The case that was true and undated

Approved four times, and never once dated

Programme C puts counters and fault reporting on Wanagiri's twenty-one special machines: bartacks, buttonholers, pocket welters, belt-loop machines. Its case claims USD 28,400 a year from raising their utilisation, and the case is correct. Sundari's committee reviewed it four times over fourteen months, approved it every time, and never started it, because something more urgent always was.

By the time the question came back, programme A was live and programme B, the planning replacement, was the programme in flight. Everything below is counted over three years, 156 weeks, from the day B's specification work began.

The claim was never wrong. What it never said was from when.

USD 25,200 of the USD 28,400, which is 88.7%, is unavailable until B is live. Not smaller. Unavailable. Knowing that a welter has been idle for forty minutes changes nothing in a factory where next week's sequence was fixed on a spreadsheet last Thursday and cannot be re-cut without re-cutting the fabric plan behind it. The supervisor can see the idle machine perfectly well without a sensor. What she cannot do is move work to it.

Why nobody caught it, and it was not carelessness

Bayu wrote C's case. The precondition, a plan that can be re-cut inside the week, is owned by Lestari. The two cases went to the same committee in different months, and no document in the building has a field for this benefit depends on that programme. A case is a claim about a benefit. A plan is a claim about a date. The sentence joining them belongs to neither, and so it was written nowhere.

Course 22.5 makes the same point inside a single programme: work done above a layer that has not been built is repeated once for every time the layer beneath it changes. This is the portfolio version, and it is harder, because the layer beneath belongs to a different department, has a different sponsor and owes you nothing. Inside one programme you can re-order the work. Across two, you have to ask.

What starting early actually costs

The instinct is that starting C now buys time. The sensors go on, the wiring is done, and when B lands the benefit is already there. Wanagiri costed both.

Wait for B. C's specification work begins when B goes live at week 32.9. Retno's 68 hours take 13.1 weeks. The installation has a 16-week lead through a network contractor and a shutdown weekend, running alongside. C is live at week 50.9.

• Three-year net: USD 25,770.30

Start now. C's specification runs first, so C is live at week 18.0, and B is pushed back by the same 13.1 weeks, from week 32.9 to week 45.9. Between weeks 18.0 and 55.6, C earns only its maintenance slice of USD 3,200 a year. When B finally lands, C's configuration has to be rebuilt against a world where work can move: USD 9,400 of supplier time and 40 more of Retno's hours. So C's full benefit does not begin until week 55.6, which is later than if it had waited.

• C's three-year net: USD 13,326.39
• B's lost benefit from being pushed: minus USD 14,334.32
• Together: minus USD 1,007.93

Waiting is worth USD 26,778.22 more than starting early, and the head start is the smaller half of why. Look at it on its own. The thirty-seven weeks of early life earned USD 2,314.79, and undoing them cost USD 9,880.00, which is 4.27 times what they earned. The rest of the difference is B, delayed by a programme that could not use the time.

The part that can start early, and the reason it usually cannot

There is a real version of the instinct, and it is worth separating from the false one. C has two halves. The configuration depends entirely on B. The physical half depends on nothing at all: where the counters go, how they are wired, which machines are in scope. Its sixteen-week lead time is the longest single duration in the programme. A cable does not care what the planning system is.

So specify the install now, run the lead time in the background, and hold the configuration until B is live. C then goes live at week 46.8 instead of 50.9. Against that, USD 14,600 of hardware is drawn about 32 weeks earlier than it needs to be, and Wanagiri's facility charges 11.5% a year, which is USD 1,033.23.

• Net: USD 26,681.44, which is USD 911.15 better than waiting for everything.

It works only under one condition, and Wanagiri failed it. The early specification has to cost the constraint nothing. If Retno writes it, it takes 6 of her hours, which pushes B by a bit over a week, and the arithmetic reverses:

• Net after B's loss: USD 24,866.36
• Six hours of the constraint cost USD 1,815.09
• Sixteen weeks of lead time bought USD 911.15

The six hours cost 1.99 times what the sixteen weeks bought. Wanagiri gave the install specification to Joko Mahendra, the maintenance chief. He could do it because it is a question about machines rather than about how the factory plans, and that is the test. If the early half genuinely needs nobody who is short, start it. If it needs an hour of the constraint, it is not the early half. It is the same half, wearing a different name.

What to take away

1. Split every claimed benefit by what has to be true for it to exist. C's split was USD 3,200 available and USD 25,200 conditional, 88.7% of the claim.
2. A case states a benefit and a plan states a date. The sentence joining them belongs to neither, so nobody writes it.
3. Starting a conditional programme early does not bring the benefit forward. It buys a small benefit and an expensive rebuild, and it delays the programme the benefit was waiting for.
4. Wanagiri's head start earned USD 2,314.79 and cost USD 9,880.00 to undo, 4.27 times over, before counting the delay it caused.
5. Under every claimed benefit write three lines: the precondition, its owner, and the date they have committed to. A blank third line moves the benefit out of the forecast.
6. Start early only the part that costs the constraint nothing. Six hours of Retno were worth 1.99 times sixteen weeks of lead time.