Chapter 6 — Risk Management: Who Owns Which Risk, What It Costs, and How Contingency Actually Works

The Hook: Forty-One Rows and Nobody's Name On Them

The Guaranteed Maximum Price (GMP) for the Northgate Outpatient Pavilion was due to Meridian Health System on a Friday. On Wednesday afternoon it sat in Tomás Reyes's office, printed, bound, and unsigned, because our chief estimator would not release it.

Tomás does not raise his voice. He had a spreadsheet open on the second monitor, forty-one rows long, and he had turned the monitor forty-five degrees toward the door so anyone who walked in would have to look at it.

Nadia Haddad walked in. Our Vice President of Operations does not usually come down to estimating.

"Tomás. The number is forty-seven five. I've seen it four times. Pri Sethi has seen it twice. What are we doing?"

"Thirty-four of these rows have a name in the owner column," Tomás said. "Seven don't."

"Then the seven are small."

"Two of them are small." He scrolled. "Row 19 is the imaging equipment package. Meridian hasn't selected the MRI unit. Every unit on their short list has a different pad, a different slab depression, a different shield wall, and a different electrical feed. Row 19 has a cost impact and a probability and no name."

Nadia looked at the screen a while. "Meridian owns their own equipment selection. That's obvious."

"It's obvious to us," Tomás said. "It is not written down. If it is not written down, then in eleven months it will be obvious to nobody, and we will be having a very expensive conversation about what everyone assumed on a Wednesday in February."

I was in the room. I am not proud of what I said, which was some version of we can sort that out later. Nadia asked what "later" cost. Tomás pulled up the row: a 60% chance of a change, an estimated impact of $185,000, and nine calendar days.

"So the expected value is a hundred and eleven thousand dollars," Nadia said. "And you're holding a forty-seven-and-a-half-million-dollar number over it."

"I'm holding it over the seven blanks," Tomás said. "The hundred and eleven thousand is the cheap part. The expensive part is that if nobody owns row 19, the first written document about it gets created after the argument starts. You know what that document is worth."

He won. It took a day and a half. Pri Sethi, Meridian's owner's representative, agreed in an exhibit to the GMP that equipment selection — and any resulting structural, electrical, and shielding modifications — was owner scope, funded from Meridian's own contingency, with a defined pricing procedure. The number went out Friday.

Eleven months later, Meridian's imaging vendor selected a different MRI unit. It needed a deeper depressed slab, added structural framing, additional radio-frequency shielding, and a larger electrical feed. Pri gave a verbal go-ahead on a Thursday. Our assistant superintendent, under schedule pressure, let the concrete crew build it Monday. No written directive. No agreed price. No time-impact analysis. No time-and-material tickets for the first four days.

That became Change Order (CO) #14. Meridian's understanding of the cost was "about $60,000." Our actual cost was $186,400. We could substantiate $121,000 with contemporaneous records. Eight weeks later we settled at $142,750 and got four of the nine days we asked for. We ate $43,650.

Here is what matters. We had already identified that risk. It was row 19, with a probability, a cost, and — after Tomás dug in — an owner. The register did its job. What failed was everything downstream: the trigger nobody watched, the procedure nobody followed, and the four days of missing tickets. A risk register does not prevent risk. It converts an argument you were going to have anyway into a conversation you have early, in writing, while it is still cheap. Then somebody still has to run the play.

🏃 Fast Track: If you already build registers, skim §6.1–6.4 and slow down at §6.5 (the honest limits of a heat map, and why the sum of most-likely values is systematically wrong), §6.6 (the threshold concept — what contingency actually is, sized both ways, with the drawdown curve), and §6.9 (the ethical lines and the safety exception).

🔬 Deep Dive: Arithmetic conventions are collected in Appendix A. The clause names in §6.3 are decoded at length in Appendix G. Contingency drawdown as a reporting discipline is picked up in Chapter 28; schedule buffer as a production discipline in Chapter 27.

What this chapter gives you. You will be able to write a defensible risk register; allocate each row to a party for a stated reason; price a risk two ways; size a contingency bottom-up and top-down and explain the gap in writing; read a drawdown curve well enough to know eight weeks early that a job is in trouble; and recognize the habits that cause competent people to leave the biggest risk off the list.

6.1 What a Risk Actually Is — and Two Things That Are Not Risks

A risk is an uncertain event that, if it happens, has a cost. Three moving parts, and you need all three:

1. It is uncertain. If it is certain, it is not a risk — it is scope, and it belongs in the estimate at full value. A tower crane you know you need is not a risk. It is $340,000.
2. It is an event. It has a trigger, a moment, a thing that happens. "The market is bad" is not an event. "Structural steel mill pricing moves more than 6% between GMP and buyout" is an event.
3. It has a consequence you can express in money, days, or both. If you cannot say what it costs, you have not finished thinking about it.

Most people stop after the first part. The result is a register full of anxieties instead of risks: labor, weather, the architect. You cannot manage an anxiety. You cannot assign it, price it, or watch for it.

A risk can also be upside. If the concrete package comes in under the estimate because two qualified bidders are hungry, that is an uncertain event with a consequence, and it belongs on the register alongside the bad ones. A register that tracks only downside teaches the team that raising your hand is always bad news — exactly the culture problem we get to in §6.9.

Risk, uncertainty, and ignorance

These three words get used interchangeably on job sites. They should not be. They call for completely different management.

That third row is the humbling one. On Northgate, the most expensive event of the whole job was not on the forty-one-row register. It was a submittal sitting in a pile. We had rows for the steel market, the steel fabricator, and the steel erector. We had no row for our own process, because the register was pointed outward at the world instead of inward at ourselves. That is the spine of Case Study 1.

💡 Aha moment. A risk register is not a forecast. It is a list of the conversations you are going to have anyway — held early, in writing, while they are still free. The forecast is a by-product.

🔄 Check your understanding. A project manager writes: "Risk: supply chain problems. Probability: high. Impact: significant." Name three things wrong with it, and rewrite it properly.

6.2 Five Answers to Every Risk — All Five Applied to One Risk

🧩 Productive struggle. Take three or four minutes before reading on. Northgate's site is 6.2 acres, sloping, urban-edge. Mass excavation is 44,000 CY cut and 12,000 CY fill, leaving 32,000 CY of net export. The geotechnical report has good borings in the building footprint, thin coverage on the south half, and the old aerial photographs show something used to be there. Write down every different thing Kestrel could do about the possibility of unsuitable subsurface material — not the best one. All of them.

