Chapter 25 — Key Takeaways

A one-page reference card. Of everything in this book, this is the page I would most want taped inside the cover of a project engineer's notebook — because the skills on it are cheap to learn, they are worth six figures on a single job, and almost nobody is taught them.

Key Takeaways

• The paper trail is the project's memory, and contemporaneous records are worth roughly ten times reconstructed ones. A record made on the day, by someone who had no idea it would ever matter, is evidence. The same facts assembled eight months later by someone who very much wants them to matter is argument — and argument loses.
• A submittal log is a schedule, not an inventory. Sorted by specification section it is a bibliography: accurate, complete, and useless on Monday morning. Sorted by submittal due date, back-scheduled from required-on-site, it is a look-ahead. Rebuilding the Northgate register that way took one afternoon in week seven and surfaced three items that were already 25 to 34 days late.
• There is exactly one fixed point in a procurement chain: the required-on-site date. Everything else is computed backward from it — installation lead, delivery, fabrication from release, the approval cycle including a resubmittal allowance, and the subcontractor's preparation time. One more subtraction gives you the award-by date, because nobody prepares shop drawings for a job they have not been awarded.
• If a computed date lands in the past, you have found a problem on a spreadsheet instead of on a job site. Ironbridge Steel's subcontract was executed March 7; the back-schedule with one resubmittal in it says February 20. The chain was fifteen days infeasible on the day everyone shook hands, and nobody knew, because nobody ran the calculation.
• Track the fabricator's release date, not your own approval date. Kestrel's milestone was April 15. Ironbridge's mill release was April 18. Only one of those was real, and it lived on a production schedule Kestrel never asked to see. Ask every long-lead supplier at buyout, in writing: what is the last date you can release, and what happens if you miss it?
• A submittal sitting on your desk is the most expensive object on the site. Eleven days at Kestrel instead of the planned five — six days of desk time — cost the mill slot, slipped steel erection twenty-three calendar days, and produced a $168,000 acceleration. That is $28,000 per day of desk time. No crew, no crane, and no machine on a job site burns money at that rate.
• A resubmittal allowance protects you against comments. It does not protect you against a question the documents cannot answer. The curtain wall carried a correctly sized 21-day allowance and still lost nine weeks, because forty percent of the north-elevation transition details were marked "by others" and "others" turned out to be nobody. That is a scope gap, and you cannot buy your way out of one with schedule float — you find it in the scope review, months earlier.
• Approval of a submittal transfers almost nothing. By submitting, you certify that you have reviewed it and verified materials, field measurements, and field criteria, and coordinated it with the rest of the work. The design professional's review is for conformance with the design concept, and it does not relieve you of responsibility for deviations you did not specifically flag in writing, or for errors and omissions in the submittal, dimensions, quantities, fabrication processes, techniques of installation, safety precautions, or coordination with other trades. A stamp is not a shield. Read your own contract's version, because the wording varies — but the structure does not.
• Therefore: every deviation gets a cover letter. Item by item, on the front, numbered, asking for specific written approval of each. Not a note in the corner of sheet 147. Approve it and you are covered for exactly what you listed; reject it and you found out before anybody cut aluminum.
• Never send the design team a submittal you have not actually reviewed. If you are transmitting packages you have not opened, you are not running a submittal process — you are running a mail room, and you have also just falsely certified that you checked it.
• "Approved as noted" and "revise and resubmit" are separated by weeks. Ask, at the preconstruction meeting, that the reviewers use "approved as noted" on the four to eight gating items wherever their comments do not go to design intent. It costs nothing and nobody is being asked to lower a standard — you are asking for a stamp choice on items where the difference between two stamps is a manufacturing slot.
• Publish a ball-in-court report every week, without editorial comment. In week seven at Northgate the answer was 138 items sitting with subcontractors, 11 with Kestrel, and 14 with the design team. It is almost never the architect first — but "out for review" is visible on a log and "never asked for" is not. The effect of publishing the report is not informational; it is social. Nobody wants to be the top row.
• Rank the aging report by what each item gates, not by days out. Twelve days out on an item with float is fine. Nine days out on the item that gates a mill slot in two days is an emergency.
• Batching destroys your own argument. Submit eighty items in one week and then complain about turnaround, and you have manufactured the problem. Level the submission curve: ten items a week reviewed in fourteen days beats forty a week reviewed in forty.
• An RFI is a request for clarification where the documents are genuinely ambiguous, conflicting, incomplete, or contradicted by a field condition. It is not a substitution request, not a means-and-methods decision you should be making yourself, not a question the drawings already answer, and not a demand with a threat attached.
• Your RFI log is a reputation. Every unnecessary RFI spends a little of the design team's willingness to take you seriously, and you will need that intact on the day you send the one that matters. Northgate: 287 RFIs on $47.5M, mean response 9.4 calendar days. Rivermont Elementary: 496 RFIs on $22.4M — roughly one per $45,000 of work — mean response 21 days, 63 items open past thirty days at substantial completion. By RFI 300 the architect had stopped reading them carefully, and he was right to.
• The single biggest determinant of a one-round answer is a proposed solution. A reviewer who must design an answer needs hours they do not have. A reviewer who must evaluate one needs minutes — and the answer that gets approved is usually the one you proposed.
• Read every RFI response with one question in front of your mind: did the answer change the contract? An RFI response that adds scope is a change, and because it arrives on a form with your own number on it, in a paragraph that sounds like a clarification, the crew builds it and nobody prices it. That is a donation, and it establishes a course of dealing that makes the next one harder to charge for.
• There is exactly one current set. Everything else is marked REFERENCE ONLY or destroyed. A superseded level-3 partition plan taped to a field-office wall cost $340,282 over nine working days and 118 partitions — on a bulletin Kestrel had received and acknowledged in writing. A document is not controlled until the last person who acts on it knows its revision. Transmittals prove you sent it; only a layout-verification habit proves somebody used the right one.
• A platform amplifies discipline; it does not create it. If your log is sorted by specification section, cloud software gives you a beautifully rendered, mobile-accessible, permission-controlled log sorted by specification section — and it will faithfully display your empty required-approval field, in color, on every device your team owns.
• CO #14 is the whole argument in one table. The work was real. The owner's representative had genuinely authorized it. Kestrel was, in the ordinary sense, right — and it did not matter.

