[Bldg-sim] cheat-gpt

[Richard Sapwell]

*Hi Chris,*

You’ve raised an important—and frankly unavoidable—question for anyone
working in building simulation, NatHERS, LEED, or broader performance
modelling:

*If an LLM can rapidly generate the appearance of a complete modelling
package, what happens to the value of the modeller?*

As someone who actively uses LLMs (ChatGPT) to support the production of
reports—particularly complex Performance Solutions and NatHERS
documentation—I see both the opportunity *and* the real risk.
------------------------------

*The Risk: “Autocorrect on Steroids” Can Mimic Output Without Doing the
Work*

You’re right: LLMs can now generate

   - unmet load tables,
   - calibration summaries,
   - error logs,
   - and even plausible-looking LEED or NCC/NatHERS documentation,

*all without any underlying simulation ever being run.*

This creates two exposures:

*1. Reviewers may not detect fabricated modelling artefacts*

Even experienced certifiers can struggle to identify fabricated tables if
they are numerically plausible.
NCC and NatHERS documentation is especially vulnerable because reports
often follow rigid templates.

*2. The modelling profession risks being judged by its paperwork, not its
competence*

If the visible output (the report) becomes easily faked, then the *perceived
value* of modelling may degrade—even though the true value is in the
physics, domain reasoning, and regulatory interpretation behind the scenes.
------------------------------

*So what is the value a competent modeller still provides?*

Fundamentally:

*1. Understanding the physics and constraints*

LLMs can *describe* thermal bridging, infiltration limits, control logic,
shading geometry—but they cannot validate that a proposed model follows the
laws of thermodynamics or matches local climate files, geometry, or
construction systems.

*2. Regulatory interpretation and compliance judgement*

“EE-04 weighted existing vs proposed floorspace” or “NCC 2019 Amdt.1 vs NCC
2022 equivalence under Building Surveyor discretion” requires *professional
reasoning*, not pattern-matching.

*3. Ethical accountability*

A human must sign the report, take responsibility for accuracy, and engage
with BS, council, or GBCI when queries arise.
LLMs cannot accept liability.

*4. Integration with design reality*

Your point about tying design modelling to *actual operation*
(post-occupancy calibration) is spot on. This is something only competent
practitioners can deliver.
------------------------------

*How do we improve—and protect—the value of our work?*

*1. Strengthen the link between models and real buildings*

Your proposal is exactly where the profession needs to move:

   - Require calibration against measured operational data
   - Use design models to support commissioning and tuning
   - Close the loop between predicted and actual performance

This pushes our value upstream and downstream—far beyond simply generating
documentation.
------------------------------

*2. Escrow the model input files*

I strongly support this.

For LEED, NatHERS, NCC performance solutions, mechanical simulations, etc.,
requiring:

   - the native model files,
   - all geometry and input assumptions,
   - weather files,
   - software versions,

*to be lodged with a trusted authority (GBCI, local regulator, building
surveyor, etc.)*

would limit the risk of “paper-only” submissions.
Reports must correspond to verifiable inputs.

This is analogous to the way:

   - structural engineers submit their calculation sets,
   - energy auditors submit calibration data,
   - certifiers maintain audit logs.

------------------------------

*3. Accreditation that requires evidence of actual modelling competence*

This could be NatHERS, CIBSE, ASHRAE, or emerging Australian equivalents.

An LLM can write a report.
It cannot apply ASHRAE 140, NCC V2.6.2.2, or interpret ambiguous DTS
pathway conflicts.

A competency-based accreditation system reinforces the value of the
practitioner, not the document.
------------------------------

*4. Transparency in documentation*

Professionals should adopt a standard such as:

   - Model version
   - Software version
   - Data sources
   - Assumptions
   - Calibration status
   - QA/QC logs
   - Sensitivity tests

This is the kind of information an LLM *cannot fake accurately*—without the
real model behind it.
------------------------------

*Controversial? Maybe. Necessary? Yes.*

Your proposals won’t just protect the profession; they will improve the
credibility of modelling outcomes and reduce the temptation for individuals
(or clients) to push for “report-only” shortcuts.

The uncomfortable truth is:


*LLMs threaten the commoditised paperwork side of our profession, but they
strengthen the value of genuine expertise and verifiable modelling.*

If we move towards transparent inputs, post-occupancy calibration, and
audited modelling files, then the future modeller becomes:

   - a systems thinker,
   - a regulatory interpreter,
   - a data integrator,



—not a “report generator.”
------------------------------

*Thanks for raising this, Chris. This is exactly the kind of conversation
the industry needs to be having right now.*

Cheers,
Richard
(Building Designer & Energy Efficiency Assessor)

ecodesigns know the elements, work with them


On Thu, 11 Dec 2025 at 00:40, Chris Yates via Bldg-sim <
bldg-sim at lists.onebuilding.org> wrote:

> Hi there,
>
>
>
> Just a philosophical one; maybe even an existential one 😐
>
>
>
> Given LLMs’ “autocorrect on steroids” abilities, what risk do we see
> ourselves exposed to as modelling professionals?
>
>
>
> Let’s give an example…
>
>    - A full LEED application including reports, unmet load tables,
>    “error” reports.
>    - Not based on modelling – all reports and outputs generated with LLMs.
>
>
>
> I think it would be relatively easy to do this, and present something that
> would stand up to review.
>
>
>
> I think this poses the following questions:
>
>    1. If our main value offering can be replaced with auto-complete, what
>    is the value of what we’re doing, anyway?
>    2. How can we improve the value?
>    3. How can we protect that value?
>
>
>
> I’d like to propose some solutions:
>
>    - Tie Design stage models into the Operation of the building –
>    calibrate
>    - Rather than just submitting reports, place model input files into
>    *escrow* with GBCI (or their local equivalent)
>
>
>
> Controversial?
>
>
>
> Cheers
>
> Chris
> _______________________________________________
> Bldg-sim mailing list
> http://lists.onebuilding.org/listinfo.cgi/bldg-sim-onebuilding.org
> To unsubscribe from this mailing list send  a blank message to
> BLDG-SIM-UNSUBSCRIBE at ONEBUILDING.ORG
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.onebuilding.org/pipermail/bldg-sim-onebuilding.org/attachments/20251211/a2304178/attachment.htm>