Case Study 24-1 — Finding Three: Writing It Down, and Then Finding Out Whether It Worked

Kestrel Construction Group, the Northgate Outpatient Pavilion, Meridian Health System, and every person named here are Tier-3 illustrative composites assembled from real projects. The regulatory framework is real. The people are not.

Setup

You already know what happened on the north elevation. Week 34, a Tuesday, 7:20 in the morning. A frame scaffold on level 3 had been partially modified overnight by a Halcyon Electric crew running conduit; a plank was lifted out of its hooks and set back approximately in place. Emiliano "Milo" Serrano, a mason tender for Ashcroft Masonry, stepped onto it, felt it go, and caught himself on the top rail. Nobody went to the hospital. Bea Salgado stopped work on the entire elevation for the day.

You also know the three findings, because the chapter opens with them:

1. No competent-person inspection tag current for that shift. The tag was two days stale.
2. A scaffold modified by a trade that did not erect it, with no re-inspection and no re-tag.
3. A crew running behind after the steel acceleration, under an unwritten pressure to make it up.

Chapter 22 walked you through that morning as a temporary-structures problem — how the plank moved, who owned the scaffold, what a modification-control clause has to say. This case study is the other half: the investigation from the system side, going past the report to the two questions almost nobody asks. What did the corrective actions cost, and did they actually work?

The cast: Ray Alvarez, senior project manager, the narrator. Beatriz "Bea" Salgado, corporate safety director, eleven years an ironworker before thirteen doing this. Marguerite "Margo" Deacon, general superintendent, thirty-one years. Nadia Haddad, vice president of operations, on the phone from Rivermont. Dani Okonkwo, field engineer, taking minutes. And on Tuesday afternoon, on a conference line for forty minutes, Arlene Wexler, Kestrel's general counsel.

What Happens

Day 6 — Monday. The argument you have already read.

Bea's three lines on the whiteboard. My sentence — "Three isn't a finding, Bea. Three's a feeling." Margo's "She's right and you know she's right." Bea's "That's what happened."

I am not going to retell it. The Monday argument is the famous scene; the Tuesday working session is the one that mattered.

Day 7, morning — the honest version of the objection

Here is what I actually believed on Monday night, stated more precisely than I managed to state it in the room.

Findings one and two are falsifiable. Finding three is not. I can photograph a stale tag with the date in frame. I can produce the inspection register and show the gap. Finding two is the same: two electricians, interviewed separately, both describe lifting the plank. That is evidence. Finding three has no photograph. There is no email that says go faster and don't stop for a scaffold tag. There is a schedule update, a payroll record showing 21% overtime, a coordination meeting where somebody said "north elevation" in a particular tone of voice, and an inference.

So my objection was not cowardice, or not only cowardice. It was that I had unconsciously imported an evidentiary standard borrowed from liability into a process whose purpose is learning. A courtroom asks whether you can prove it. An investigation asks whether you have a hypothesis good enough to act on. Confusing the two is the most common way a competent investigation dies at the last step.

Bea's answer, on Tuesday morning, was the substitution test from §24.10:

Bea: "Take Milo out. Put any competent mason tender in the world on that ladder at 7:20 that morning. Does he read the tag?"

Ray: "No."

Bea: "Then the finding isn't about Milo, and it isn't about the electricians either, because run the same test on them. Any competent electrician, 9 p.m., conduit to finish, competent person went home at three, no number to call. Does he wait until morning?"

Ray: "No."

Bea: "Then you have three people behaving normally and a bad outcome, which means the thing that changed is not a person. And the only thing that changed on that elevation in the last eleven weeks is the sequence, and you and Nadia changed it."

That is not proof. It is a disciplined inference, and it is the standard an investigation is supposed to run on.

Day 7, afternoon — the discoverability question, answered by somebody qualified to answer it

My second objection was the real one, and every project manager in the country has it: if we write it down, it is in the file, and the file is discoverable.

We called Arlene Wexler. Her answer took four minutes and I have repeated it in every safety meeting I have run since. This is a management framework, not legal advice — privilege doctrine, discovery rules, and the treatment of post-incident investigations vary substantially by jurisdiction, and how you structure an investigation is a question for your own counsel before the incident, not after.

Arlene: "Start with what you think you're hiding. The acceleration is in the change directive. It's in three schedule updates, in the payroll, and in the coordination minutes for five straight weeks. If anybody ever litigates this, a competent expert finds it in ninety minutes without your report. You are not deciding whether it gets found. You are deciding what it gets found next to.

"Next to an investigation that doesn't mention it — what does an expert say about that on the stand? Or next to an investigation that identified the connection and seven corrective actions with owners, dates, and verified closure. One of those is a company that finds its own problems. The other is a company that either didn't look or looked and left it out. I would much rather defend the first one.

"And here's the part you're not thinking about. The General Duty Clause turns on recognized hazards. If this happens again in nine weeks — and Bea says it will — the question won't be whether you wrote it down. It'll be whether you knew. You knew on Monday. Write it, fix it, close it, and the knowing is an asset instead of a liability."