• Read the gap, not the settlement. Every cost line with a zero gap was one where a third party generated a document as part of its own routine — batch tickets, supplier invoices, a testing lab report, a subcontractor's executed change order. Every line with a large gap was one where a Kestrel employee had to stop and write something down in the moment, under pressure, and did not. If your documentation depends on people remembering to document under pressure, it will fail under pressure — which is exactly when you need it.
• The exchange rate on that failure: under two hours of paperwork was worth $43,650. Forty minutes of time-and-material tickets was worth about $15,480 — roughly $387 a minute. And the $43,650 consumed 2.4 percent of Kestrel's entire $1,804,800 fee on a $47.5 million project, on work everyone agreed had been performed.
• The settlement premium is not comfort. Pri Sethi paid $21,750 above what Kestrel could prove because she knew the work was real and she had started it. That premium is discretionary, it shrinks as the relationship ages, and it does not exist at all in mediation, arbitration, or litigation. Hold it this way: $142,750 was not a good outcome that fell a little short. It was a $43,650 loss that a slightly less generous counterparty would have made a $65,400 loss.
• Write facts, not conclusions, blame, sarcasm, or predictions of disaster. The record will be read, in an order you did not choose, by someone whose profession is finding your worst sentence. Writing nothing is not the answer — writing nothing is what cost $43,650. Write well: dates, times, quantities, names, crew sizes, what was said and by whom, what you did about it. And note that the well-written version of a daily report is the same number of words as the bad one. The entire difference is that the author knew what the document was for.

