In short: The master schedule is a statement of what will be built, in sellable units, in time buckets, at a quantity somebody has agreed to. Its central mechanism is forecast consumption governed by a demand time fence, which decides at what point booked orders stop being netted against forecast and start being the whole of demand. Get that rule wrong in one direction and you plan a phantom 46 percent of extra volume, get it wrong in the other and a shortage stays invisible until the week it ships. The freezing literature from the late 1980s says stability is worth more than the cost penalty it carries, and the schedule's other job is handing order management an uncommitted quantity it can sell.
Sales wants to know whether they can take 400 units in week five. Three people answer, and all three are reading a real number. Demand planning says the forecast for week five is 100 a week and there is room. The plant says the next batch lands in week two and after that there is nothing scheduled until week eight. Customer service says the system shows stock. Everyone is looking at a different object, and the object that would settle it is the one that has not been maintained since the last planner left.
The object has to be buildable and it has to be in sellable units
A demand plan is a statement about what customers will buy. A master production schedule is a statement about what the plant will build, expressed in the items you actually ship or configure from, in buckets that match how the plant is scheduled, at quantities that someone with authority has agreed to.
Those two statements differ for good reasons. The demand plan is unconstrained by design, because you want to see the demand you are turning away. The schedule has to be feasible, so it lands on lot sizes, it respects the fact that the line runs a family once a fortnight, and it carries the deliberate decision to build ahead of a seasonal peak. Vollmann, Berry and Whybark set this out as the anchor of the manufacturing planning and control framework across editions of their textbook from 1984 onward, and the framing has aged well: aggregate planning above, master schedule in the middle, detailed material planning below.
Hax and Meal had made the structural argument in 1975, that production planning decisions belong at different levels of aggregation with different frequencies and different data, and that a single model spanning all of them will be wrong at both ends. The master schedule is the level where the aggregate decision becomes an item and a week, and that translation has to happen somewhere explicit. When it does not, it happens implicitly inside the MRP run, using whatever forecast happens to be loaded.
Forecast consumption and the fence that governs it
Here is the mechanism the whole object turns on.
Take an item with 220 on hand, a lot size of 500, and a forecast of 100 a week for the next eight weeks. Booked customer orders sit at 130, 95, 60, 40, 15 and 5 in weeks one to six, tailing off the way an order book does. The demand time fence is set at three weeks.
Inside the fence, demand is the booked orders and only the booked orders: 130, 95, 60. The reasoning is that three weeks out, everything real has already arrived, and any remaining forecast is a number that failed to convert. Outside the fence, demand is the greater of forecast or orders in each bucket, which gives 100 for weeks four through eight because the forecast is still larger than what has been booked.
Project the balance. Week one closes at 90. Week two would go to minus 5, so the schedule places 500 in week two and the balance closes at 495. It runs down through weeks three to seven at 435, 335, 235, 135 and 35, then week eight would go negative, so a second 500 goes in there. Two scheduled quantities, week two and week eight, and now the question about 400 units in week five has an answer that everyone can read off the same record.
Notice that the item was never short, the forecast was never touched, and the plant got two clean batches. The fence did all the work.
Two ways to get the demand line wrong
Adding forecast to orders. The most common configuration error, usually created by someone who wants to be safe. Week one becomes 100 plus 130, which is 230, and the schedule pulls a batch in immediately to cover demand that does not exist. Across the eight weeks the demand line totals 1,145 against a real 785, an overstatement of about 46 percent, and every bit of that inflation explodes down through the bill of materials into component orders. The plant builds ahead, the stock sits, and the following quarter's forecast accuracy review blames the demand planner.
Using forecast alone inside the fence. The mirror error, and quieter. Week one shows 100 against 130 already booked, so the record says there is cover when 30 units of it have been sold twice. Nothing in the system flags it, because from the schedule's point of view the balance is positive. The shortage surfaces in the week it ships, when a picker cannot fill a line.
The fence length itself is an operating decision rather than a default. Set it at the point where new orders inside the window stop being material, which for most businesses is close to the order lead time customers actually give you. Measure it: take a quarter of order entry data, and for each week count what proportion of the volume shipped in that week was booked less than one week before, less than two weeks before, and so on. Where that curve flattens is where your fence belongs. A fence longer than the real order horizon starves the schedule of forecast and the plan collapses to whatever is booked. A fence shorter than it double counts.
Freezing, and what the research actually found
The planning time fence sits further out and does a different job. It is the boundary inside which the system will not automatically create or reschedule master schedule quantities, so changes have to be made by a person who then owns them.
Sridharan, Berry and Udayabhanu tested this directly in Management Science in 1987, and the result was more encouraging than the folklore. Freezing a larger share of the horizon produced substantially more stability with a cost penalty that grew slowly, so there is a wide middle range where you buy most of the available stability cheaply. Their related work found that the choice of how much to freeze mattered more than the choice of how often to replan.
Zhao and Lee extended the question to demand uncertainty in the Journal of Operations Management in 1993 and found the expected interaction: the higher the forecast error, the more the stability benefit of freezing is worth, because an unfrozen schedule under high error simply transmits noise into the material plan. That runs against the instinct that volatile demand needs a more responsive schedule.
The operational point underneath the papers is about authority. A time fence with no approval step attached to it is decoration. The fence works when a change inside it requires a named person to accept the consequence, and when that acceptance is recorded, because the record is what lets you answer the question of why the plant lost four setups last month.
What the schedule hands to order management
The other output nobody configures properly is the uncommitted quantity. Of the 500 arriving in week two, 155 units are already spoken for by orders booked in weeks two and three. The rest of that batch is available to sell, and it stays available until the next scheduled quantity arrives in week eight.
That single figure is the honest answer to the sales question at the top of this piece, and it is a property of the master schedule rather than of the stock record. How that quantity becomes a date a salesperson can promise, including cumulative and capable to promise variants, belongs to I7.
Two related mechanics are worth naming so they do not get invented locally. Where the item you sell is a configuration rather than a stocked part, the schedule is maintained on a planning bill with option percentages and the final assembly schedule is a separate, much shorter horizon object. Where the plant runs mixed model, the schedule quantity and the sequence are different decisions with different owners, and conflating them puts sequencing arguments into a monthly meeting where they cannot be resolved.
Where this stops
The master schedule assumes there is something stable above it. If the aggregate plan changes every month, the schedule inherits that movement and every fence in the system becomes a way of delaying the arrival of bad news rather than absorbing it. The monthly consensus process that is supposed to hold the level above still is a different discipline (B1), and the schedule cannot substitute for it.
Stability is also genuinely expensive on the days it costs anything. The freezing results describe averages across many runs. On the particular week when a large customer moves an order and the fence refuses it, the cost is concentrated and visible, and the benefit is diffuse and invisible, which is why fences get overridden by people acting reasonably. The defence is not a stronger rule. It is the count of how many overrides happened last quarter, put in front of the people who authorised them.
And the schedule says nothing about whether the plant can build what it holds. A master schedule with an item and a week and a quantity is arithmetic on an inventory balance until it is checked against a resource profile, which is what rough cut capacity planning is for (EE7). A schedule that has never been checked that way is a forecast with better formatting.
This week, take your master schedule as it stood at the start of last quarter, compare it to what was actually built, and split the difference into changes made inside the demand time fence, changes made between the fences, and changes made outside both. The first bucket names your real fence length, and it usually turns out to be shorter than the parameter says.