Day 7, late afternoon — Nadia sets the condition

Nadia Haddad's position was not the one I expected. She did not argue about the report at all. She argued about the price.

Nadia: "I'll sign the report. Here's my condition. If finding three is real, then it means an acceleration costs more than we priced it at, and I'm the person who prices accelerations. So I don't want a finding. I want a form, I want a threshold, and I want the controls in the number before I sign the directive. If this ends up being a paragraph of regret in a report nobody reads, I've spent political capital for nothing."

That is where the safety impact review came from. Not from a safety department. From an operations executive who understood that a finding without a mechanism attached is a feeling, which is what I had accused Bea of in the first place — and I was right about that, and wrong about which finding it applied to.

Day 7, evening — Margo's contribution, which was different from everybody's

Margo did not care about the report and said so.

Margo: "Write whatever you want. Here's what I want. Ashcroft's superintendent has been eleven days behind for three weeks and nobody at Kestrel has ever said one word to him about it in writing. Not one. He's just been sitting in a meeting every Thursday hearing 'north elevation' in a certain tone. So he found the hours the only place a subcontractor can find them, which is by taking them off the front of the shift.

"If we expect a sub to recover eleven days, that goes in a letter. With a revised sequence. And what it costs. Because then he can write back and say no I can't, and then it's a conversation instead of a pressure."

That became corrective action six. It is the cheapest item on the list, it costs literally nothing, and in my experience it is the one contractors are least willing to do — because putting a recovery expectation in writing invites a subcontractor to price it.

Day 8, Wednesday — Margo delivers the toolbox talk in §24.7, announcing the three field actions.

Day 10, Friday — the report issues with all seven corrective actions. I signed it.

The seven corrective actions, priced

Four of them have nothing to do with scaffolds. That is the point of the whole exercise.

💰 What it cost.

Now put that number somewhere useful. $88,800 is 8.3 calendar days of schedule exposure at $10,650/CD. It is 4.9% of Kestrel's $1,804,800 fee. It is 0.19% of the $47,500,000 GMP.

And here is the sentence I want you to be able to say in a budget meeting. It came out of general conditions, which is inside the GMP, and unused GMP contingency and savings split 75% owner / 25% Kestrel. So the entire seven-action program, funded for the rest of the job, reduced Kestrel's bottom line by roughly $22,200.

Twenty-two thousand dollars. I have watched executives spend longer than that arguing about a jobsite copier lease.

Did It Work? — measuring the corrective actions instead of admiring them

This is the part almost every organization skips. A corrective action gets marked "complete" when the document exists, which measures nothing. The question is whether the behavior changed and stayed changed, and answering it requires a measure defined at the time you write the action.

The verification measures

Look at the two bolded rows and hold them. We come back to them.

The trend data

Here is the leading-indicator dashboard past week 37, where §24.11's table stops.

Weeks 61–65 are the peak of the job — 210 craft workers in week 61, the highest headcount Northgate ever carried. Every pressure indicator went up and every reporting indicator went down, exactly as in weeks 31–34.

But look at the magnitude. In weeks 31→34, near-miss reports fell from 4 a week to 1 — a 75% collapse, and we did not notice it until a man's boot found a plank. In weeks 54→61, under higher headcount and comparable overtime, reports fell from 9 to 7 — a 22% sag, and Wei Chen flagged the schedule pressure index in the week-58 report, nine days before it peaked. We took two days out of the ceiling-grid sequence and rebalanced two crews instead of pushing.

Then in month 16 — weeks 66 through 71 — the finishes package slipped 11 days and we ran the acceleration you priced yourself in §24.14: second shift, twenty workers, six weeks, $186,000 priced plus $52,000 of controls funded as a condition of the recommendation. Restricted night scope, dedicated task lighting, a night superintendent, a walked handoff. It produced no recordable, two near-miss reports (both about the day-to-night handoff, both closed inside a week), and exactly one violation of the restricted scope — a hot-work request at 9:10 p.m. that the night superintendent refused. He called me at home to tell me he had refused it. That call is the whole program working.

🔍 What this evidence is, and what it is not. One project, small counts, two pressure periods compared under different headcounts twenty-seven weeks apart. That is not a controlled study and I will not pretend it is. What I will say is that it is the only kind of evidence available at the scale a project manager operates at, and it is enormously better than the alternative. The alternative is the recordable rate, and ours cannot answer this question at all. Northgate finished at 5 recordables over 412,000 MH, TRIR 2.43 — three before week 34, two after. Had the corrective actions done nothing, the expected count in the remaining 254,000 hours at the pre-incident rate would have been about five. We got two. Is that a result? No. At counts that small, two versus five is a coin-flipping exercise, and anyone who tells you otherwise is selling something. The leading indicators are the only place a real answer lives.

Analysis: what held, and what decayed

Fourteen months later I went back through all seven with Bea, honestly, and this is the part I would want if I were you.

