Haute Lumière

Commerce · IV.10 · MMXXVI · daylight

La Bourse  /  Volume IV  /  Nº IV.10  /  Workbook — the student

WORKBOOK — THE STUDENT

Chapter IV.10 · Measuring Regeneration

For the person learning this for themselves. A term of practice, a term project, and a self-assessment. Applied to a life, not to a firm — because the discipline of asking what a claim costs to prove is a personal discipline before it is a professional one.

---

WHY THIS WORKBOOK IS DIFFERENT

Most of what you will be taught about regeneration is about whether it works. This chapter is about whether you can show that it works, which is a different skill and a rarer one. The first is ecology. The second is engineering with a budget attached, and almost nobody teaches it.

The whole term reduces to four questions you will learn to ask in under a minute, of any claim, in any domain:

1. What is the effect, and what is the scatter it sits in?
2. How many samples, and how far apart in time?
3. What does that cost, and what is the thing worth?
4. If κ is above one, what structure changes it?

By the end you will be able to look at a press release about regenerative farming and say, in a sentence, whether anybody could possibly have measured what it claims. That is a genuinely useful ability and very few graduates have it.

---

PART ONE — DISCOVERY

Weeks 1–4: find a claim somebody is already proving well

Exercise 1.1 — The archive hunt (one afternoon)

Find three long-running measurement series — anywhere, any domain. Rothamsted's Classical Experiments running since 1843 with physical samples archived. LUCAS Soil across the European Union on a repeated common protocol. Your own weight, a tide gauge, a phenology record, a rain gauge at a village hall.

For each, write one paragraph on what made it survive. You will find the same three answers every time: somebody standardised the method, somebody kept the physical material, and somebody returned to the same place.

Exercise 1.2 — The instrument that replaced a month of labour

Read Biggs et al. (2015) on environmental DNA for great crested newts. Write 400 words on the sequence that made the swap legitimate: they quantified the detection probability of the new method against the old, and only then did a regulator accept it. Note what would have happened if the sequence had been reversed.

Exercise 1.3 — The five-minute κ (repeat weekly, all term)

Pick one environmental or social claim a week from anywhere — a label, a report, a news story. Estimate, roughly, what proving it would cost and what it is worth. You will be wrong by a factor of three most weeks. It does not matter. The habit is the deliverable, and by week eight your factor of three will be a factor of 1.5.

Output for Part One: a one-page log of nine claims with your estimated κ beside each.

---

PART TWO — THE ARITHMETIC

Weeks 5–9: learn to compute before you argue

Exercise 2.1 — Reproduce the chapter

Run python3 lib/verify.py IV.10. Then, without looking at it, reproduce by hand:

• k = (z₀.₉₇₅ + z₀.₈₀)² = 7.848880
• independent sampling: n = 2 × 7.848880 × 8.00² ÷ 1.50² = 446.51 → 447
• paired at ρ = 0.90: σ_d² = 12.80, n = 44.65 → 45
• MDD = 2.801585 × √(12.80 ÷ 45) = 1.4942 t C/ha
• T = 1.4942 ÷ 0.30 = 4.98 years

If any of the five comes out differently, find the error before reading on. This is the only exercise in the workbook that must be completed exactly.

Exercise 2.2 — Move one input at a time

Rebuild the sample-size calculation in a spreadsheet and vary one input while holding the rest:

Write one sentence per row. The one you will remember is ρ: from 0.50 to 0.95 takes n from 224 to 23.

Exercise 2.3 — Build the κ table yourself

Compute κ at $20/t for 2, 50, 200, 500, 1,000 and 5,000 hectares using C_fixed = $17,000.00, $2,000.00 per mobilisation, $55.00 a sample, 45 cores per stratum and one stratum per 200 ha. You should get 163.826, 6.553, 1.638, 0.905, 0.578 and 0.366.

