Case Study 35-1 — The Northgate VDC Ledger: Two Findings, Two Misses, and What It Actually Bought

All people, companies, and projects in this case are Tier-3 illustrative composites. Every dollar figure is illustrative arithmetic from one job, not an industry statistic.

Setup

Project: Northgate Outpatient Pavilion — 132,000 GSF, four stories, Meridian Health System. CM at Risk, $47,500,000 GMP, 565 calendar days, substantial completion September 18, Year 2. Extended general conditions $5,150/CD, liquidated damages $5,500/CD, total daily exposure $10,650/CD.

The people.

The occasion. In April of Year 2, Owen Baptiste is building next year's corporate VDC budget and asks me one question: what did Northgate's program cost, and what did it return — with the misses left in. This case is that answer, and the two findings behind it.

What Happens

Finding one — the level-two corridor, December of Year 1

You know the geometry from Chapter 10. Level 2, Corridor C, grids 4 through 9: 30 inches of available cavity against 39 inches of systems, under a transfer girder that cannot move. What matters here is not the nine inches. It is when we learned about them.

Four options were priced, twice each — dollars and days. Three died on arithmetic, and one of those deaths is the point. Option 3 was to reshape the supply trunk to 60" × 11" and upsize the fan: the answer most engineers reach for, $96,000 direct plus 21 calendar days at $10,650 — $319,650 — and it does not work. The new stack is 34 inches against 30 available and the new plan width is 103 inches in a 96-inch corridor. Nobody had to buy that option to find out it failed. A model ran both checks in under an hour.

The reroute that was selected cost $46,100 and zero days.

Now price the counterfactual — discovered at rough-in in February of Year 2, sheet metal fabricated, hangers set, sprinkler main permitted and hung:

Twelve days, not twenty-one, because the field version is a smaller reroute performed badly rather than a fan change. That is worth saying out loud: the field version of a conflict is usually a worse solution, not the same solution later.

One honest qualifier, and Grace insists on it. The model did not solve Corridor C. It found the problem in December, proved which options were impossible before anyone paid to explore them, and put the decision in front of four companies on a date they could still act on. Sofia Marchetti, Tavares Hill, Devlin Achebe, and Dale Whitcomb solved it, in a room, arguing. The instrument does not make the decision. It buys you the calendar in which a decision is still cheap.

Finding two — the crane, last week of June of Year 1

The 4D run is in the chapter hook: the 275-ton crawler finishes the building and cannot leave it, because twenty-two panels set in week eight reduce the mat-road opening from 22'-0" to 11'-6" and the machine is 14'-8" wide over extended tracks.

What is not in the hook is the meeting the next morning, which is where the money was actually made or lost.

Margo Deacon: "Walk it out the west bay."

Wei Chen: "Priced it. Three hundred eighty feet of new mat road across the detention basin, which is already excavated, plus a temporary crossing over the sanitary lateral. $61,400 and three days. And I do not love putting that machine over a lateral we just bought."

Margo: "Then take it apart where it stands."

Wei: "Also priced. Disassemble in place, assist crane on the east pad, remove by components. $79,200 and six calendar days, and Hank loses his last two picks while the assist crane is set up."

Hank Duffy, on speaker: "I would rather you moved the panels."

Which is what happened. Three options, priced the same way as the corridor — dollars and days:

Option C's build-up is the one in §35.5: double handling on fourteen panels, a temporary hardstand and storage frames, a second precast mobilization, revised erection drawings, and ninety hours of Wei's and Grace's time re-logicking the schedule and re-running the 4D.

The same problem in week nine of erection — the last week of October — was priced at $178,000 direct plus 11 CD on the critical path: $295,150. Avoided: $261,250.

💰 The line worth stealing. Options A and B are not wasted work. They are the reason $33,900 is a decision and not an assumption. If you take one habit out of this case, take that one: price the options you intend to reject.

The ledger, April of Year 2

$861,750 ÷ $643,500 = 1.34. Net gain $218,250.

Owen does not accept a ratio. He takes the number apart. "Your RFI line compares this job to jobs that are not this job." Correct — strip it. "Your prefab line is a buyout differential, not a saving." Partly correct — strip it too. "And two lines are half your return. What if neither had happened?"

So we ran it hostile:

On the most hostile honest reading this program returned sixty-five cents on the dollar. On the reported reading it returned $1.34. Both of those are true, and the range is the finding. A coordination program is not a profit center; it is a priced instrument that narrows the tail of your rework distribution. You buy it the way you buy builder's risk insurance — knowing that in the good year you paid for nothing.

What the model did not catch

Two conflicts. $49,600 between them. Both are in the ledger as a deduction, and both are here because a case study that only lists wins is a brochure.

Miss one — AHU-5's maintenance path. $31,200. Cause: level of development.

The LOD matrix requires owner equipment at 400, with vendor data, for coordination. AHU-5 — the level-four mechanical room unit — was an approved substitution made in October, and the vendor's dimensional package arrived nine weeks after the model needed it. The generic LOD 300 placeholder stayed: right box, right location, right connection points, and no coil-pull envelope, no filter-access zone, no drain-pan removal path.

