False Precision

Understand first *why* false precision is so hard to resist, because the pull is psychological as much as technical. Humans trust numbers, and we trust precise numbers most of all. A figure like '43.7 degrees C' carries an air of measurement - someone, surely, must have determined it carefully to arrive at that decimal. A range like 'somewhere between 40 and 47, we cannot say where' feels weaker, vaguer, less expert, even though it is the more honest and more useful statement. So there is a constant temptation, in reports and conversations and design decisions, to collapse the honest range down to the confident single value - to say the number and drop the spread - because the single value sounds like knowledge and the range sounds like hedging.

The format of the artefact makes it worse. A future weather file is not presented as an estimate with error bars; it is presented as a file, byte-for-byte identical in structure to a real measurement, with every hour filled in to full precision. Nothing in the file itself flags that its 43.7 is a construction and a station's 43.7 is a record. The simulation software then treats both the same way, runs the same physics, and returns a result to the same number of decimals - 'peak operative temperature 31.4 degrees C, 128 hours of overheating' - carrying not a hint that the whole chain rests on one uncertain scenario. Precision propagates; uncertainty evaporates. By the time the result reaches a slide, it can look like a measurement of the future.

This is why false precision is not laziness or stupidity - careful, expert people fall into it constantly. The tools are built to produce precise numbers; the culture rewards confident ones; and the honest object, the range, is harder to state, harder to draw and easier to forget. Naming the trap is the first defence: every time you see a single, precise number attached to a future climate or a future-weather result, train yourself to ask reflexively - *where is the range this number came from, and why has it disappeared?* The number is almost never wrong because it is too imprecise. It is dangerous because it is far more precise than the thing it describes.

To respect the range you have to know why it exists - and it is not one uncertainty but a cascade of them, each stacking on the last. Trace the chain that produces a future weather file and you find at least four distinct sources of spread.

The emissions scenario. How hot it gets depends first on how much greenhouse gas humanity emits over the coming decades - and that is not a physical unknown but a matter of human choices, economies and politics that no model can predict. So projections are made under several scenarios (from deep cuts to high emissions), and by late century these diverge enormously. Choosing one scenario is not finding the truth; it is selecting one branch of a future that has not been decided. This alone means there is no single future to be precise about.

Model spread. Even for one emissions scenario, the world's climate models do not agree. They represent clouds, oceans and feedbacks differently, and so return different warming for the same inputs. The honest practice is to look across an ensemble of models, which gives a range for a single scenario - not because any one model is careless, but because the system is genuinely hard and legitimately uncertain.

Downscaling to a site. Global models work at coarse resolution - grid cells far larger than a building. Getting from that to the weather at a specific plot requires downscaling, which adds its own uncertainty, especially for local effects like urban heat, coastal influence or terrain that the global model never saw.

The morphing method. A future weather file is often built by 'morphing' a historical file - stretching and shifting its numbers to reflect projected change. That method embeds assumptions (that the pattern of a future year resembles a warmed version of a past one) which are reasonable but not certain, and different methods give different files. Add internal year-to-year variability on top, and the picture is complete: by the time a single hour of a single future file shows you 43.7 degrees C, that figure has passed through four or five layers of choice and uncertainty. The decimal is real; the certainty it implies never existed.

If the future is a range, then designing to a single value is designing to a point that will almost certainly not be the one that arrives - and worse, it invites brittle optimisation, a building tuned so exactly to one predicted future that it fails if reality lands somewhere else in the range. The disciplined alternative is to make the range the design object, not an inconvenience to be averaged away.

In practice this means several habits. Where you can, run more than one future: test the design against a moderate scenario and a high one, against a couple of models rather than a single file, and see not one answer but the spread of answers. What you are looking for is not the precise number but the *shape* of the risk - the direction (hotter, more humid, more extreme), the range (how wide the plausible outcomes are), and the severity (how bad the worse end gets). A design decision that holds across the whole spread is robust; one that only works at the mild end of the range is a gamble dressed as a plan.

This reframes the goal from optimisation to robustness. Rather than tuning a building to perform perfectly in one predicted 2050, you design so it performs acceptably across the range of plausible 2050s and 2080s - and, crucially, so it stays *survivable* at the severe end even if that is where reality lands. Passive survivability - a building that stays within safe limits in a heatwave even when cooling and power fail - is the ultimate range-based design move: it does not bet on a number, it guarantees a floor. Sensitivity testing helps too: change your key assumptions and see how much the answer moves; where a small change in scenario swings the result wildly, you have found a fragility worth designing out. The mindset shift is total. You stop asking 'what will the future be?' - a question with no honest single answer - and start asking 'across the plausible futures, does my design keep people safe and comfortable, and where does it break?' That is a question you can actually answer, and answering it well is the whole craft. The binding quantification stays, as ever, with qualified engineers, validated tools and the codes; your job is to insist the analysis reports a range and to design against it.

Honest analysis can still be undone at the last step - the moment you tell someone else the result. A client, a committee, a colleague hears a range and often wants it collapsed into a number they can act on, and the pressure to oblige is real. Resist it, but resist it skilfully, because communicating uncertainty badly is its own failure: state it so vaguely that people ignore you, or so gloomily that they give up, and you have helped no one.

The craft is to make uncertainty *actionable* rather than paralysing. Do not say 'we cannot know the future' and stop - that is true and useless. Say instead what you *do* know: the direction is clear (it will get hotter and more humid), the range is this wide, the severe end looks like this, and here is a design that holds up across all of it. Uncertainty about the exact number is entirely compatible with confidence about the decision - you can be sure the building needs to survive a worse heatwave without being sure of the precise degree, just as you buy insurance against a fire whose date you cannot predict. Frame it as risk management, which people understand, rather than as fortune-telling that failed.

Concretely: present ranges, not points ('40 to 47, most likely around here'), and never quietly drop the spread to sound more authoritative. Attach the scenario to every number ('under a high-emissions path, by the 2050s') so no figure floats free of its assumptions. Show the severe end explicitly, because that is what safety turns on. And be scrupulously honest about the difference between what the analysis showed and what it could not - the model told you about heat, perhaps, but said nothing about the power failing or the flood arriving. Above all, refuse the false-precision sentence in your own mouth: 'the building will be 2.3 degrees warmer in 2050' is a lie of confidence even when 2.3 sits inside the true range, because it claims a knowledge that does not exist. The honest sentence is longer and less impressive and far more useful - and in a warming country where these decisions protect real, vulnerable people, the honest sentence is the only responsible one to say.