Action Items — What to Do on Your Job This Week

1. Re-sort your submittal log by required-approval date and look at the top twenty rows. If your log does not have that column, add it — even if you only populate it for the twenty items you know are long-lead. That is one afternoon, and on Northgate that afternoon was worth $168,000.
2. Pick your three longest-lead items and run the full back-schedule with a resubmittal cycle in it. If any computed award-by or submittal-due date has already passed, you have found today's most important problem, and you have found it early enough to do something about it.
3. Call your three long-lead fabricators and ask one question: what is the last date you can release against your production slot, and what happens if you miss it? Write the answer in the log as its own column. That is the date that is real.
4. Produce a ball-in-court report and send it to the whole team — including the architect and the owner's rep — with no commentary. Five rows, one page. Then do it again next Monday, and the Monday after.
5. Take the last five RFIs your team sent and score them against the six elements. Count how many contained a proposed solution. Then look up how many rounds each one took. The correlation will make the argument for you.
6. Take the last ten RFI responses you received and sort them: clarification, interpretation, or direction that adds scope. If you find one in the third pile that got built without a change order, you have just measured what this habit is costing you.
7. Walk into the field office and the two busiest work areas and check the revision block on every drawing posted on a wall against the current drawing log. Fifteen minutes. That is the $340,282 check.
8. Add one column to your submittal log: issued to installer — date / signed by — and one rule: no scope starts until that cell is filled. It closes the loop that quality management depends on.
9. Name one person as the owner of the log. Not "the project engineers." One name. A log with shared ownership degrades in about three weeks, and the degradation is always the same: statuses first, then dates, then people start keeping private lists.

Common Mistakes — and the Fix

Decision Framework

1. The back-schedule — run this on every long-lead item

Work right to left. The only fixed point is the far right.

  SUB          GC          A/E        [REVISE +     RELEASE     FABRICATE   SHIP /    ON
  PREPARES ->  REVIEW  ->  REVIEW  -> RE-REVIEW] -> TO SHOP  -> / MANUF. -> DELIVER-> SITE
   ^                                                                                   ^
   |___________________ compute in THIS direction ____________________________________|

  START HERE:  required-on-site date, from the CPM predecessor of the installing activity
     − installation lead   (rigging window, crane availability, pad poured, opening still open)
     − delivery / transit  (plus oversize permits, offload, staging)
     − fabrication FROM RELEASE  (not from order — release means approved and cleared to build)
     − approval cycle:  A/E review period
                      + ONE full resubmittal cycle (revision + re-review)
                      + your own review and transmittal days
     − subcontractor preparation time
     = SUBMITTAL DUE TO YOU
     − preparation time again, or the sub's mobilization
     = AWARD-BY DATE (execute the subcontract or PO no later than this)

  Then check ONE more thing the log will not tell you:
     ► the FABRICATOR'S RELEASE DATE against its production slot.
       Ask for it in writing. It is usually earlier than your approval milestone,
       and it is the only one of the two that is real.

Any computed date in the past is today's problem. Any date within two weeks of today is this week's.

2. The anatomy of an RFI that gets answered in one round

Six elements. All of them. In this order.

One optional seventh line, used carefully: "Kestrel proposes to proceed on this basis unless directed otherwise." Appropriate when your proposal is genuinely no-cost, low-risk, and reversible. Not appropriate when it costs money, changes design intent, or touches a life-safety or code-governed condition.

3. Reading the response — did the answer change the contract?

An RFI response is generally not a change order. The instruments that authorize changed work — change order, construction change directive, architect's supplemental instruction — vary by contract form, and public work is usually stricter and less forgiving. Look up yours, once, and write the notice period on the inside cover of your log.

4. The current set — know the instrument, know the money

5. The two tests

The record test: Write it so a stranger can use it in a year. Not so your PM understands it — they were there. Not so it is defensible — that comes for free if the facts are there. So that a person who has never met you, never walked the job, and has no memory of March can pick it up, read it once, and know what happened. Every good record I have written passed that test; every bad one failed it the same way, by assuming a reader who already knew.

The archive test: Can a person who has never seen the project find the answer to "what happened on the north elevation between April and June" in under an hour? If yes, you have an archive. If no, you have a storage unit.