Supply Chains as Ecologies

Volume IV — Production and Regeneration

THE PLATE

A woman reading in a rattan chair beside a tall white window, light falling across her and the floor. — Plate IV.07The Second Room.Everyone draws a supply chain as a line because a line fits on a page. The thing itself is a population, and it is lit from a room you have not entered.

THE LETTER

You already know that the phrase supply chain is wrong. You have known it since the first time a part you had never heard of, from a company you had no contract with, in a country you did not know you were exposed to, stopped your line for a fortnight.

What you may not have is the arithmetic that replaces it — and without the arithmetic the correction stays an attitude. It is easy to say a supply chain is an ecosystem. It is more useful to say: there are about six thousand firms in my third tier, I can see thirty-six percent of them, mapping them costs seven point two million dollars a year, and the standard I have written into my contracts is enforceable to a depth of one and six-tenths tiers. That last sentence is a plan. The first one is a metaphor.

This chapter does that arithmetic. It asks four questions, all of which have numerical answers you can compute for your own firm before the end of the week: how many firms are actually upstream of you, what it costs to see them, what a single node's failure does to everybody else, and how deep a standard you write can actually reach before enforcing it costs more than the standard is worth.

Two of those answers are pleasant and two are not. The unpleasant ones are the reason the chapter exists, because the standard corporate response to supply risk — qualify more suppliers, write a stronger code of conduct, audit further upstream — does two of those three things at once and they work against each other. Diversity is bought with leverage. Nobody prices the trade, so nobody notices the bill.

We are going to price it.

— The Editors

DISCOVERY

What is already working

Start where it has already worked, because in every one of these cases somebody found out what the real structure was and then used it, and in none of them did the answer arrive as software.

Aisin Seiki, 1 February 1997. A fire destroyed the plant in Kariya that made the brake proportioning valve — the P-valve — for almost every vehicle Toyota built. Aisin was effectively the sole source: a single small part, a few thousand yen, on which the entire group's output depended. Toyota's plants stopped within two days. What happened next is the most instructive week in the literature. Toyota did not have a contingency plan for this. Instead the group distributed the blueprints laterally, and more than two hundred firms — most of them with no experience of hydraulic brake components, several of them Toyota's competitors' suppliers — began working out how to make the valve. Sixty-two of them succeeded. Line production resumed in about five days, and by the end of the second week output was near normal. Nishiguchi and Beaudet, who documented it closely, make the point that matters here: nothing in the formal contracts produced that recovery. What produced it was a dense lateral network of firms that already knew each other, already had a habit of mutual help, and did not have to negotiate terms before starting work.

That is an ecology behaving like one. Not redundancy — Toyota had none — but recruitability: a population with enough shared capability that a lost function can be reassembled out of neighbours.

Toyota again, after 2011. The Tōhoku earthquake found the same group blind below tier two, which the Aisin recovery had been lucky enough not to need. The response was to build the database now known as RESCUE, which records supplier and part relationships several tiers deep across hundreds of thousands of records. The result reported afterwards was not a better forecast. It was a reduction in the time to locate an affected part from weeks to about a day. Visibility does not prevent a disruption. It changes the time constant of the response, and the time constant is the whole cost.

The Accord on Fire and Building Safety in Bangladesh, from 2013. More than two hundred brands funded a single inspection body covering upwards of sixteen hundred factories and two million workers. It found well over a hundred thousand safety hazards and drove the great majority of them to remediation. Set aside, for one paragraph, the politics of how it came about. Look at the structure: two hundred buyers stopped each auditing the same factory separately and paid once, together, for one inspection whose results they all accepted. That single decision changes the economics of enforcement by close to an order of magnitude, and we will compute exactly how much in the next movement.

