Practice Resources · R5

Delivery-Method Notes

The standard of care has not moved. Carriers still assess negligence by outcome and professional judgment rather than by tool, and no serious body has proposed otherwise. That sounds reassuring and is not, because it means the question of who carries the exposure was already answered, by the delivery method, before anyone bought anything. Delivery methods are risk-allocation instruments first. These are notes on the four seams the new work presses hardest, and on how differently each method holds them.

Independent and personal

This document is my own work and my own opinion, written in my own time. It is not issued by, endorsed by, reviewed by, or connected to my employer. It does not describe, quote, or draw on any firm's internal practice, and nothing in it should be read as representing HOK, its practices, or its clients. Everything on this site is written the same way.

I am an architect, not a lawyer, and this is a reading of practice rather than a reading of law. It deliberately names no contract document or clause, because the families vary by country and the editions change. Take the questions to your own counsel and your own insurer, with your own agreements in front of you. Contract terms, professional liability policies, and jurisdictional rules all vary, and any of them can override what follows.

Four seams, before any of the methods

S1
Standard of care, and whether you carry it alone
The standard itself did not change and there is no reason it should have. Ordinary care, exercised by a similarly situated professional under similar circumstances. Tools do not appear in that sentence. What varies between methods is not the standard but the company you keep when it is tested, and whether anyone else in the room has an interest in finding the problem before a claim does.
S2
What actually counts as the deliverable
Drawings, a model, or a continuously maintained data environment. The seam is whether a machine-assisted artifact is a working aid or a representation the other party is entitled to act on. Most agreements in force were written before that question needed an answer, so the answer is usually implied rather than stated.
S3
Reliance, in both directions
Who may rely on your output, and whose output you are entitled to rely on. This is the seam that surprises people, because the first exposure most firms meet will not come from their own use of these tools. It will arrive inside somebody else's deliverable, already looking finished.
S4
The record
Who owns it, who is obliged to maintain it, and what it can prove in two years when someone asks how a decision was reached. A machine-assisted check that found nothing is worth recording as carefully as one that found something, and only one of those feels worth the trouble at the time.

Design-bid-build

The method allocates risk by separating the parties. You design, someone else builds, and the distance between those two acts is where the money is.

S1Standard of care: You carry it alone and visibly. There is nobody in the room with a commercial reason to find the problem early, so a finding becomes a claim without passing through anything that might have absorbed it. That exposure profile is the method, and it does not soften because your own checking got faster.
S2The deliverable: The issued drawings are the contract. The model is usually excluded in writing, and that exclusion is protective. A machine-assisted analysis that never enters the issued set stays a working aid rather than a representation. Keep it that way deliberately, because the alternative is keeping it that way by luck.
S3Reliance: Narrow, one-directional, and late. The contractor relies on your set. You rely on almost nothing from them until submittals arrive, by which point the design has hardened. The tool that shortens your loop does nothing at all to shorten theirs.
S4The record: Yours to own and yours to keep, and nobody asks for it until they ask for all of it. In a method with no early collaborative forum, the project file is the only place your reasoning survives.

Where it presses

Continuous conformance checking produces findings earlier than this method has any mechanism to act on them. You will know in schematic design what the agreement only lets anyone fix through a change order. That is not a defect in the tool. It is the method working exactly as designed, and it is the strongest argument for a different one that I know of.

Ask before you sign

If my check finds it in month three, what in this agreement lets anyone act on it before month nine?

Construction manager at risk

A builder joins during design and eventually holds a price. You buy constructability input early, and in exchange you gain a reader whose interest in your documents is commercial rather than neutral.

S1Standard of care: Still entirely yours, but now tested continuously by a party reading your set for exposure rather than for intent. That is genuinely useful and it is not disinterested. Both things are true at once and the method works better when everyone says so out loud.
S2The deliverable: Pricing documents become a de facto contract artifact well before any permit set exists, and they are priced while explicitly incomplete. What your model implies therefore carries more weight here than in separated delivery, because somebody is putting a number against the implication.
S3Reliance: Two-way and uneven. They rely on your documents to price the work. You rely on their estimate to advise the owner. If either number was produced by a process neither party has examined, the reliance running in both directions is unearned.
S4The record: Split down the middle. You hold the design record, they hold the cost record, and the two are usually reconciled for the first time in a dispute. Date your assumptions, because the estimate certainly dates theirs.

