Haute Lumière

Commerce · IV.05 · MMXXVI · daylight

La Bourse  /  Volume IV  /  Nº IV.05  /  Workbook — the student

A watercolour of rolling hills in late summer, a farmhouse and a line of trees on the far ridge.
Plate IV.05 · Workbook — the studentThe Ten Thousandth.The ten thousandth panel costs less than the first. Nobody economised. Ten thousand people each learned something small and none of them could tell you what.

WORKBOOK — THE STUDENT

Chapter IV.05 · Energy Systems That Compound

For the person studying this alone, or in a seminar, with no procurement budget and no factory. You are not too early. The regression you learn to run here is the same one that decides billion-dollar contracts, and you can run it this afternoon on data you already have.

WHY THIS WORKBOOK IS DIFFERENT

The chapter was written for somebody with a spend to commit. You may have neither the spend nor the counterparty. It would be easy to conclude that the method waits until you do.

It does not, and the reason is the most useful sentence in this workbook: a learning curve is a measurement instrument before it is a purchasing strategy, and the instrument works on anything that gets repeated. Your own practice. A lab protocol. A commute. A recipe. A codebase. The energy system is the case where the stakes made somebody keep the records for forty-seven years, which is why it is the example — not because it is the only place the equation holds.

So you will do exactly what the executive does. You will run the regression, print the interval, and refuse to quote a point estimate. On a system you actually control, which for the next decade is a considerable advantage.

PART ONE — DISCOVERY

Weeks 1–4: find the curves you are already on

Exercise 1.1 — The repetition inventory (60 minutes)

List everything you have done more than twenty times in the last two years and could plausibly have timed or costed. Problem sets. Essays. Shifts at a job. Lab runs. Rehearsals. Commits. Translations. Kilometres.

For each, write three columns: what is the unit, roughly how many have I done, and do I have any record of how long the early ones took. The third column is where this exercise usually collapses, and the collapse is the lesson: you have been on curves your whole life and kept no series. Start one today for the thing you will still be doing in a year.

Exercise 1.2 — Read the chapter's series with your own hands (45 minutes)

Open lib/verify/IV_05.py. Find the eleven-point module-price series and read the inputs before the results — that ordering is deliberate and it is the habit being taught. Then answer on paper:

Exercise 1.3 — The appreciative interview (45 minutes, with another person)

Find somebody who has done one thing many thousands of times — a nurse, a printer, a chef, a machinist, a violinist. Ask exactly this, without variation: tell me about the difference between the tenth time you did this and the thousandth. Do not ask what got easier. Ask what got different.

Write down every mechanism they name. You will get things no equation predicts: they stopped needing to look; the tools moved closer; they learned which step to do wrong on purpose. That list is what the exponent is made of, and it is the thing the chapter's own sources — Nemet, Kavlak — went and found in a factory rather than in a model.

PART TWO — THE ARITHMETIC

Weeks 5–8: run the regression and own the interval

Exercise 2.1 — Your own learning curve (2 hours)

Take the one activity from 1.1 with any usable record. Put cumulative units on one axis and time-or-cost per unit on the other, both in logs, and fit a line. Ten points is enough. Report four things and never fewer: slope, standard error, R², and the implied learning rate with its interval.

A worked one to check yourself against. Suppose twelve hours on the first unit, forty units completed, a learning rate of 15 percent:

  doublings over 40 units                 5.32
  hours on the 40th unit                  5.05
  total hours saved against no learning   222.9

Two hundred and twenty-three hours is five working weeks, recovered by nothing but repetition, in a task most people would describe as just getting on with it.

Exercise 2.2 — Break your own fit on purpose (45 minutes)

Drop your three earliest points and re-fit. Then drop your three latest and re-fit. If the learning rate moves by more than a few points, your series contains a regime change — the product changed, the tools changed, you changed — and the honest move is to cut the series at the change rather than average across it. Write down where you cut it and why.

Exercise 2.3 — The sensitivity that matters (45 minutes)

Project your own curve forward two doublings at the bottom and the top of your interval. Then project it at your central rate but with half and double the volume. Which spread is larger?

For solar the answer is the deployment one: the measured learning interval moves the 16,000 GW answer by 1.30× while a deployment range of 4,000 to 16,000 GW moves it by 1.67×. If your own answer comes out the same way, you have just learned the chapter's central practical point on your own data: an argument about the rate is usually a displaced argument about the volume.

Exercise 2.4 — The honest negative, worked (60 minutes)

Compute the firming arithmetic from the chapter yourself, from the four inputs, without looking at the answer: $250/kWh installed, 300 cycles a year, 85 percent round-trip, fifteen-year life, 7 percent cost of capital. You should get a capital recovery factor of 0.1098, an annual charge of $27.45, delivery of 255.0 kWh per kWh of capacity, and a shifting cost of $0.1076/kWh. Against prompt solar at $0.0440, a delivered shifted kilowatt-hour is $0.1516 — 3.4×.