The Responsible Minerals Initiative. Tin, tantalum, tungsten and gold enter the world's electronics from thousands of mine sites and pass through roughly three hundred smelters and refiners. The initiative audits the smelters, not the mines, and publishes a conformant list that hundreds of companies rely on rather than reproduce. That is not a compromise. It is the correct engineering answer: in a network shaped like an hourglass, you enforce at the waist.

Four cases, one pattern, and it is not the pattern people usually take from them. In each, somebody stopped treating the upstream as a line of contracts and started treating it as a population with a shape — dense here, narrow there, capable of reassembling itself under some conditions and not others — and then acted on the shape. Chapter II.05 gives the mathematics of why shape governs what a shock does; we use its conclusion here and do not repeat its derivation.

THE ARITHMETIC

What works, what does not, and where the line sits

Take a mid-sized vehicle-systems manufacturer: six hundred million dollars of revenue, three hundred and sixty million of external spend, sixty part families, four qualified sources per family. Everything below is computed in lib/verify/IV_07.py with its inputs printed.

First, the population. With four sources per family you have 240 direct suppliers. Each of them buys from others. Take a branching factor of five — conservative for a firm buying assemblies rather than commodities:

  tier 1        240 firms
  tier 2      1,200
  tier 3      6,000
  tier 4     30,000
  tier 5    150,000

This is why the word chain fails. By tier two you are not managing counterparties, you are managing a population, and populations are governed by distributions rather than by contracts.

Second, what it costs to see them. Identifying a node — confirming it exists, where it is, what it makes for whom — runs at roughly twelve hundred dollars once travel, verification and data upkeep are counted. But visibility decays as it descends, because you can only ask tier three through tier two, and roughly sixty percent of any tier will name its own suppliers:

  tier   firms      gross cost/yr    visible    cost per node FOUND
   1       240          288,000      100.00%              1,200
   2     1,200        1,440,000       60.00%              2,000
   3     6,000        7,200,000       36.00%              3,333
   4    30,000       36,000,000       21.60%              5,556

Tiers one to three cost $8,928,000 a year — 2.48 percent of external spend — and buy you thirty-six percent of the third tier. Extending to tier four raises the bill to $44,928,000, or 12.48 percent of spend, for a fifth of that tier. That is the honest price of the thing every risk committee asks for. It is not infinite, which is the good news. It is also not free, and nobody who demands full supply chain transparency has ever been handed that table.

Third, what one node does when it fails. Two measured cases.

Thailand, 2011. The floods caused damages and losses the World Bank put at 1.425 trillion baht, which at the 2011 average of 31.3 baht to the dollar is $45.5 billion — 13.17 percent of Thai GDP. Thailand assembled about forty-five percent of the world's hard drives. Global shipments fell from 176.6 million units in the third quarter to 120.0 million in the fourth, a fall of 32.05 percent. Divide:

  transmission coefficient  =  32.05%  /  45%  =  0.712

Seventy-one percent of one flood plain's share of world output became world shortfall inside a single quarter. Average drive prices rose by eighty to a hundred percent and did not return to their prior level for roughly a year — a fifty-dollar drive selling at ninety or a hundred. The damage was local and insurable; the shortage was global and nobody had underwritten it.

Naka, 2021. A fire started in a plating machine on the 300mm line of the Renesas Naka fab on 19 March. The line restarted on 9 April — twenty-one days dark — and pre-fire output was restored by 15 June, eighty-eight days after the fire. Renesas made roughly thirty percent of the world's automotive microcontrollers. Across 2021 the chip shortage in aggregate removed about 10.5 million vehicles from world production. At a declared world-average price of $35,000 that is $367.5 billion of output blocked. World semiconductor sales in 2021 were $555.9 billion; automotive took on the order of twelve percent, or $66.7 billion:

  blocked output / upstream industry revenue  =  367.5 / 66.7  =  5.51 x
       at a 10% automotive share   6.61 x
       at a 14% automotive share   4.72 x

The value of the cars that were not built was five and a half times the entire annual revenue of the industry that supplies their chips. The fire is one contributor among several — the sequence also included drought in Taiwan, a Texas freeze and a demand whipsaw — and the number above is the whole shortfall, not the fire's share of it. That is the point being made, and it does not need the attribution: the upstream node is small precisely in proportion to how much it gates.