Then find the floor: c_ha = 2 × 45 × 55 ÷ 200 = $24.75, so κ_floor at $20/t is 24.75 ÷ 79.20 = 0.312.

Exercise 2.4 — The honest negative, computed by you

For a two-hectare holding at $20 a tonne, compute the value of a five-year claim ($158.40) and κ against a $25,950.00 verification (163.826). Then do it at $50 a tonne (65.5). Then with a stripped $5,000.00 audit at $20 a tonne (31.6).

Write two hundred words answering: what should a standard say to this farmer? Do not resolve it comfortably. The honest answers are all uncomfortable.

Exercise 2.5 — The eDNA replicate table

Compute n ≥ ln(0.05) ÷ ln(1 − p) for p = 0.30, 0.50 and 0.70. You should get 9, 5 and 3. Then compute the site count for an occupancy shift from 0.30 to 0.45: 159.59 → 160 sites, and the two-date cost at five replicates and $55.00 each: $88,000.00.

Output for Part Two: a working spreadsheet that reproduces every figure in the chapter's Arithmetic, and a half-page note on which input you found most surprising.

---

PART THREE — DREAM AND DESIGN

Weeks 10–13: build the apparatus

Exercise 3.1 — Design a real sampling plan for real land

Find an actual parcel you can walk — an allotment, a college field, a friend's smallholding, a park. Sketch it. Stratify it. Place your paired points on the sketch and write the coordinates. Cost the plan using the chapter's stack. Then compute the minimum interval at which resampling would tell you anything.

You will almost certainly find the parcel is far too small for the claim to be worth proving. That finding is the exercise, not a failure of it.

Exercise 3.2 — The three-dollar stake, done for real

Buy or make four permanent markers. Install them on a piece of ground you have access to, with permission. Record the coordinates and photograph each from the same bearing. Write the register: identifier, coordinates, date, installer, bearing of the photograph.

Total cost: under twenty dollars. You have now built the single highest-return instrument in this chapter, and you will understand the 327.6× figure in a way that no amount of reading produces.

Exercise 3.3 — Assemble a pool on paper

Using H = C_fixed ÷ (κ_max·V_ha − c_ha) with C_fixed = $16,500.00 and c_ha = $2.74, compute H at $10, $20 and $50 a tonne: 3,324.9, 1,303.1 and 461.4 hectares.

Now find, in a real place you know, whether 1,303 hectares of comparable land could be assembled and who already convenes the people who own it. Name the body. If there is none, that is the finding.

Output for Part Three: a sampling plan, a photographed marker register of four points, and a one-page pool memo naming a real convening body.

---

PART FOUR — DESTINY AND DELIGHT

Weeks 14–16: make it hold, and enjoy it

Exercise 4.1 — Write the failure list before you need it

For your paper pool, write the four failure modes the chapter names, in your own words, with what you would do about each: uniform deductions driving out the best members; early resampling producing an unpublishable near-zero; the calibration subsample trimmed in a cost review; the aggregator becoming the only counterparty.

Exercise 4.2 — Go back to your markers

At the end of term, walk back to the four markers with your register and the photographs. Find each one. Note how long it took and what had changed.

This is the delight exercise and it is not sentimental. The feeling of finding the peg exactly where the register said it would be is the feeling the whole chapter is built to produce, and you cannot get it from the reading.

Exercise 4.3 — Teach it in five minutes

Explain κ to somebody with no background, using one example, in five minutes, without a slide. If they can afterwards tell you what a κ of 163.826 means for a two-hectare farm, you have understood the chapter.

---

THE TERM PROJECT

One piece of work, carried the whole way

Choose one claim and take it all the way to a verification design.

It does not have to be soil carbon. It can be a river's water quality, a school's tree canopy, a hedgerow's bird occupancy, a community garden's soil, a building's energy use. What matters is that it is real, local, and that somebody would care about the answer.