There are exactly five things you can do with any risk. Not five steps — five alternatives. Most people treat them as a checklist and do a little of each, badly.

1. Avoid. Change the plan so the risk cannot occur. Here: shift the footprint onto the north half where the borings are good, or redesign the foundations as drilled piers bearing below the suspect zone so fill quality stops mattering. Avoidance is the only response that takes a risk to zero, and it is usually the most expensive, because it costs you the thing you wanted. Piers run roughly $600,000 more than 148 spread footings on a job this size.

2. Transfer. Move the financial consequence to somebody else — by contract, bond, or insurance. Here: a differing-site-conditions clause moves the cost to Meridian, or a "contractor has satisfied itself as to all site conditions" clause pushes it to the earthwork subcontractor. Transfer does not reduce risk. It decides who writes the check. And it is never free: the party accepting it prices it, and the less they can control or bound it, the more they charge.

3. Mitigate. Reduce the probability, the impact, or both. Here: drill eight additional borings and four test pits across the south half for $34,000 and have the geotechnical engineer map the old fill before the GMP is set. That does not stop bad soil from existing. It converts an unbounded unknown into a known quantity you can price and carry as scope.

4. Accept. Carry it, on purpose, with money behind it. Here: after the extra borings you conclude there is a 35% chance of roughly $420,000 of undocumented fill, and you fund it. Acceptance is a legitimate adult response — but only when it is active: written, priced, funded, assigned. Acceptance without funding is not acceptance. It is hope with a spreadsheet.

5. Exploit. Act to make an upside more likely. Here: if the borings come back clean and native material is suitable, rebalance cut and fill and cut the export. At roughly $11.83/CY, reducing export by 6,000 CY puts about $71,000 back in the job. Exploiting means writing the row that says if the soils are good, here is what we do and who decides.

Two rows deserve a second look. The mitigate row turns a $147,000 expected exposure into a $16,000 one for $34,000 of drilling — the best trade on the table, and available only before the GMP is set. Two months later that same $34,000 buys nothing, because by then you are digging. The third row looks free and is not: it costs $95,000 to $140,000 inside the sub's bid, and it transfers a risk the sub may not survive. We return to that in §6.9, where it stops being a pricing question and becomes an ethics question.

🔍 Why this works. Mitigation before commitment is cheap for a structural reason, not a psychological one. Before the GMP, information changes the price: whatever you learn is converted into scope and priced at unit rates in a competitive market with several bidders. After the GMP, information changes only the loss: whatever you learn becomes a change order priced in a market with exactly one bidder — you — negotiating with a counterparty who has every reason to disbelieve you. The same fact is worth different money depending on when you learn it.

🔄 Check your understanding. A subcontractor's bid comes in $110,000 higher than the others because the subcontract makes them responsible for a condition they cannot inspect. Which of the five responses is the general contractor using, which one is the subcontractor using, and who is actually holding the risk?

6.3 Risk Allocation Is the Core Contractual Act

Every textbook states the principle: a risk should be borne by the party best able to control it. It produces efficient contracts, because the party who can prevent a loss has the incentive to prevent it, and nobody pays a premium to insure something they cannot influence.

Every practitioner makes the observation: in practice, risk sits with whoever has the least negotiating leverage. A public owner with forty bidders and a standard contract can push almost anything downhill. A general contractor with a signed subcontract form and a sub who needs the work can push it further. The risk rolls until it hits somebody who cannot say no.

Both are true at once, and holding both in your head is most of what makes a good negotiator. When you read a clause, ask two questions: does this sit with the party who controls it? and if not, what did the shifting cost? Because it always costs something. It just may not show up as a line item.

⚖️ What the contract says. Risk allocation is not an abstraction floating above the contract. It is the contract. Strip out the recitals and the boilerplate and a construction agreement is a list of answers to one question: "and if that happens, who pays?" Every clause below exists because somebody, once, had a very expensive argument about a specific event.

Two things to notice.

The "should" and "usually" columns agree more often than cynics claim. Standard industry forms have been beaten into shape by a century of litigation and mostly land in sensible places. When a contract deviates hard from the "should" column, that deviation is the most important thing in the document, and it is where your money is. Read the modifications to a standard form before you read the form.

The "cost to shift" column is real money even when it never appears as a line item. Delete the differing-site-conditions clause and no invoice arrives. What happens instead is that every bidder quietly loads the earthwork package, all the bids come in higher, and the owner pays for the risk whether or not the bad soil exists. The owner has bought insurance from a contractor who is a worse insurer than an actual insurer and charges more for it. That is the most common way owners waste money in construction, and they almost never see it happen.

🔄 Check your understanding. Cover the allocation table. Unusually severe weather stops work for nine days on a job with a standard set of general conditions. Who gets the time, who gets the money, and what would have to be different in the contract for that answer to change?

6.4 Building the Register: Rows, Not Worries

A risk register has five columns of analysis and five of management. Most registers I see have the first five and none of the second — which is why they get built once during preconstruction and never opened again.

Where the rows come from, in descending order of value: your own job cost history — the last twenty closeout reports are the best risk database on earth and almost nobody reads them; if drywall subs defaulted on two of your last thirty jobs, your base rate is about 7%, not "unlikely." A structured document review with a pen, hunting conflicts, missing details, deferred design, and performance specifications that quietly hand you a design obligation (Chapter 7). The site visit — access, neighbors, slope, utilities, soils, what was there before. Trade partner interviews, because Sofia Marchetti at Cardinal Mechanical will tell you about a plenum-depth problem four months before your model does, if you ask. And last, not first, a checklist — terrible at finding your unique risks, excellent at catching ordinary ones you forgot.

The Northgate register at GMP conversion

Here is the working register that came out of Tomás's forty-one rows after consolidation — one register, shown in two halves for readability. Same fifteen rows, same IDs.

Half 1 — assessment

Half 2 — management

Read the Owner column again. Every entry is a human being with a name, not a department. "Operations" cannot watch a trigger. Margo Deacon can. And read the Trigger column: a trigger is the observable event that converts a row from maybe into now, and it must be something a person can see and declare on a Tuesday. "Manning falls 15% below the curve for two consecutive weeks" is a trigger. "If things start going badly" is not.

🔄 Check your understanding. R-11 has a probability of only 12% but earned a place on a fifteen-row register while dozens of likelier items did not. Why?