Make that formal. Define the exposure ratio as the value of downstream output a node gates, divided by the node's own revenue from it:

  all chips in a car      $500 gates $35,000           70 x
  one automotive MCU      $3.50 gates $35,000       10,000 x
  one hard drive          $50 gates a $700 machine      14 x
  one Aisin P-valve       $10 gates a $15,000 car    1,500 x

A single microcontroller is one hundredth of one percent of the vehicle's price and stops it completely. Criticality is close to the inverse of spend share — which means a supplier list ranked by spend, the list every procurement function actually maintains, sorts your exposures in almost exactly the wrong order. Kraljic said as much in 1983 with a two-by-two; the division above says it with a number.

Fourth, how many sources — and here the arithmetic stops being comfortable.

The case for multi-sourcing assumes failures are independent. They are not. Sources share a region, a port, a mode of transport, a sub-tier supplier, a certification body, a sanctions regime. Model that directly: let each qualified source fail independently with probability p a year, and let all of them fail together with probability q — the common-mode term.

  P(all m fail)  =  q  +  (1 - q) · p^m        p = 0.05,  q = 0.015

At $18,000 a year to keep one supplier qualified, audited and administered, and $28 million as the cost of one line-stopping event:

  m   suppliers   P(all fail)   E[loss]/yr    admin/yr     TOTAL/yr
  1         60      0.064250     1,799,000   1,080,000    2,879,000
  2        120      0.017462       488,950   2,160,000    2,648,950
  3        180      0.015123       423,448   3,240,000    3,663,448
  4        240      0.015006       420,172   4,320,000    4,740,172
  8        480      0.015000       420,000   8,640,000    9,060,000

The optimum is two. And the reason is the floor: q = 0.015 is a wall no number of suppliers passes. The second source buys 4.68 percentage points of failure probability; the third buys 0.23; the fourth buys 0.012; the eighth buys effectively nothing. Put that in the only unit that decides anything — the cost of one event at which each additional source pays for itself:

  1 -> 2   pays if a stoppage costs more than      $23,083,088
  2 -> 3   pays if a stoppage costs more than     $461,661,769
  3 -> 4   pays if a stoppage costs more than   $9,233,235,373

Dual-sourcing is right for almost everyone. Quad-sourcing is right for a firm whose stoppage costs half a billion dollars — a large carmaker, a vaccine manufacturer, a grid operator. For the firm in this example it is a $2.1 million annual donation to the idea of resilience. Chapter III.10 computes what the buffer costs alongside this; the two decisions are usually made by different people and should be made together.

Now the honest negative, and it is a subtraction rather than a caveat.

Every source you add divides your share of each supplier's revenue, and your share of a supplier's revenue is your leverage over its conduct. Write λ for the leverage you retain one tier further up, and take it as falling roughly as 0.60 / m. Enforcement cost grows geometrically with the tier while the value of enforcing falls geometrically:

  C_n  =  c · N₁ · b^(n-1)             cost of auditing tier n
  V_n  =  V₁ · λ^(n-1)                 value of compliance at tier n

  they cross at   n*  =  1  +  ln(V₁ / C₁) / ln(b / λ)

With $6,000 an audit, $12 million a year as the value of full compliance at tier one, and four sources per family:

  ln(V₁/C₁)   = ln(12,000,000 / 1,440,000) = ln(8.3333)  = 2.1203
  ln(b/λ)     = ln(5.0 / 0.150)            = ln(33.3333) = 3.5066
  n*          = 1 + 2.1203 / 3.5066        =  1.605 tiers

