out with exactly one action

Not a flag. Not a severity. One of five things to do:

Unfirm it — a firmed planned order holding a date somebody set by hand and then forgot.

Close it — dead, and the demand it carries is entirely false.

Verify a partial — something ran, something did not, and the remainder is a question for the floor rather than the planner.

Reschedule it — the work is still real and the date is a fiction. Give it an honest one.

Leave it — genuinely fine, and saying so out loud stops it being re-examined every month.

A worklist where every line has one verb is a worklist somebody finishes.

The where-used view shows what each order is actually over-buying

Knowing an order is stale is half of it. The other half is which purchased components it is inflating.

The tool reads the bill of materials backwards from the stale order, so each line names the bought parts sitting in a rack because of it. That is what turns “close some old orders” into a number somebody in finance recognizes.

A snapshot diff separates the living from the dead

Some past-due orders are still confirming activity — material moving, operations closing. Those are late but alive.

Comparing two snapshots tells them apart from the ones where nothing has happened for months. The dead ones are the ones inflating stock, and they are the ones worth the attention.

What this page is

A working demo of a tool that is built to order.

It runs on Northpoint Manufacturing, an invented company with invented part numbers and invented order history. No real production data is behind any of it.

It is one of the factory tools sold to manufacturers on the manufacturer page: built to order for one plant on its own extract, or run once for the written answer. The invented data is what lets anyone open it first.

What it will not do

It closes nothing. Every action is a recommendation for a person. An automated close on a misread partial would destroy a real record.

It sets no staleness threshold for you. How late is too late depends on your production cycle, and nothing in this data knows it.

It does not value your inventory. It names the components an order is over-ordering. Attaching money to that is your own costing, done in your own system.

It reads a snapshot, never a live system. The demo runs entirely in your browser on fixed data.

Built to order for your plant.

This one runs on invented data. Built to order for your plant, on your own extract, and yours outright: the code, the accounts, and a copy that runs with the internet off. Nothing is built before it is bought and there is no license to renew. A factory tool starts at $10,000 per plant. That is a floor per plant for a tool built to order on your own extract, never a quote. What yours costs is agreed before any work begins.

Or run once. The same tool on one extract of yours, and what you get is the written answer it produced rather than the tool. The audits below are that form, and each names the tool behind it.