6.5 Analyzing the Rows: Heat Maps, Expected Value, and What Monte Carlo Actually Does

6.5.1 Qualitative: the probability × impact matrix, and its four honest limits

The most common risk tool in construction is a 5×5 matrix. It is fast, needs no data, and everyone can read it. These are the bands we used:

And the register plotted on it:

The matrix does two things well: it gets everyone looking at the same picture in ninety seconds, and it puts R-13 — incomplete finish documentation, score 4 × 4 = 16 — in the top-right corner where it belongs. That row had the largest expected value on the register, and the matrix found it.

Now the honest part, because it also lies in four specific ways. It does arithmetic on labels — "3 × 4 = 12" looks like math, but those are rank positions, not quantities, and a score of 12 is not twice as bad as a score of 6. It compresses ranges: R-01's impact is $420,000 and R-02's is $310,000, and both score "4." It ranks things backwards, and you can watch it happen — R-11 scores 2 × 5 = 10 and R-07 scores 3 × 3 = 9, so the matrix says R-11 is worse, while the expected values say the opposite ($76,800 against $115,000). And it is blind above the top box: R-11's band is "over $500,000," but a drywall default at 55% complete on 412,000 SF of gypsum board runs from about $400,000 to well over $1,400,000 depending on when it happens and how much of the sub's second tier walks out with him. The matrix caps at exactly the place where the tail that could kill you begins.

Use the matrix to communicate. Never use it to decide. Once a decision is worth more than about $50,000, do the arithmetic.

6.5.2 Quantitative I: expected monetary value

Expected monetary value (EMV) is the simplest useful piece of risk math in construction:

EMV = probability of the event × cost if it occurs

EMV is a portfolio number, not a project number. You will never spend $147,000 on R-01. You will spend $0 or $420,000. EMV is honest across many rows and many jobs and meaningless for one row on one job. Note too that R-04 (30% × $260,000 = $78,000) and R-11 (12% × $640,000 = $76,800) are financially identical on paper and not at all identical in life: one costs you a bad quarter, the other a bad year.

Now the most useful thing you can do with a register — separate the money by whose it is:

$326,400 + $139,500 + $735,650 = $1,201,550. Hold onto that $735,650 — §6.6 does something important with it.

💰 Money check. Allocation is what kept CO #14 from being much worse. Row R-03 was assigned to Meridian in a written GMP exhibit, so when the change arrived eleven months later the argument was never about whether Meridian owned it — only about how much. That exhibit was worth roughly $142,750. The $43,650 we lost was the documentation half of the job, not the allocation half. You need both, and they are different skills.

6.5.3 Quantitative II: three-point estimating and PERT

EMV works for events that either happen or do not. It is the wrong tool for something that will happen but whose magnitude you do not know — a duration, a unit price, a quantity. For those, use three points.

PERT weighted mean = (O + 4M + P) ÷ 6 where O = optimistic, M = most likely, P = pessimistic

Worked on Northgate's structural steel erection, 985 tons:

PERT mean = (62 + 4 × 72 + 95) ÷ 6 = (62 + 288 + 95) ÷ 6 = 445 ÷ 6 = 74.2 work days Rough standard deviation ≈ (P − O) ÷ 6 = (95 − 62) ÷ 6 = 5.5 work days

What it means: 72 work days is not the expected duration. It is the most common one. The expected duration is 74.2 WD, because the bad tail is longer than the good tail. Build the schedule on 72 and you have roughly a four-in-ten chance of hitting it, not the five-in-ten everyone assumes.

The same method works on money. Northgate's 32,000 CY of net export, with an optimistic $9.60/CY (a nearby fill site takes the material), a most likely $11.25/CY (the estimate's haul and disposal), and a pessimistic $16.40/CY (the fill site closes; longer haul plus tipping fees):

PERT unit cost = (9.60 + 4 × 11.25 + 16.40) ÷ 6 = 71.00 ÷ 6 = $11.83/CY

At 32,000 CY that is $378,560, against a most-likely estimate of 32,000 × $11.25 = $360,000. The export line is understated by about $18,560 if you carry the most-likely price. That is not an error. It is the ordinary, systematic optimism of using the middle of a right-skewed range — and it happens on every line of every estimate you will ever write.

6.5.4 Quantitative III: what Monte Carlo actually does

Monte Carlo simulation sounds exotic and is not. You give the computer a range for each uncertain item instead of a single number. It picks one value at random from each range, adds them up, and writes down the total. It does that a few thousand times. Then it shows you the distribution of all those totals — not one answer, but a picture of how the answers spread. That is the entire concept; you do not need to build one to understand what it tells you.

Here is why it matters, on Northgate's foundation-to-steel path. Five activities, in work days:

The schedule you would publish is 6.2 work days optimistic before anything goes wrong at all — and that is one path with five activities on it. Now the simulation. Eight iterations, each a single random draw from each range:

Eight runs, average 175.4 WD; three of them — 1, 3, and 7 — came in at or under the published 168. Run it five hundred times and the picture stabilizes:

What it means: the schedule that says 168 has roughly a 31% chance of being met. Getting to 80% confidence takes 187 work days — nineteen more, or about 27 calendar days. At Northgate's total daily exposure of $10,650/CD, that gap is $287,550 of unfunded exposure sitting inside a Gantt chart that looks perfectly reasonable.

Why is the sum of most-likely values systematically too optimistic? Two mechanisms that stack.

Skew. A construction activity can go badly wrong in ways it cannot go wonderfully right. Excavation can take three times as long as planned; it cannot take one-third as long, because there is a physical floor on how fast you move 44,000 CY. Every activity's distribution has a long right tail and a short left one, so the mean sits above the mode. Add fifty such activities and the errors do not cancel — they accumulate in one direction.

Merge bias. Interior finishes on Northgate cannot start until enclosure, MEP rough-in, and elevator installation are all complete. Three paths converge. If each has an 80% chance of finishing on time and they are independent, the chance all three land on time is 0.80 × 0.80 × 0.80 = 0.512 — about 51%. Every feeder looked safe; the merge point is a coin flip. Add a fourth and you are at 41%. This is why a schedule with many parallel paths converging on one milestone is far riskier than the same activities in a line, and why the critical-path calculation alone will not show it to you. Chapter 14 builds the network; this is what the network does not tell you.

🔄 Check your understanding. Your scheduler shows you a 42-activity CPM (critical path method) schedule finishing exactly on the contract date with two days of float. Without running any simulation, what do you already know about the probability of finishing on time?