Your code of conduct is one and six-tenths tiers long. Not because anyone is lax — because of a division. And the diversity is what shortened it:

  m = 1   λ 0.600   C₁   360,000   n* = 2.654
  m = 2   λ 0.300   C₁   720,000   n* = 2.000
  m = 4   λ 0.150   C₁ 1,440,000   n* = 1.605
  m = 8   λ 0.075   C₁ 2,880,000   n* = 1.340

Quadrupling the source count costs 1.049 tiers of enforcement reach. The resilient configuration and the influential configuration are different configurations, and the firm that has bought resilience has, in the same motion and without a line item, sold most of its ability to govern conduct upstream. Reaching tier three from here would require V₁/C₁ ≥ (b/λ)² = 1,111 — a value of compliance of $1.6 billion a year, a hundred and thirty-three times what it is. No audit budget closes that.

What does close it is not an audit. Three levers, computed:

  as found                              n* = 1.605
  shared audit across 8 buyers          n* = 2.198
  deep partnership, λ back to 0.60      n* = 2.000
  both together                         n* = 2.981
  both, at a waist where b = 1.5        n* = 5.583

And the waist is real, not hypothetical. Roughly eight thousand mineral sources pass through about three hundred smelters. Auditing the sources costs $48 million a year; auditing the waist costs $1.8 million — a 96.25 percent saving — and shared across eight buyers it is $225,000, against a tier-three compliance value of $270,000. It pays, at the exact tier that the geometric arithmetic above declares unreachable. The Responsible Minerals Initiative is not a compromise with the ideal. It is the ideal, correctly located.

DREAM

What becomes ordinary

In a firm that has done this, the supplier list is not the map and nobody pretends it is. There is a second document, kept by the same people, that says how many firms sit behind each family, which of them are shared with competitors, and where the population narrows to a waist. It is updated quarterly and it is three pages long, because a shape is small even when a population is not.

Criticality is ranked by exposure ratio and not by spend, so the part that appears at the top of the risk register is a three-dollar microcontroller, and nobody in the room finds that odd any more. Spend is still tracked; it answers a different question, and the two lists are printed side by side precisely because they disagree.

The number of sources per family is a computed number with a written justification, not a policy. Families whose stoppage costs tens of millions carry two sources. Families whose stoppage costs hundreds of millions carry four. Families whose stoppage costs a fortnight of inconvenience carry one and a good relationship, and the relationship is understood as the resilience mechanism rather than as a sentimental leftover.

Audits are not duplicated. When a factory is inspected, the inspection is purchased once by a group of buyers and read by all of them, and a firm that insists on its own private audit of the same site is asked, in the normal course of budgeting, what the extra five thousand dollars is buying that the shared report does not contain. Usually there is no answer, and the shared report stands.

The deepest tiers are reached by price, not by inspection. A supplier four tiers up that meets the standard borrows against the buyer's credit rather than its own, and the discount it receives is larger than what verification costs it. Compliance stops being an imposition travelling down a chain of contracts and becomes a rate, and rates propagate through populations without anybody enforcing anything.

And when something burns down — something always burns down — the first useful question is answered in a day rather than a month, because the map exists. Nobody calls this resilience. They call it knowing where things come from.

DESIGN

The structure that gets there

Four mechanisms, in the order they pay.

One: map by exposure, not by depth. Do not attempt tier three across the board; the table above prices that at $7.2 million a year for a third of a tier. Instead compute the exposure ratio for every family, take the top decile, and map those to exhaustion. Six families mapped four tiers deep costs on the order of a tenth of mapping sixty families three tiers deep, and it covers the exposures that can actually stop you. Denominator stated plainly: this leaves ninety percent of families mapped one tier, and you should say so in the paper rather than let the board infer coverage you do not have.

Two: set the source count from the event cost, in writing. One line per family: stoppage costs X; the m-th source pays above Y; therefore m. That single table converts the most political argument in procurement into arithmetic, and it defends both directions — it is as much a defence of single sourcing where the event is cheap as it is of quad sourcing where the event is catastrophic.

