Entry 0196·November 16, 2026·Scheduling·Throughput, Variability & Constraints

Your Planned Hours Were Negotiated, Not Counted

A plant's continuous improvement lead walked me through his loss data, and the runtime column would not hold still.
Truth · observed pattern

The runtime column nobody could reproduce

A plant's continuous improvement lead walked me through his loss data, and the runtime column would not hold still. Same line, same shift length, and the hours the line was expected to produce came back as 7.75 on one record, 7.58 on another, and lower than both elsewhere. He was not defending the numbers. He had already tried to fix them and had been talked out of it.

The arithmetic should be dull. First shift runs 6:00 to 2:30, which is 8.5 hours. Lunch is 30 minutes, the break is 15, so the line is down for 45 of those minutes by agreement. Subtract and you get 7.75 hours. That figure is the denominator under every downtime percentage, every utilization number, and every loss report in the building.

It was not 7.75. Under two earlier plant managers it had been 7.58, and he could name the difference: gowning time, granted at startup and again at the break. Not measured, not reported as a loss, simply removed from the hours the line was held accountable for. His own position was the correct one. Set it at 7.75 for every line and require every minute to add up to it.

Ten minutes a shift sounds like nothing, and the minutes are not the point. The point is that the denominator turned out to be adjustable, so nothing computed from it can be trusted to mean the same thing twice.

Available time is a fact. Planned time is a decision.

Two different quantities get called the same thing. Available time comes from the shift schedule and the labor agreement; it is a matter of record. Planned time is a field somebody populates, and whoever populates it decides what the operation is answerable for.

Loss accounting only works when the denominator is fixed and every minute inside it is classified. A minute is either run time or it belongs to a named loss. Nothing else is allowed. The moment a minute can be granted instead of classified, the books balance by construction, because planned time has been defined as whatever runtime plus recorded downtime happened to be.

That is the tell, and it is easy to check. If planned time on your tracker always equals runtime plus recorded downtime exactly, with no remainder, the accounting is not closing, it is being closed for you. Real operations produce an unexplained residual every week. A correct system reports that residual as its own line. A system with an editable denominator absorbs it and reports nothing.

The comparability damage is worse than the arithmetic. Each grant was made for a local reason, at a particular time, by a particular manager. One line carries gowning time, another does not, a third was set up after the practice changed. Put those lines side by side in a weekly review and rank them, and you are ranking a mixture of physical loss and inherited bookkeeping that nobody in the room can separate, because the grants were never recorded as grants. They were recorded as planned hours.

I looked at the same operation's scrap accounting separately, where a single plant level rate hid the distribution underneath it. This is the same class of defect in a different place, and the fix is different. A scrap rate fails by being too aggregated. A planned hour fails by being editable.

Changeover is the loss that pays for the ambiguity

Changeovers at this plant are recorded as downtime, which is already better treatment than the break received. Recording them is where the discipline stops. There is no standard time for a changeover, and there is no changeover type. A wipe down, a quick wet pass, and a full teardown all arrive in the same bucket carrying whatever minutes the crew spent.

So the number exists and has nothing to compare against. Forty minutes is neither good nor bad; it is just forty minutes. You cannot tell a crew they ran long, you cannot schedule the next change with confidence, and you cannot build a case for tooling that would shorten it, because the saving has no reference to be measured against.

This is where a loose denominator stops being an accounting curiosity and starts costing money. A mid shift product change is exactly the event a fixed denominator exists for. Run one product or three, the line is still staffed for 7.75 hours, so the changeover minutes have to come out of production time and appear as a loss with a name on it. When the denominator flexes instead, part of the changeover gets funded by shrinking what the shift owed, and the cost of running shorter batches never fully surfaces. On lines taking two or more product changes a day, that is not a rounding difference.

Where the money is, and where the report is

Recorded downtime on these lines runs around 13 percent of the shift. The continuous improvement lead's own estimate of actual downtime is 20 to 25 percent, and he is the person closest to the data. The gap exists because operators record downtime by hand, on paper, while running the line.

Read that plainly. The largest single loss class in the plant is the part that never gets recorded, and it is roughly the size of everything that does. Any ranked improvement list built from this data is ranking the visible half, and the ordering inside that half is partly an artifact of which lines were granted which minutes.

The obvious fix is an instrument. Tap the bagger's own cycle signals so the machine declares its run and stop states, then add a reporting layer so the data can be cut by line, shift, and item instead of sorted by hand before anyone can ask a question. That work is worth doing, and it was already in pilot on three lines.

It will not fix this. Automated capture inherits whatever denominator it is given. Point it at a planned time field that two former managers negotiated and you get negotiated numbers faster, at higher resolution, carrying more authority than before, because now they came from the machine. The instrument makes measurement cheaper. It does not make the definition true.

None of this is a savings claim. Closing the denominator reduces no cost by itself. It changes which projects are fundable on evidence, which comparisons between lines are legitimate, and whether the biggest loss in the plant has a number at all.

Fix the definition before you buy the instrument

Four moves, in this order.

Write planned time once, per shift pattern, as scheduled length minus contractual break, and publish it. No line gets a different rule and no manager gets to adjust it.

Require the minutes to close. Every minute between shift start and shift end lands in exactly one category, and the categories sum to the shift. Carry an explicit unexplained category so the residual is reported rather than absorbed. Expect it to be uncomfortable in month one. That is what it is for.

Give changeover a type and a standard time for each type, so the recorded minutes finally have something to be measured against.

Then automate the capture, against a definition that no longer moves.

Ownership decides whether any of this happens. The continuous improvement lead had already reached the right answer and could not install it, because what came back was a negotiation with plant leadership about whether gowning time counts. A denominator is a definition, and a definition is a leadership decision, not an analytics one. Name the person who owns it and the date it takes effect, or the correct answer stays an opinion held by the one person who cannot act on it.

Warn the organization before the first honest month. Reclassifying granted minutes as loss makes every percentage look worse while the floor runs exactly as it did. Unsaid in advance, the correction reads as a performance decline, somebody defends the old number, and the grant returns with a fresh justification.

Before I accept any loss percentage, I ask what total it divides into and who last changed that total. It is a cheap question, and it has reordered more improvement lists for me than any analysis built on top of the answer.

What a well run time ledger reads

Planned time for a shift pattern is one published number, identical on every line running that pattern, equal to scheduled hours minus contractual break. The loss categories on any line sum to the shift, including an unexplained bucket that is reported rather than held at zero. Changeovers are typed, each type carries a standard time, and recorded minutes are reported against it. Downtime capture comes off machine state rather than a clipboard, and the variance between the two is small enough to state. Every percentage in every review carries its denominator in the same sentence.

The next useful test

Pull last week's loss report for your highest volume line and compute the denominator yourself from the shift schedule and the break agreement. If your number and the report's number differ, you have found a grant. Ask who made it, when, and in exchange for what, and get that answer in writing before you fund anything the report recommends.

Published November 16, 2026
Related reading in Throughput, Variability & Constraints