6.6 Contingency, Properly

🏗️ From the field. On the convention-center expansion, years before Northgate, I carried a budget line called "Misc. Contingency — $400,000." It was not attached to anything; a job that size ought to have some room in it, and $400,000 felt like room. By month nine it was gone, and I could not tell you today — or my CFO then — what it had gone to. Forty or fifty small draws, none of them wrong and none of them written down against a named risk. Then the actual risk hit: a curtain-wall sequencing problem that cost $310,000 to work through. There was nothing left to absorb it, so it came out of fee. I did not lose that money in month eleven. I lost it in month one, when I wrote a line item with no rows behind it.

🚪 Threshold concept. Contingency is not padding. It is a priced, owned, drawn-down reserve for identified risk. Money set aside without a named risk attached to it is either fat or a lie — and both get spent.

Before you understand this, contingency is a cushion — the number you add at the end so the estimate feels safe. Its size is a matter of nerve: a confident estimator carries 2%, a cautious one carries 5%, and the difference is temperament. When it gets spent, it was needed; when it does not, you got lucky. Nobody can say, at any given moment, whether what remains is enough, because there is nothing to compare it against.

After you understand this, contingency is a funded liability. Every dollar traces to a row on a register: a named event, with a probability, an impact, an owner, a trigger, and a response. It has a schedule of its own — you can say what it should look like at 40% complete and be right or wrong about it. It is drawn down by written authorization, one row at a time, and every draw retires or reduces a specific exposure. And because each dollar has a name, you can answer the only question that matters in month seven: is what is left enough for what is left?

One question separates the two worlds, and you should ask it of any contingency you are ever shown: "Show me the rows." If there are no rows, there is no contingency. There is only a number somebody felt good about.

The five buckets, and why confusing them costs money

"Contingency" gets used for five different things that are funded differently, owned differently, and spent under different rules. Mixing them is the most common money mistake in preconstruction.

Two of those get abused constantly. Spend Northgate's $575,200 escalation allowance on a productivity problem and you will have nothing left when steel actually moves — and a hard time explaining it in an open-book GMP. And estimating contingency should shrink as drawings progress; if your design contingency is the same percentage at 100% construction documents as it was at schematic design, you are not estimating, you are guessing at a constant rate. For rough orientation only — these vary widely by market, project type, and company, and your own history beats any published range — design-stage contingencies commonly start in the high teens or low twenties at conceptual level and step down into the single digits as drawings complete; construction contingency on a negotiated commercial building commonly lands in the 2–5% range of the cost of work; owner's contingency is typically larger than either, because owners change their minds.

Sizing it two ways, on Northgate

Top-down. The traditional method: a percentage of the cost of work.

That is $1,320,000 ÷ $43,800,000 = 3.01%. Fast, defensible in a boardroom, and completely uninformative — it tells you nothing about what the money is for, so it can never tell you whether it is enough.

Bottom-up. Add the expected values of the rows Kestrel actually owns: $735,650 (§6.5.2).

So the register says $735,650 and the GMP carries $1,320,000 — a gap of $584,350, with the contingency at 1.79 times the bottom-up number. Which one is wrong? Neither. Here is the reconciliation, and the reconciliation is the deliverable. Any estimator can produce two numbers; a good one explains the space between them in writing.

1. An expected value is a mean, and you do not fund to the mean. Fund a portfolio of risks at its average and you run out roughly half the time. You fund to a confidence level. When Wei Chen ran the ten Kestrel-owned rows through a simulation treating each impact as a range rather than a point, the output looked like this:

$1,320,000 sits just above the P90 of the identified risks.

2. The register is incomplete, and you know it. Go back to §6.1. The register captures risk, captures some uncertainty as a range, and captures none of the ignorance. Kestrel's closeout data showed that somewhere between a fifth and a third of contingency draw went to events that had never appeared on any register. So $1,320,000 is roughly P90 of the known risks and, allowing for what is not on the list at all, closer to P70–P75 of all risk. Tomás describes it as "ninety percent confident about the things I thought of, seventy percent confident overall, and honest about the difference."

3. Mitigation costs belong in the estimate, not the contingency. The $34,000 of extra borings, Grace's clash-detection hours, the deposit holding the curtain-wall fabrication slot — those are certain costs incurred to reduce uncertain ones. They are scope. Fund mitigation out of contingency and you will underfund mitigation, because the money feels like it belongs to somebody else.

🔍 Why this works. Bottom-up and top-down are not competitors; they are a check on each other, and the value is entirely in the disagreement. If bottom-up comes in far below top-down, either your register is thin or your percentage is a habit nobody has questioned in ten years. If bottom-up comes in above top-down, you have found a job whose risk profile does not match your standard percentage — a job you should either price differently or not bid. A percentage can never tell you that, because it carries no information about the specific job. It only carries your company's history, averaged.

Who owns contingency in a GMP, and how it gets spent

⚖️ What the contract says. In a CM at Risk contract with a GMP, the construction contingency is a line inside the guaranteed maximum price, and three consequences follow. It is the contractor's to manage subject to the contract's rules — Kestrel decides which row to draw against, and on Northgate any single draw above $50,000 needs Nadia Haddad's signature. It is the owner's money if it is not spent — because the GMP is a maximum, unused contingency does not become contractor profit; on Northgate it flows into the savings split, 75% owner and 25% Kestrel. And it cannot be used for scope changes — contingency covers risk within the agreed scope, while an owner's change is a change order to the GMP, a different document with a different approval path and often a different pot of money. Confusing the two is how a contractor ends up funding an owner's change out of its own reserve and discovering it in month nine (Chapter 31).

Most negotiated GMP contracts require monthly contingency reporting. Take it seriously even where it is not required, because the reporting is what makes the reserve manageable. Kestrel's contingency-use log has six columns: date, register row ID, description, amount, approver, remaining balance. Nothing leaves the account without a row ID. Ray's rule, learned the expensive way: no row, no draw.

The drawdown curve

Contingency is not spent evenly, and it is not supposed to be. But there is one guardrail everybody can understand: you should never have drawn a larger share of your contingency than the share of the work you have built. At 40% complete you should have spent no more than 40% of the reserve. Cross that line and the arithmetic is against you for the rest of the job.

Here is Northgate's actual drawdown against that guardrail. Bar scale: $60,000 per block.

The burn ratio is the number to watch:

Burn ratio = (percent of contingency drawn) ÷ (percent of work complete)

At 40% complete: $482,000 ÷ $1,320,000 = 36.5% drawn, ÷ 40% complete = 0.91.