Three: buy your audits in a group. This is the highest-return move available and it requires no technology. Eight buyers sharing one inspection take the cost per node from $6,000 to $750 and move enforcement reach from 1.605 tiers to 2.198 — more than half a tier of additional reach, bought with a phone call to your competitors. Every mature version of this exists already: the Accord, the Responsible Minerals Initiative, Sedex, amfori BSCI. Joining one is a procurement decision, not a strategy.

Four: recover the leverage you sold, selectively. Leverage λ is the other term in the denominator and it is the one nobody manages. It rises with share of the supplier's revenue, with contract length, with the presence of joint engineering, and with whether you are the customer they would protect in a shortage. You cannot have deep leverage with two hundred and forty suppliers. You can have it with the thirty that matter — which is what deep partnership means operationally, and why the Aisin network could reassemble a lost part in five days while a four-sourced commodity family took a year to recover from a flood. Concentrate the relationship where the exposure ratio is high and the event cost is moderate; concentrate the redundancy where the event cost is extreme. They are different families and the mistake is applying one policy to both.

And the sequence. Exposure ratios first, because they are a spreadsheet afternoon and they reorder everything downstream. Source counts second, because they follow from the ratios. Shared audit third, because it is the cheapest win and it takes a quarter to join. Leverage recovery last, because it is a multi-year renegotiation and it should be aimed only where the first three have already shown it matters.

DESTINY

How it holds when you stop pushing

It holds on three things and fails on four.

It holds because the map has a second owner. A supply map maintained by one person decays at the rate that person changes jobs. Put it in the quarterly operations review with a named owner and a named deputy and it survives.

It holds because the shared audit has a subscription. A cost that recurs automatically is a cost nobody re-litigates. The single best structural feature of the group-inspection model is that it arrives as an invoice rather than as a proposal.

It holds because the financing is priced. Once a supplier's borrowing rate depends on its verified standard, compliance becomes self-interested, and self-interest does not need a champion.

Now the failure modes, named without softening.

The map goes stale silently. Suppliers change their own suppliers without telling you, and a map that is not refreshed reports the structure of two years ago with complete confidence. Date every node. A node not re-verified within twelve months reads as unknown, not as unchanged.

The exposure ranking gets quietly re-sorted by spend. It happens whenever the list is handed to someone who was not in the room when it was built, because spend is the ordering every system defaults to. Write the ratio into the column header.

The shared audit becomes the ceiling rather than the floor. A group standard is the minimum everyone accepted, and it will be lower than what your best suppliers already do. Use it as the base layer and keep a thin private layer for the twenty relationships where you actually have leverage — this is the one place duplication is worth paying for.

And the largest one: the arithmetic gets used to justify stopping. n\ = 1.605 is a statement about a particular configuration, not a law of nature, and it is trivially misread as upstream conduct is not our problem.* The same equation says shared audit plus partnership depth reaches 2.981 and a waist reaches 5.583. A number that shows a limit also shows what moves it; a firm that quotes only the first half is using arithmetic as an alibi.

DELIGHT

What it feels like

The pleasure is specific and it arrives the first time the map answers a question faster than a phone call. Something has happened somewhere — a port, a fire, a border — and instead of the usual fortnight of emails, somebody opens a document and says: four families, two of them dual-sourced within the region, one of them is going to hurt. The relief in the room is not about the news. It is about no longer being in fog.

There is a second and better pleasure, which is the first time a competitor's procurement director agrees to share an audit. It feels faintly illicit and it is entirely ordinary — you are both paying twice for the same inspection of the same factory and both of you knew it. The conversation takes ten minutes. The savings are permanent, and the reach of the standard you both believe in goes up by half a tier the day you sign.