Then do the thing that separates a student from a spreadsheet: change one assumption and say which one you were least entitled to. Cycles per year is the usual answer, and it is not a measurement.

PART THREE — DESIGN

Weeks 9–12: build something that compounds

Exercise 3.1 — Choose a practice with an exponent (90 minutes)

Pick one thing you intend to still be doing in ten years, and design it for learning using Malhotra and Schmidt's mechanism: small, repeated, standardised, and measured. Write down the unit, the cadence, and the one number you will record every time. The number must take under a minute to record or you will stop.

The counter-example is in the chapter and it is worth keeping beside you. The French reactor programme — 58 units, 5.86 doublings, a cost increase of about 3.5×, a learning rate of −23.8 percent — is what happens when the unit is large, bespoke and assembled outdoors by people who will never build another. Ask of your own practice: is this a module or is it a reactor?

Exercise 3.2 — The volume commitment, personally (60 minutes)

Find one place where you are negotiating price and could be committing volume instead. A gym, a season ticket, a bulk order, a longer lease, a standing arrangement with a tutor or a collaborator. Compute the two paths on paper.

You will usually find the committed path is cheaper and you avoided it because it was irreversible, not because it was expensive. That is worth noticing about yourself. Irreversibility is what the other party is actually buying, and it is the only thing you have that they want.

Exercise 3.3 — The soft-cost audit (45 minutes)

Residential solar costs about $3.00/W in the United States and about $1.00/W in Australia for identical hardware — a ratio of 3.0× made entirely of forms, inspections and queues. Find the equivalent in your own life: something where the thing itself is cheap and the process of obtaining it is expensive. Course registration. A visa. A grant application. A medical referral.

Write the process out as steps with a time against each. You have now produced the only document that has ever reduced a soft cost, which is a list.

Exercise 3.4 — Read a claim the way this chapter reads one (75 minutes)

Find any public document making a cost-decline claim about a technology — a consultancy deck, a ministerial press release, an investor presentation, a campaign briefing. Almost any of them will do. Then answer six questions in writing, and answer them in this order, because the order is the method:

  1. What is in the numerator? Module cost, capital cost, levelised cost, or an installed system price? These are four different numbers and the chapter shows three of them for one technology — 39.4 percent, 23.0 percent and 12 percent for onshore wind, all defensible, none interchangeable.
  2. What is the independent variable — time or cumulative volume? If the document does not say, it is using time, because time is easier.
  3. Is there an interval? If not, what would the claim look like at the bottom and the top of the published spread — 10 percent to 47 percent for solar modules is the band a careful reviewer would apply.
  4. What is the denominator of the deployment assumption? A projection at 16,000 GW and one at 4,000 GW differ by 1.67× at the same learning rate.
  5. Where is the integration cost? If the document compares a generator to a system, it has crossed a boundary without saying so.
  6. What does the author lose if this is wrong? Not a cynical question. A forecast nobody is exposed to is a different kind of statement from one somebody has underwritten.

Write your six answers on one side of one page. That page is the single most transferable artefact in this workbook, and you will use the same six questions for the rest of your working life on claims that have nothing to do with energy.

Exercise 3.5 — The term of practice (ongoing, ten minutes a week)

For the remaining weeks of the term, keep one series. One line a week: date, cumulative units, time or cost. Ten minutes, no more. At the end of the term, re-run Exercise 2.1 on the fuller series and compare the interval with the one you got from your patchy historical data.

The interval will narrow, and watching it narrow is the point. An interval is not a confession of ignorance. It is a measurement of how much you know, and it improves with exactly the same discipline that the cost itself does.

PART FOUR — DESTINY AND DELIGHT

The term project, and what to keep

The term project — the series nobody kept

Find a real cost or performance series that nobody has fitted, assemble it from public sources, fit it, and publish the result with its interval and its denominator. Candidates: the price of a hearing aid; battery-electric bus costs in one city; sequencing cost per genome; the cost of a cataract procedure in one health system; heat-pump installed cost in one country; desalination cost per cubic metre.

The deliverable is four pages.

  1. The series, with every source named and every gap declared.
  2. The fit: slope, standard error, R², learning rate, interval.
  3. A second route that shares none of the first route's assumptions, and the disagreement between them stated in points — not reconciled, stated.
  4. What the number did not look at. One page. This is the page that makes the other three trustworthy.

You are marked on the fourth page.

Six terms to own before the examination

Write each definition from memory, then check it. A term you can only recognise is a term you cannot use in a room.

Self-assessment — six questions

Delight — what to keep

Keep the plot. Print it, if you still print things. There is a specific pleasure in the moment a mess of your own recorded effort resolves into a line, and it is not the pleasure of being right — it is the pleasure of discovering that something you thought was chaos had a slope all along.

And keep the habit that produced it, which is smaller than it sounds: write down how long it took. That is the whole instrument. Forty-seven years of the world's most consequential cost series exists because somebody in 1976 wrote down a price, and somebody else did it again the next year.