Read the Northgate table again. The ratio sat at 0.65 through 30% complete and jumped to 0.91 in a single reporting period. That jump was the steel acceleration — $168,000 out of contingency in one month — and it was the moment the conversation about the job changed. It told us nothing Margo Deacon did not already know from walking the deck. What it did was make the problem legible to people who were not on the site, four weeks earlier than a cost report would have. That is the entire value of the curve.

Note also what happens after 60%: the ratio falls, not because we stopped spending but because the work kept building while the big risks retired. A falling burn ratio in the back half of a job is the signature of a project that got its bad news early. A rising one in the back half is the signature of a project that has not gotten its bad news yet.

The savings split, and the incentive it creates

Northgate finished with $438,000 of unused contingency, and contingency was not the only source of savings:

💰 Money check. Kestrel's fee on Northgate was $1,804,800. The savings share of $242,500 is a 13.4% increase on that fee ($242,500 ÷ $1,804,800 = 0.134) — earned without a single additional square foot of work. That is why a savings split is a genuinely powerful incentive, and why you must understand what it incentivizes.

It produces real buyout effort, real risk mitigation, real value engineering, and a contractor who treats the owner's money like its own because a quarter of it is its own. It can also produce three bad behaviors you should name out loud on your own jobs: starving legitimate risk response (a superintendent who needs $40,000 of temporary heat to protect a February slab pour is asking for money the PM is trying to preserve — say no for savings-split reasons and the split has just bought a $200,000 callback); hoarding to the last month, refusing to release money for real problems early and discovering at 85% complete that they compounded; and under-resourcing quality and safety, the two places where money saved today reappears tomorrow with interest and where the damage is not always measured in dollars.

The counterweight is structural, not motivational: separate the person who manages the reserve from the person who benefits from it, and make the register the arbiter. If a draw retires a named row it is legitimate and the answer is yes; if it does not, the answer is no. Neither answer is about anybody's bonus.

Contrast Curtis Boone's job. Rivermont Elementary School #12 is a hard-bid lump sum, so there is no split: 100% of savings is Kestrel's and 100% of overrun is Kestrel's. A cleaner incentive and a much sharper knife — and one where nobody but Curtis ever sees the contingency, which is exactly the situation in which a contingency stops having rows behind it. Case Study 2 follows what happened.

🔄 Check your understanding. At 55% complete, a job carrying an original $900,000 contingency has $310,000 left. Is it in trouble? Show the arithmetic and say what you would do next.

6.7 Schedule Risk and Cost Risk Are the Same Risk Seen Twice

Twelve of the fifteen rows on the Northgate register carry both a dollar figure and a day figure. That is not a formatting habit. It is the second theme of this book showing up as arithmetic: the schedule and the budget are the same conversation. On Northgate the exchange rate is fixed and everybody knows it:

Every schedule risk is therefore a cost risk with a conversion factor. When R-04 says "18 CD," it is also saying $191,700 — before you count the acceleration you would spend trying to avoid it.

A warning about double counting. If a row's cost impact already includes extended general conditions, do not add the daily rate again on top of it. I have reviewed registers where the same 14 days got counted three times: once in the row's cost impact, once as a separate "extended GC" row, and once in a schedule-contingency calculation. The total looked defensible and was wrong by a quarter of a million dollars. Double-counting is as damaging as omission, because it produces a number nobody believes — and once nobody believes the number, nobody funds any of it.

Where schedule contingency belongs

Cost contingency lives in one place, at the project level. Schedule contingency should live the same way, and almost never does. The natural instinct is to pad each activity: the erector says 72 work days, you carry 80 "to be safe," and you do that forty times. Three things then happen, reliably. The buffer gets consumed silently — a crew given 80 days for 72 days of work takes 80 days, because nobody reports being early when being early looks like having sandbagged. You cannot see it — a schedule with hidden padding in forty places has no honest critical path, so you have destroyed the diagnostic value of the schedule to buy protection you will not get. And you still finish late, because when a real problem hits one activity, the buffer that would have covered it is sitting inside thirty-nine other activities doing nobody any good.

The alternative is a single, visible, project-level buffer that belongs to the project and is drawn down like money. On Northgate, Wei Chen's baseline CPM carries an internal substantial-completion target of August 24, Year 2 against a contract date of September 18, Year 2 — 25 calendar days of project buffer, in one place, reported monthly, with a name on it.

💰 Money check. Those 25 calendar days are worth 25 × $10,650 = $266,250 of protected exposure — a schedule reserve with a dollar value, reported alongside the $1,320,000 cash reserve and drawn down the same way. When the steel delay consumed 23 of the 25 days, everybody could see exactly how much protection was left. Two days. That visibility is why the acceleration decision got made in a meeting instead of discovered in a cost report. The alternative — 25 days scattered across forty activities, invisible and already spent — leaves nobody able to say how exposed the job actually is.

If this argument feels like it wants a whole production philosophy behind it, it does. Chapter 27 picks it up and builds the Last Planner System on top of it; Chapter 14 shows you how to calculate the float that buffer protects.

🔄 Check your understanding. Two schedulers each add 25 calendar days of protection to the same 565-day project. One puts five days into each of five late activities; the other puts all 25 days in one block before substantial completion. Six months in, a single activity runs eleven days long. Describe what happens in each case.

6.8 Risk Across the Life of a Project — and Across the Life of a Company

The curve everybody draws and nobody acts on

Two things move in opposite directions over the life of a project: your ability to influence the outcome, which starts near total and decays to nothing, and the cost of making a change, which starts near nothing and grows without limit.

📊 Diagram (described). Picture two lines crossing. The horizontal axis runs through the project's phases: concept, schematic design, design development, construction documents, buyout and GMP, construction, closeout. The influence line starts at the top left and falls steeply, flattening near zero once steel is in the air. The cost-of-change line starts flat along the bottom and curves upward, steepening through construction and going nearly vertical at closeout. They cross around the end of design development. Everything left of the crossing is cheap to fix and hard to see; everything right of it is easy to see and expensive to fix.

This is theme three of this book — the project is built twice, once on paper and once in the field, and the first build determines the second — expressed as a risk statement. Take one row and walk it across the curve. R-03, the imaging equipment package:

Same decision, same building, a range of roughly zero to $186,400 depending only on when.

The practical instruction is uncomfortable: your best risk management happens at the moment you know the least. You must make the biggest decisions on the thinnest information, because that is where the leverage is. This is why preconstruction is not a formality and not a sales exercise (Chapter 11). It is where the job is won or lost, and the register is one of the very few tools that lets you act well while you still know almost nothing.