And the deepest one is quieter. When you start ranking by exposure rather than spend, the small firms come into focus — the three-person shop that makes the one part, the family business with the only heat-treatment furnace of its kind. They were always load-bearing. You simply had no column that showed it, and now you do, and they can tell you have it. That changes how they treat you in a shortage, which is the only insurance that has never had a premium.

OPERATIONALIZE THIS

At the level of finance

The instrument: a deep-tier payables finance programme with the discount priced off the verified standard.

This is not a pilot concept. Buyer-anchored supplier finance is ordinary treasury machinery, and the version that ties the rate to a verified environmental and labour score has been run at scale — the IFC's Global Trade Supplier Finance programme with Levi Strauss & Co. did exactly this from 2014, giving better pricing to factories with better verified scores. What follows is that structure, stated for a treasurer, with the number that decides it.

The mechanics.

Structure. The buyer's bank pays approved supplier invoices early against the buyer's credit rating. The supplier receives cash at day five instead of day sixty-five and pays a discount set by the buyer's risk, not its own.
The pricing tier. Two rates, not one: a base rate for any approved supplier, and a lower rate for a supplier holding a current verified standard from the group inspection scheme. The differential is the enforcement mechanism and it costs the buyer nothing — it is the bank's pricing, not the buyer's cash.
Balance-sheet treatment. Get this right early or it will unwind the programme. Trade payables stay trade payables only while commercial terms are unchanged and the supplier's participation is genuinely voluntary; extend terms as a condition of access and the auditors will reclassify the balance as borrowings. Disclose the programme's size and the payment terms — IAS 7 and IFRS 7 now require supplier-finance disclosure, and a reclassification discovered at year end is how these programmes die.
Extension to deep tiers. Tier-one suppliers nominate their own critical suppliers into the same facility. This is the whole point: it is the only mechanism in this chapter that reaches tier three without the buyer holding a contract there.
Counterparty. Your existing cash-management bank first, because the credit work is already done. A fintech platform only once volumes justify it, and never as the sole route — a single platform holding the payment rails for your deep tiers is a new single node with a very high exposure ratio.

The number that decides it. One division, and it belongs on the front page:

  supplier's own working capital rate        12.00 %
  rate against the buyer's credit             5.00 %
  spread                                       700 bp
  payables financed                          60 days

  benefit per dollar of invoice  =  0.07 × 60/365  =  1.1507 %

  breakeven annual trade  =  verification cost / 1.1507 %
        solo verification   $6,000  ->  $521,429
        shared verification   $750  ->   $65,179

Above the breakeven, a supplier pays for its own verification out of the financing spread and the standard needs no enforcement at all. Below it, no audit budget will ever reach. And the whole strategic question of this chapter reduces to which side of that line your tier two sits on: 1,200 tier-two nodes carrying half your spend average $150,000 of trade each — below the $521,429 solo breakeven and comfortably above the $65,179 shared one. The group inspection scheme is not an ethics programme. It is the thing that moves your second tier across the line where compliance becomes self-financing.

The first ninety days.

Day	Action	Artifact
1–15	Compute the exposure ratio for all 60 families	The ranked list, beside the spend list
16–30	Map the top decile to exhaustion; date every node	The map, three pages
31–45	Set source count per family from event cost	One line of arithmetic per family
46–60	Join a group inspection scheme; cancel duplicate audits	The subscription
61–75	Agree two-tier pricing with the cash-management bank	Facility term sheet
76–90	Onboard tier one; open nomination of tier two	First tier-two supplier funded

APPRECIATIVE QUESTIONS

Twelve, for a room

Discovery — what is already working

Think of a time a supplier of ours went out of their way during a shortage and protected us when they did not have to. What was the relationship like before that day, and who built it?
Which part that costs us almost nothing would stop us completely — and who in this room already knew that without looking it up?
Where do we already know something about our second or third tier that never made it into a system? Who holds it, and how did they come to know it?