Where it presses

The estimate. Machine-assisted quantity and cost work is arriving faster than anyone's protocol for checking it, and a guaranteed price is the single place where a confident wrong number does the most damage. Ask how the number was produced before you advise an owner to rely on it. That question is now part of the advice, not a discourtesy to the person who brought it.

Ask before you sign

Before I advise the owner to rely on this number, what do I actually know about how it was produced?

Design-build

One entity carries design and construction together. The method exists because the feedback loop in separated delivery is too long, and it solves that socially, by putting the parties under a single agreement rather than by making anyone faster.

S1Standard of care: Unchanged as a standard, materially changed as a position. Inside a design-build entity you advise a party whose commercial interest is close to the owner's but not identical to it. Machine-assisted analysis that finds a problem finds it for the entity. Where the finding goes next is a professional question, and no tool will answer it for you.
S2The deliverable: Performance criteria rather than prescriptive documents. This is the sharpest of the four seams, because a performance criterion is already a statement of intent written to be tested, which is precisely the input these tools use best. It is also where an unstated criterion is most likely to be quietly satisfied by its cheapest available reading.
S3Reliance: The owner relies on the entity rather than on you individually, and often holds fewer independent means of checking it. That asymmetry is the standing criticism of the method. Machine-assisted work sharpens it, because the entity's capacity to produce defensible-looking analysis grows considerably faster than the owner's capacity to interrogate it.
S4The record: Held by the entity, and worth asking about early, because entities formed for one project are dissolved at the end of one project. Establish before closeout who keeps the record and for how long.

Where it presses

Whether the tools are pointed at the design intent or at the margin. Both are legitimate objectives and a well-run project serves both. They are not the same objective, and a system that optimizes whatever it is aimed at will never volunteer which one it was aimed at.

Ask before you sign

When my check finds something that costs the entity money and saves the owner money, what is the path for that finding?

Integrated project delivery

A multi-party agreement with shared risk, shared reward, and usually a mutual waiver of claims. The method's bet is that if the parties cannot sue each other, they will surface problems while the problems are still cheap.

S1Standard of care: Individual, always, whatever the liability structure does. This is worth stating plainly because the collaborative framing can blur it. The waiver allocates money between the parties. It does not distribute your responsibility to the public, and your seal is still your own.
S2The deliverable: A shared model in a common data environment that everyone maintains. This is the only one of the four where the deliverable is genuinely continuous rather than episodic, which makes it the natural home for continuous checking and simultaneously the hardest place to establish who authored what.
S3Reliance: Everyone on everyone, by design and on purpose. An unexamined machine-assisted input therefore propagates further and faster here than anywhere else, and the same waiver that makes the method work also removes the adversarial friction that would ordinarily have surfaced it.
S4The record: Shared, which sounds like the solution and is in fact the hardest version of the problem. Authorship inside a continuously edited model is a convention a team has to establish and maintain. It is not a fact the software records for you.

Where it presses

The best and worst cases for this technology are both here. Best, because the method already rewards information that arrives early. Worst, because a structure built to remove friction between the parties has correspondingly little friction to offer a confident wrong answer.

Ask before you sign

In a model everyone edits, what is our convention for marking what was machine-assisted, and did we agree it before or after the first disagreement?

With an agreement in front of you

01Find the word deliverable and read what it actually names. If it names the issued drawings, your model is a working aid. If it names the model, everything inside it is a representation somebody may act on.
02Find the reliance language, or notice that there is none. A right-to-rely provision is the cheapest protection available against somebody else's machine-assisted error arriving inside your scope.
03Ask who owns the record at closeout and who is obliged to keep it, for how long. A defense you cannot retrieve is not a defense.
04Separate what the agreement requires you to disclose from what you intend to disclose anyway. Only the first of those is a legal question, and the second is the one that determines how the conversation goes.
05Decide which of the four seams your method leaves widest, and write your own protocol for that one rather than for all four. That is what a studio policy is for.

Version 1.0 · Updated August 2026

None of this is new law and none of it is legal advice. The standard of care did not move. What moved is how quickly any of us can now produce analysis that looks finished, and the delivery method you signed already decided who that helps and who has to check it. Read the agreement before you buy the tool.

Free to adapt. If your experience of one of these methods contradicts what I have written, that is the most useful thing you could send me. hello@frankcauthen.com

The protocol this document keeps pointing at is the Studio Policy Starter. The argument underneath all of it is in The Tighter Loop.

← All practice resources