Enterprise risk: a single bad job can end a company

Everything so far has been about one project. Now step up to the level where Nadia Haddad and Owen Baptiste live, because a project manager who does not understand this is dangerous to the company that employs them.

Kestrel does roughly $410 million a year with about 340 salaried employees. General contracting is high-volume and thin-margin; net margins in the low single digits are ordinary. Take 2.5% as an illustration:

One job, out of the twenty-odd Kestrel runs, takes 39% of the year's profit. That arithmetic is what makes construction executives conservative, and it is why the answer to "should we bid this?" is so often no.

Three enterprise exposures compound it. Bonding capacity: Kestrel's surety program is $150 million aggregate and $60 million single project — not a credit line but an annual judgment based on audited financials, working capital, and backlog quality. A large loss consumes working capital, a surety that sees working capital fall reduces capacity, and the next job you wanted is one you are no longer allowed to bid. A bad job costs you its own loss plus the jobs you cannot chase for a year (Chapter 34). Concentration: Northgate is roughly 12% of revenue in one project, one owner, one sector, one geography; add two more healthcare jobs for the same system and you have a problem in all four dimensions at once, because when healthcare capital spending pauses — and it does — three jobs go quiet together while the salaried staff still get paid. Go/no-go discipline: the cheapest risk response available to any contractor is the one people find hardest to use — do not bid the job. Every risk in this chapter can be avoided completely at zero cost, right up until you sign. Kestrel's go/no-go review is a scheduled meeting with a written scorecard, not a hallway conversation, precisely because the decision has to survive the enthusiasm of the people who want the work (Chapter 15).

🔄 Check your understanding. Your register shows a well-managed set of rows totaling $600,000 of expected value against a $900,000 contingency. Company net profit is $6,000,000; the job is $52,000,000, about 14% of annual revenue. Name a risk that is invisible on the project register but real at the company level.

Transfer versus retention: when to buy and when to eat it

Chapter 5 covered the mechanics of bonds and insurance. This is about the decision. Three tests, in order: can the loss exceed what you can absorb? — if a single event could take more than your annual profit or your working capital, transfer it regardless of the arithmetic, because solvency is not a line on a spreadsheet. Is the premium less than the expected loss plus what you would pay for certainty? — insurers profit, so premiums usually exceed pure expected loss, and the gap is what you pay for variance reduction. Does the transferee actually control the risk? — transfer to a party who controls it is efficient and cheap; transfer to one who does not is expensive theater.

Kestrel's builder's risk options on Northgate, priced for the full 565-day term:

Going from A to B saves $196,000 − $154,000 = $42,000 in premium. What does it cost in retained loss? Kestrel's claims history on comparable jobs suggests about 1.4 covered events over a job this size and duration:

| Loss band | Expected events | Average loss in band | Retained under A ($25k) | Retained under B ($100k) | |---|---:|---:|---:|---:| | Under $25,000 | 0.8 | $11,000 | 0.8 × $11,000 = $8,800 | 0.8 × $11,000 = $8,800 | | $25,000–$100,000 | 0.4 | $58,000 | 0.4 × $25,000 = $10,000 | 0.4 × $58,000 = $23,200 | | Over $100,000 | 0.2 | $340,000 | 0.2 × $25,000 = $5,000 | 0.2 × $100,000 = $20,000 | | Total expected retained loss | 1.4 | | $23,800 | $52,000 |

Additional expected retained loss under B: $52,000 − $23,800 = $28,200. Premium saved: $42,000. Net expected advantage of Option B: $13,800.

The arithmetic says take the higher deductible — but $13,800 on a $47,500,000 job is 0.03%, which is noise, and nobody sensible decides on noise. The real decision is about variance and about who is deciding. At the project level, Option A looks right: one bad night — a sprinkler head sheared during MEP rough-in, flooding three floors of new gypsum board — costs the job $100,000 out of contingency under B instead of $25,000 under A, and that single event eats 7.6% of the entire reserve. At the company level, Option B is clearly right: across many jobs a year the retained losses average out, the $42,000 saved on each job is real repeatable money, and the company has a balance sheet to absorb a $100,000 hit while a single project has no balance sheet at all.

Ray's rule: take the higher deductible when the retained amount is small relative to the reserve that absorbs it, and when you run enough jobs for the averages to work. Which is another way of saying this is a company decision, not a project decision — the same lesson as the enterprise section arriving from a different direction. One caution that costs contractors real money: a deductible is not the only thing you are buying. Coverage terms, exclusions, sublimits for water damage and testing, whether the policy covers soft costs and delay, whether it responds during commissioning, and who is a named insured all matter more than the deductible in the events that actually hurt.

6.9 The Human Side, the Safety Exception, and the Line You Do Not Cross

Three cognitive habits and one social pressure

Everything in this chapter assumes people will honestly estimate probabilities and impacts. They will not, and the ways they fail are consistent enough to plan around.

Optimism bias. People systematically believe bad outcomes are less likely for them than for others in the same situation. Ask a room of project managers whether their current job will finish on time and nearly all say yes. Ask what fraction of jobs like theirs finish on time and the same room gives a much lower number — same people, same population, twenty seconds apart.

Anchoring. The first number spoken dominates every number that follows. If Nadia says "I think this is a two-percent contingency job" before Tomás opens the register, every argument afterward becomes a negotiation around two percent — including the correct ones. The anchor does not have to be credible. It only has to be first.

The planning fallacy. People forecast from the inside view — this plan, this crew, this sequence, all going as intended — instead of the outside view, which asks what actually happened on the last fifteen jobs like this one. The inside view is always more optimistic, because a plan is a description of things going right. Nobody plans the eleven days a submittal will sit on a desk.

And the social pressure. The person who raises a risk in a meeting is, in that moment, the person slowing everybody down — the pessimist, the one who is "not a team player," the one making the boss's Friday harder. Meanwhile the person who says "we'll figure it out" is agreeable, confident, and wrong, and nobody will remember eleven months later that they were the one who said it. The incentives in a live meeting are aligned almost perfectly against telling the truth about risk.

🔍 Why this works. A written register with named owners defeats all four, and the mechanism in each case is structural rather than motivational — which is why it works on people who have never heard of any of these biases.

Against optimism bias, it forces disaggregation. Optimism operates on aggregates: "it'll be fine" is a statement about a whole project. It is far harder to be casually optimistic about fifteen specific named events you must put a number on one at a time. Breaking a job into rows does not make anybody less optimistic — it removes the vagueness optimism needs in order to live.