Dream — what becomes possible

If we could answer "who makes this, and where" in a day rather than a month, what decision would we make differently first?
Imagine our competitors and us paying once, together, for the inspection of a factory we all use. What would we do with the money and the goodwill?
If our best suppliers borrowed against our credit rather than their own, what would they build that they currently cannot afford to build?

Design — what we build

Which ten families deserve a deep partnership rather than a fourth source — and what would we have to give up to make that a real partnership?
What is the smallest map that would have answered our last three surprises?
Who would we call first to propose a shared audit, and what is the one sentence that would make them say yes?

Destiny — how it holds

What would tell us, twelve months from now, that our map had quietly gone stale — and who would notice it first?
What would have to be true for the exposure ranking to survive the next reorganisation of procurement?
If this became ordinary across our industry, what would we want to be known for having done first?

WORKS CITED

Accord on Fire and Building Safety in Bangladesh (2013–2018). Annual Reports and Quarterly Aggregate Reports. Amsterdam/Dhaka.

Choi, T. Y., Dooley, K. J. and Rungtusanatham, M. (2001). "Supply Networks and Complex Adaptive Systems: Control Versus Emergence." Journal of Operations Management, 19(3), 351–366.

Choi, T. Y. and Hong, Y. (2002). "Unveiling the Structure of Supply Networks." Journal of Operations Management, 20(5), 469–493.

Haraguchi, M. and Lall, U. (2015). "Flood Risks and Impacts: A Case Study of Thailand's Floods in 2011 and Research Questions for Supply Chain Decision Making." International Journal of Disaster Risk Reduction, 14, 256–272.

International Finance Corporation (2014). Global Trade Supplier Finance: Sustainability-Linked Pricing with Levi Strauss & Co. IFC, Washington DC.

International Labour Organization (2016). Purchasing Practices and Working Conditions in Global Supply Chains. INWORK Issue Brief No. 10. ILO, Geneva.

Jacobides, M. G., Cennamo, C. and Gawer, A. (2018). "Towards a Theory of Ecosystems." Strategic Management Journal, 39(8), 2255–2276.

Kraljic, P. (1983). "Purchasing Must Become Supply Management." Harvard Business Review, 61(5), 109–117.

Locke, R. M. (2013). The Promise and Limits of Private Power: Promoting Labor Standards in a Global Economy. Cambridge University Press.

Nishiguchi, T. and Beaudet, A. (1998). "The Toyota Group and the Aisin Fire." Sloan Management Review, 40(1), 49–59.

Responsible Minerals Initiative. Responsible Minerals Assurance Process: Conformant Smelter and Refiner Lists. Successive editions.

Sheffi, Y. (2005). The Resilient Enterprise: Overcoming Vulnerability for Competitive Advantage. MIT Press.

Simchi-Levi, D., Schmidt, W. and Wei, Y. (2014). "From Superstorms to Factory Fires: Managing Unpredictable Supply Chain Disruptions." Harvard Business Review, 92(1), 96–101.

World Bank (2012). Thai Flood 2011: Rapid Assessment for Resilient Recovery and Reconstruction Planning. World Bank, Bangkok.

World Semiconductor Trade Statistics (2022). Annual Semiconductor Sales, 2021. WSTS.

Within this edition. Chapter II.05, Complexity and Economic Networks, for the percolation and cascade mathematics used but not repeated here. Chapter III.10, Reserves and Resilience, for buffer sizing, the loss function and the newsvendor — the decision this chapter's source-count arithmetic sits beside.

Note on figures. Population, visibility cost, transmission coefficient, exposure ratios, source-count optimum, enforcement depth n*, the waist saving and the financing breakeven are all computed in lib/verify/IV_07.py, which prints every input with its unit and source before dividing. The 2021 vehicle shortfall is the whole shortfall for the year and is not attributed to the Renesas fire alone; the average vehicle price and the automotive share of semiconductor sales are declared assumptions and their sensitivity is printed.