Coordination ran clean, and could not have run any other way. A soft-clash rule tests for intrusion into a clearance zone somebody modeled, and nobody had modeled this one — the element carrying it was two levels below its matrix cell. Cardinal's service group found it at startup in August of Year 2, when they tried to pull a coil.

The rule was not broken by a modeler. It was broken by a schedule. Nobody decided to rely on an element below its required LOD; a placeholder survived a sign-off because its replacement was late and nobody ran a report that would have said so.

Grace's fix costs almost nothing and is now standard on every Kestrel job: before any area is signed off, run an LOD compliance report against the matrix — every element below its matrix cell, listed by system and trade, with a name and a date next to it. An element below its required LOD blocks sign-off for that area. Twelve minutes a level.

Miss two — the level-three multi-trade rack. $18,400 and four days. Cause: the building moved between scans.

Northgate ran six progressive verification scans. Between scan three and scan four, the level-three slab deflected under the wet lightweight topping and finished about 1¼ inches high over a forty-foot stretch at grid lines 6 to 9. The corridor rack for that run had already been released for fabrication against scan three.

The rack was built correctly, to a model that was correct, for a building that had moved.

And then the part that should worry you more than the $18,400.

When Grace investigated why nobody in the field flagged the divergence before three fabricated sections arrived on a truck, the answer was not about deflection at all. Beacon Communications — low voltage, cable tray, the smallest detailing allowance on the job at $19,000 — had detailed level three the way they have always detailed: on paper. The model Beacon submitted for federation was produced afterward, by a subcontracted service, from those paper sheets, to satisfy the requirement.

So Beacon's foreman had never opened the model. Beacon's tray was on that rack, and the first person to notice it would not land was a Cardinal fitter on a lift.

Grace's audit of Beacon's remaining areas found eleven more locations where Beacon's model and Beacon's paper did not agree. None had cost a dollar yet. That is the danger of failure mode #5 in §35.10: a reverse-engineered model looks exactly like compliance right up to the moment it doesn't. It costs nothing, repeatedly, and then it costs everything once.

Kestrel's response was contractual, not technical, and it is the correct lever: Beacon's detailing allowance on the next job was doubled, and fabrication release for their scope was tied to coordination sign-off, so detailing on paper stopped carrying a schedule advantage.

Analysis

1. The return is a distribution, not an average. One finding — an afternoon in June — is 30 percent of the gross savings and 41 percent of the whole program cost. Evaluate a coordination program on expected value and you will underbuy it, because the value lives in the tail. That is the logic behind risk R-07 in Chapter 6: a 50 percent chance of $230,000 is a $115,000 expected value, and the expected value is what buys the budget — but the reason you buy it is the $230,000.

2. Both misses have the same signature. Neither was a modeling error. In both, something outside the model changed and the model did not know: late vendor data in one, a deflecting slab in the other, and a trade working from paper underneath both. The chapter's eighth rule — coordinate the model, verify the building — is right and incomplete. There is a third verb: verify the process. Ask which trade is not actually modeling, and how you would know.

3. The ledger is credible only because it was built as we went. Every line was identified and priced on the day it was found. An ROI case reconstructed at closeout is an argument, not an accounting — so ask which lines were priced contemporaneously and which were remembered.

4. What would have to be true for this program not to pay? A less congested building, fewer parties, a generous plenum, a repetitive floor plate, and no constrained site — which is to say, a different job. That is §35.10's small-project note, and it is not a hedge. It is the management skill.

Discussion Questions

1. Ray claims "the field version of a conflict is usually a worse solution, not the same solution later." Test that against Corridor C: name three specific ways the February field resolution would have differed from the December reroute, and which one the owner would eventually have noticed.
2. Owen's hostile test bottoms out at 0.65. Suppose that were the real number. Construct the strongest case for buying the program anyway, then the strongest case against — and say what evidence would settle it.
3. The AHU-5 miss happened because a late vendor package left a placeholder in the model through sign-off. Grace's fix is an LOD compliance report. Name one other control that would have caught it, and say which you would rather rely on.
4. Beacon's reverse-engineered model cost Kestrel nothing it can prove. Write the argument you would make to a project executive for doubling Beacon's next detailing allowance when the evidence of harm is eleven discrepancies that never hurt anyone.
5. Options A and B for the crane were priced and rejected. What does pricing options you intend to reject cost, and when is that cost not worth paying?

Your Turn

Rebuild the ledger for a project half Northgate's size and half its congestion — a $24,000,000, 70,000 SF, three-story medical office building with no imaging suite, no surgery, no medical gas beyond four exam-room outlets, a 42-inch typical ceiling cavity, and an open site.

Produce three things, on one page:

1. A scaled cost table. Do not simply halve it. Decide which lines scale with project size, which scale with the number of parties, and which have a floor that does not scale at all.
2. A returns table with the lines you can honestly still claim. The crane trap does not exist on an open site. What replaces it, if anything?
3. A one-sentence recommendation — full program, scoped subset, or decline — with the two model uses you would pursue first.

Then answer the only question that matters: where in that scaling exercise did the ratio cross 1.0, and what pushed it there?