Against anchoring, it reverses the order of operations. Build the register before anyone states a total and the total becomes a result instead of an anchor. Tomás's forty-one rows existed before the words "three percent" were said in any room. That sequencing is the whole trick, and it costs nothing but discipline.

Against the planning fallacy, it demands the outside view. A register column asking "have we seen this before, and what did it cost?" cannot be answered from the plan. It can only be answered from history — which is why closeout reports matter more than checklists.

Against the social pressure, it changes what the room rewards. When every row carries a name, raising a risk stops being a personality trait and becomes a job duty. Nobody is "the pessimist" — they are the person whose name is on row 7, doing what row 7 requires. The accountability runs both ways: the question at the end of a bad job is not "who was negative?" but "whose row hit unmanaged?" That is a question with an answer, and answerable questions change behavior in a way culture memos do not.

🪞 Learning check-in. Put the book down for two minutes and answer these honestly, in writing if you can.

• Of the five responses in §6.2 — avoid, transfer, mitigate, accept, exploit — which is your personal default? Most people have one. Engineers over-mitigate. Lawyers over-transfer. Optimists over-accept without funding. Knowing your default is worth more than knowing all five, because your default is what you reach for under pressure at 4:45 on a Friday.
• Think of a time you saw a problem coming and did not say it out loud. What did the room's incentives look like in that moment? What would have had to be different — not about you, about the structure of the meeting — for you to have said it?
• Look at the Owner column in the Northgate register. How do you feel about your own name sitting in a column like that, next to an event that has not happened yet? If the honest answer is "uncomfortable," that is the correct answer, and it is the reason the column works.

You are six chapters in. If the last five have felt like a tour of documents — delivery methods, contract types, legal instruments — this is where they turn into a single method. Every one of those documents exists to answer the question you have now learned to ask of every row: who owns this, and what does it cost?

The row you cannot transfer

⚠️ Safety alert. Row R-15 is the safety row: 15% probability, $150,000 of impact, five calendar days, an expected value of $22,500 — the fourth-smallest number in the column. On the 5×5 matrix it scores 2 × 3 = 6 and lands in a box marked "Medium," alongside a commissioning delay and an asbestos survey. That is the matrix failing, not the risk being medium.

Two things are true about safety risk that are true of no other row.

First, it is the one risk you cannot transfer. You can transfer subsurface conditions to the owner, escalation to an allowance, subcontractor default to a surety, and water damage to an insurer. You cannot transfer an injury. Insurance pays medical costs and workers' compensation pays indemnity, and both are real and necessary — and neither is a transfer of the risk. The bond does not bring anyone back. Nothing you can buy makes an injury not have happened. The only legitimate responses to safety risk are avoid and mitigate. Accept is not on the menu, and neither is transfer.

Second, safety risk is generated by the other rows. In week 34 of Northgate, on a Tuesday morning on the north elevation at level 3, a mason tender named Emiliano "Milo" Serrano stepped onto a scaffold plank that had been lifted overnight by a different trade running conduit and not re-secured. The plank shifted. He went down onto the platform and caught himself on the top rail. No injury. Bea Salgado stopped work on the elevation for the day.

The investigation found three failures, not one: no competent-person inspection tag for that shift; a scaffold modified by a trade that had not erected it, with no re-inspection; and a crew running behind after the steel acceleration under an unwritten "make it up" pressure.

The third finding is the one nobody wanted to write down, and it is the one this chapter is about. The schedule risk on row R-04 and the acceleration decision it triggered became a safety hazard on the north elevation eleven weeks later. Nobody decided to trade safety for schedule; nobody would have. The pressure moved through the organization on its own — from a submittal log to a mill slot to an acceleration budget to a crew that felt behind. If your register treats safety as one row among fifteen rather than as an output of the other fourteen, you will miss it every time, because it does not arrive labeled.

Two consequences for how you build a register. Put a safety column on it. And ask of every schedule and cost response: what does this do to the people doing the work? An acceleration row reading "$168,000, recovers 17 days" is incomplete. The honest version reads "$168,000, recovers 17 days, adds a second crew to a congested elevation and increases exposure — mitigation: additional competent-person coverage, revised scaffold inspection frequency, a stand-down before the sequence change." Chapter 24 builds that discipline out fully; the point here is that it starts on the risk register, not on the safety plan.

Two ethical lines, named

Line one: contingency that is really hidden profit. You have just spent a chapter learning that contingency is legitimate, necessary, and routinely too small. That same knowledge makes it easy to bury margin in a reserve nobody examines. The line is drawn by contract type and by disclosure. In a lump-sum contract, your contingency is your money and your risk: you bid a number, you perform for it or you do not, what is inside is nobody's business, and if you carry 5% and never spend it you earned it by taking the risk. In an open-book GMP or cost-plus contract, contingency is a disclosed line item in a document the owner relies on, subject to a savings split that returns unused money to them. Carrying money there that you have no intention of spending, that traces to no risk, and that you expect to convert into your 25% share is not conservatism. It is a misrepresentation in a document the owner is using to set their budget and their financing.

The test works both directions: could you show the owner the rows? Not would you — the answer to that is often no, for legitimate competitive reasons. Could you. If every dollar traces to a named event with a probability and an impact you actually believe, you are on the right side of the line even if you never show anybody. If it traces to nothing, you are on the wrong side even if you are never asked.

Line two: risk transfer to a party who cannot price or absorb it. Go back to §6.2's third response row and make it concrete. You hand a $180,000 drywall subcontractor with a $2 million bonding line a subcontract flowing down a no-damage-for-delay clause, an unlimited indemnity, a broad "subcontractor accepts all site conditions" provision, and liability for consequential damages including your liquidated damages. Four moves, all common, all enforceable in many jurisdictions, all costing you nothing to insert.

Here is what happens next, in order, and it is the same every time. The sub does not price it — they do not read it, or they read it and know they cannot win the job if they price it, so the number that arrives carries no risk load. The risk occurs — a delay, a condition, an impact they could never have controlled. The sub cannot absorb it — a $180,000 subcontractor cannot fund a $400,000 consequence, and there is no argument to have, because the money does not exist. The sub defaults, walks, or stops paying his own suppliers — and now you own the original risk plus a replacement subcontractor at a premium, plus lien claims from his second tier, plus twenty-one days you did not have.

Risk transferred to a party who cannot absorb it is not transferred. It is hidden — from them, and from you. That is why this is a competence question before it is an ethics question, and why the two answers coincide. The line: flow down risks the subcontractor can see, price, and survive. Beyond that you are not managing risk; you are arranging for it to arrive later, wearing a different name, with a lien attached.