Deliver five things, roughly 3,000 words in total plus the working:

1. The claim, stated in one sentence with its effect size and its units.
2. The noise, estimated from literature or from your own pilot samples, with the source of the estimate named.
3. The design — sample count, paired or independent, the interval, and the arithmetic that produced all three.
4. The cost stack and κ, computed at three plausible values for what the claim is worth.
5. The structure, if κ exceeds 0.25 — which of the three applies, what it moves the number to, and a real convening body named.

The mark is on the arithmetic and on the honesty of the negative. A project concluding this claim cannot be verified at any price by this party, and here is the pool that could earns full marks. A project concluding it all works out without printing its per-hectare term earns very few.

---

SELF-ASSESSMENT

Score yourself honestly. Five is I could do this in front of somebody.

Anything below three is a two-hour fix. Do the two hours.

---

CARRYING IT FORWARD

The four questions travel. Ask them of a medical claim, a policy evaluation, an educational intervention, a training programme, a marketing attribution model. What is the effect against the scatter. How many and how often. What does the proof cost against what the thing is worth. And if the ratio is bad, what structure changes it.

You will find that most of the world's confident claims have never had the second question asked of them, and that asking it politely, with the arithmetic in your hand, makes you extremely useful to have in a room.

---

FIVE ERRORS YOU WILL MAKE, AND THE FIX FOR EACH

Every one of these was made by somebody competent before it was written down here. Making them is not a failure of ability; not recognising them afterwards is.

You will compute a sample size and forget to say per what. Four hundred and forty-seven cores per date is a very different sentence from four hundred and forty-seven cores. The fix: write the unit into the variable name before you write the number.

You will use an unpaired formula on paired data, or the reverse. The tell is that the answer changes by roughly a factor of ten, which is exactly the distance between 447 and 45. The fix: before choosing a formula, answer out loud whether the same physical point is being read twice. If the honest answer is probably, it is not, and you use the unpaired form.

You will treat the accrual rate as certain. It is the input with the widest real range in the whole calculation — roughly 0.1 to 0.6 t C/ha/yr for improved cropland. Halving it doubles the required interval to nearly ten years. The fix: compute the design at the low end of the range as well as the middle, and quote both.

You will quietly round. A figure your prose rounds without declaring it reads, to anybody checking, as a figure nobody computed. The fix: print the number to the precision you computed it at, and if you want a rounder one for the reader, give both — 1.4942, call it one and a half tonnes.

You will present κ without its denominator's price. κ is not a property of a practice; it is a property of a practice at a price. A κ of 163.826 at $20 a tonne is 65.5 at $50. The fix: never write κ without writing the price beside it, in the same sentence.

---

A NOTE ON READING THE LITERATURE

You will meet three kinds of paper this term, and it helps to know which you are holding.

Papers that measure a thing. They report a value, a method and an error, and their usefulness to you is almost entirely in the error. Read the methods section for the sample count and the design before you read the abstract.

Papers that measure a method. Rarer, more valuable, and the engine of everything in this chapter — Biggs and colleagues on eDNA detection probability, Ficetola and colleagues on replication and false absences, Brus and de Gruijter on design-based against model-based sampling. These are the papers that let you swap an instrument without losing the ability to say what you know.

Papers that measure a market. Badgley and colleagues on over-crediting in California's forest offsets is the model. They take a class of claims and ask what the claims would have had to be true for. Read at least one of these before you believe any aggregate figure about how much carbon a programme has abated.

If you read one of each kind every month for a year, you will be better equipped than most people currently employed to evaluate these claims professionally, and that is not an exaggeration.

---

APPRECIATIVE QUESTIONS FOR YOUR SEMINAR

1. Which measurement in your own field has survived longest, and what did the people who started it do that made it survive?
2. Where have you seen an error bar published as prominently as the value, and what did that make possible?
3. If everyone in this room pooled the measurement capacity they already have access to, what could we collectively verify that none of us could alone?
4. What would you want to be able to prove in ten years, and what would you have to install this year for that to be possible?