Held: 1, 2, 3, and 5. Four of seven, and for the same reason in every case: each one changed the path of least resistance rather than asking somebody to remember something.

• The tag became a precondition of access — the ladder was barricaded. You did not have to remember; you had to move a barricade, and moving it without a tag was a visible act in front of your crew.
• The modification procedure held because it came with a phone number a human answered. Halcyon's crews called it thirty-one times. You did not fix their behavior; you gave their behavior somewhere to go.
• The dashboard held because Wei Chen automated the inputs. Overtime out of payroll; negative float and out-of-sequence starts straight out of the schedule update; stacking off the look-ahead. Ninety minutes a week, because it did not depend on anyone collecting anything.

Decayed: 6, and partially 4 and 7.

Corrective action 6 — written recovery expectations — decayed most and decayed first. Five situations arose where Kestrel in fact expected a subcontractor to recover days. We wrote three of those letters and did not write two, and I know exactly why: a written recovery expectation invites a priced response, and both times the sub would have come back with a number. Margo was right on day 7 and I was still finding reasons fourteen months later. The cheapest control on the list is the one that failed, and it failed because it was not expensive in dollars — it was expensive in leverage.

Corrective action 4 — the safety impact review — held on Northgate and travelled unevenly. Nine directives crossed the $25,000-or-3-day threshold; seven had a signed review before issue. Two were signed after, and one of those was mine — a two-day resequence I directed on a Friday afternoon and papered on Monday because I told myself it was small. It was not small. It moved a mechanical crew into a zone above a ceiling crew. Across the company a year later, four of Kestrel's eleven active jobs were running the review as designed, three a version of it, and four not at all. A procedure adopted after an incident travels at the speed of the people who were in the room. Bea and I were in the room. Four jobs had neither of us.

Corrective action 7 decayed for the most interesting reason of all — it was killed by its own success. The potential-severity trigger was correct. Then reporting climbed from about 3 a week to 11, and 63 events met the trigger. We fully investigated 44; the other 19 got a form and a closure, because four hours × 63 is more investigative capacity than a project has. We built a reporting system that worked and an investigation system that could not absorb what it produced, and that failure looks exactly like backsliding when it is really a capacity problem. The fix we adopted in month 15, and the one I recommend: two-tier triage. Potential-severity events get the full treatment; repeat-mechanism events get a single pattern review across all of them; and you stop pretending you will do forty-four of anything properly.

And the decay pattern generalizes. Look at the two lists again:

That table is worth more than the seven corrective actions. Design your controls against it before you write them, not fourteen months after.

One last thing, and it is the reason I still think signing that report was the best thing I ever did on a job. Findings one and two cost $11,100 one-time and fixed a scaffold. Finding three cost almost nothing to write and changed how Kestrel prices a schedule decision. Nobody resisted findings one and two because a corrective action that costs $2,400 gets approved in a meeting. Everybody resisted finding three because a corrective action that adds six hours and a signature to every future acceleration decision the company ever makes does not cost money — it costs optionality, forever. That is the real price of a systemic finding, and it is why organizations that are good at safety are usually organizations that have already decided to pay it.

Discussion Questions

1. Arlene Wexler argued that the acceleration is discoverable whether or not you write it in the report, so the only real choice is what it gets found next to. Test that. Under what circumstances would it be wrong, and what would you want to ask your own counsel — before your next incident, not after?

2. Corrective action 6 cost nothing and decayed first, because writing a recovery expectation invites a priced response. Design a version that survives that objection: something a project manager will actually do on a Friday afternoon in month 14, when he is behind, the sub is behind, and a letter will produce a number he does not want.

3. Corrective action 7 was killed by its own success. Build the two-tier triage. Define the potential-severity trigger precisely enough that a superintendent can apply it without calling anyone, estimate the hours per month it consumes at 11 reports a week, and say who does the work.

4. The verification table measures behavior, not documents. Pick any three corrective actions and write a better measure than the one in the table — harder to game, and collectable without adding a person.

5. Northgate finished at a TRIR of 2.43 with a functioning system. In the next case study, a job with no functioning system carried a TRIR of 1.79 for over a year. Both are correctly calculated and one is deeply misleading. Before you read case study 2, write down what you would tell an owner who wants to prequalify contractors on project TRIR — and what you would offer instead.

Your Turn

Take the seven corrective actions above and score every one of them against the held/decayed table at the end of the analysis, before you know the outcome. Four columns: does it change a physical condition; does it give an existing behavior a legitimate channel; does it run off data somebody already produces; is its owner someone who will still be here in a year. One point each.

Now score the last three corrective actions you personally wrote — on a job, in a shop, in a kitchen, anywhere. Same four columns. If most of them score one or zero, you have just found out why the same problem keeps coming back, and you did not need an incident to find out.

Then do the harder half. Take the one that scored lowest and redesign it so it scores three or four without spending more money than the original. If you cannot, write one paragraph explaining what it would actually take — a person, a budget line, an authority somebody has to give up — and take that paragraph to whoever can grant it. That paragraph, not the corrective action, is the real deliverable.