The version that actually works is unglamorous. Write scope sheets that say clearly what the sub owns. Price the risks you want them to carry as visible, biddable line items. Bond the packages where a default would hurt. And when a sub tells you they cannot carry something, believe them and put it back where it belongs — because you are going to own it either way, and owning it on purpose costs less than owning it by surprise.

6.10 📋 Try It: Six Risks on Willow Street

You are the project manager for the Willow Street Community Center: $6,800,000, 24,000 SF, two stories, wood-framed second floor over a structural steel and CMU first floor. Design-bid-build, lump sum, 425 calendar days, liquidated damages of $1,200/CD, City of Rivermont Parks & Recreation as owner, prevailing wage. Your cost of work is $5,850,000 and your company's standard contingency is 4% of cost of work.

Your tasks. (a) Compute the expected monetary value of each risk. (b) Total them. (c) Compare the total against a top-down 4% contingency and state which number you would carry and why. (d) Choose exactly two risks to transfer, say precisely how, estimate what each transfer costs, and say what it does to your numbers — then name one risk that looks transferable and is not, and explain why.

Work it before you open the answer.

Spaced Review

Answer these from memory before reading the responses. The recall is the point.

1. From Chapter 4 — who owns quantity risk, productivity risk, and escalation risk under each pricing structure? Say it out loud before you look.

Notice that this table is a risk allocation table. Chapter 4 taught it as contract types; this chapter shows you it was always the same subject. Choosing a contract type is choosing a risk allocation, and the price follows from it.

2. From Chapter 5 — indemnity and insurance are both risk transfer. What is the difference, and why does it matter on a register?

Indemnity moves the obligation between two parties to the contract, and your recovery depends entirely on the other party's ability to pay. Insurance moves the money to a third party with a balance sheet, subject to the policy's terms and limits. That difference is exactly why R-11 is answered with a bond rather than a promise: an indemnity from a subcontractor who is insolvent when the risk occurs is a piece of paper. Whenever a register row's response is a contractual promise, ask the follow-up — promised by whom, and can they pay?

3. The deep callback, from Chapter 3 — risk allocation drives price. Chapter 3 argued that delivery method and contract type are one decision and that the decision is the price. This chapter showed the mechanism twice. Delete a differing-site-conditions clause and no invoice appears — but every bidder loads the earthwork package and the owner pays whether or not the bad soil exists (§6.3). Transfer truss delivery to a supplier who controls the mill and the transfer costs less than the expected value, because a risk is cheap to the party who can control it and expensive to everyone else (§6.10). Same principle, two directions. Price is what risk allocation looks like on an invoice.

Project Checkpoint: The Willow Street Risk Register

In Chapter 5 you built the legal-framework checklist for Willow Street — bonds, insurance certificates, lien-notice deadlines, and the authority-having-jurisdiction list. That checklist told you which instruments exist. This one tells you which risks they are for.

Your deliverable: a project risk register for the Willow Street Community Center, fifteen rows, plus a written contingency reconciliation.

Part 1 — the register. Fifteen rows minimum, using the two-table format from §6.4 if it helps readability. Every row needs all eleven fields:

Build it from the package in Appendix K. At least three rows should come from actually reading the drawings and specifications rather than from a generic checklist — the 8-inch water main, the wood-over-steel-and-CMU transition, the commercial kitchen, the gym floor, and the prevailing-wage requirement are all sitting there waiting to be found. Include at least one upside row and exactly one safety row, and write next to the safety row why its only permitted responses are avoid and mitigate.

Part 2 — the reconciliation, in writing. Size the contingency two ways. Bottom-up: total the expected values of the rows you own, separating out any rows the City owns and any a subcontractor or supplier owns after transfer — do not fund somebody else's risk. Top-down: apply a percentage to your $5,850,000 cost of work; use 4% unless you can justify something else.

Then write 200 to 300 words explaining the difference. A good reconciliation names at least three things: that an expected value is a mean and you fund to a confidence level; that your register is incomplete and roughly how incomplete you believe it to be; and that mitigation costs belong in the estimate rather than the reserve. Finish with the number you would actually carry and the sentence you would say to your operations executive to defend it.

Date both parts and keep them. You will update this register at buyout in Chapter 16, draw against it in the cost reports in Chapter 28, and grade yourself against it at closeout in Chapter 40 — which risks materialized, which never did, and which were never on the list.

Next: Chapter 7 puts you in front of the actual drawings and specifications with the order-of-precedence rules and asks you to find ten discrepancies. Bring this register. Most of what you find belongs on it.

Chapter Summary

A reference framework, not a recap. Come back here when you are staring at a register.

The five-question test for any risk row. If you cannot answer all five, the row is not finished.

1. What is the event? A thing that happens at a moment, not a condition of the world.
2. What does it cost, and how many days? Both numbers, always. They are the same risk seen twice.
3. Who owns it — by name? A person, not a department. "Us" without a name means nobody.
4. What is the response? Avoid, transfer, mitigate, accept, or exploit — pick one as primary.
5. What is the trigger? The observable thing that says it is happening now.

The four numbers to carry in your head.

The five money buckets, one line each. Estimating contingency covers unfinished design and shrinks as drawings complete. Construction contingency covers identified execution risk within scope, and every dollar traces to a row. Owner's contingency covers scope the owner adds and is not yours. Escalation allowance covers named commodity movement and nothing else. Management reserve is the company's, not the job's.

The threshold, restated. Contingency is a priced, owned, drawn-down reserve for identified risk. A number without rows behind it is either fat or a lie, and both get spent by month nine. The test: could you show the owner the rows?

Six failure modes and their fixes.

The two lines. Contingency that traces to no risk in an open-book contract is a misrepresentation, not conservatism. Risk flowed down to a subcontractor who cannot price or survive it is not transferred — it is hidden, and it will come back with a lien attached.

And the one with no arithmetic. Safety risk cannot be transferred, and it is generated by the other rows on your register. An acceleration decision made in a conference room becomes a hazard on a scaffold eleven weeks later. Write it down that way.

What's Next

You now know how to name a risk, price it, assign it, and fund it. What you do not yet know is where most of the rows come from — and the answer is that they are sitting in the documents, in the gaps between the drawings and the specifications, waiting for somebody to read carefully enough to find them. Chapter 7 teaches you to read a set of construction documents the way an estimator does: the CSI MasterFormat structure, what a specification governs versus what a drawing governs, and the order of precedence that decides who pays when the two disagree. Bring the register. You are going to add to it.