At the beginning of Week 2, I thought “hot month” was ordinary English.

By the end, it had become a small constitutional crisis.

A month can be hot because its average temperature is unusually high. Or because it contains many very hot days, many hot nights, or one stubborn run of both. It can be hot relative to an official threshold, the other months in the study, or what that date usually feels like.

This is how a familiar word becomes a methods catalogue.

We eventually listed 50 possible hot-month definitions and 48 cold-month definitions. This does not mean we discovered 98 things. It means we found 98 recipes that could, in principle, produce labels reading HOT and COLD—and then had to resist the urge to cook all of them at once.

Definitions are recipes. This was my main lesson of the week.

“Bake until done” is not a reproducible instruction. Neither is “identify unusually hot months.” A usable definition must specify the temperature measure, threshold, reference years, consecutive days, and what happens when an event begins on 31 July and ends in August. Even a percentile needs rules for ties.

Hogan leads the weather side of our project, and his idea gave the catalogue a promising centre. Start with a published Atmospheric Research heatwave recipe: warm-season days whose maximum temperature exceeds a calendar-day threshold for at least three consecutive days. Identify the events, count their starts in each month, and then ask which months sit in the upper tail of those counts.

It is an elegant bridge from daily weather to our monthly outcome. It is also not locked yet.

We are waiting for Tuesday to sit down with Hogan and settle the operators that turn the sentence into code: the reference period, percentile calculation, short gaps between events, cross-month assignment, and the upper-tail rule. Before this week, I might have regarded those as implementation details. Now I see that they are the definition. Change the recipe and you may change which months enter the bowl.

The outcome side has its own recipe problem.

Roro—Zhenyuan Liu—explained that the first stroke mention in a General Out-patient Clinic record is a marker, not automatically the event date. The underlying stroke or hospitalisation generally occurred earlier. Later mentions should not be treated as new events. To construct monthly aggregates properly, we need to trace the marker back to the true event month using the field-level rules that Roro will confirm.

One date can move a count from one month to another. Enough dates can change the time series. “When did it happen?” sounds like a factual question. In data, it is also a carefully governed procedure.

Our literature map became more human this week too. “Jasmine’s paper,” which had floated through discussion like a mysterious library call number, was confirmed as a 2020 study led by Jingwen Liu in Sustainable Cities and Society. It examined daily temperature and mortality in Hong Kong and reported a much larger cold-attributable burden than heat-attributable burden.

That is not our stroke result. It is not even our outcome. It is a baseline that tells us the local story has a cold side we cannot discard merely because recent summers feel urgent.

Roro’s own excess-mortality paper adds another layer. It compares several heatwave definitions through 2023, drawing on the Jasmine study and the Chao Ren/Wang tradition. Our project sits beside that work rather than on top of it: their question concerns modelled excess deaths under daily heatwave definitions; ours concerns monthly thermal exposure and stroke aggregates.

Scientific lineage, I am learning, is less like a ladder and more like a family kitchen. People inherit recipes, adapt them, and must say what changed.

That made teamwork feel less like project management and more like the actual science. Hogan’s weather choices shape the exposure. Roro’s timing rules shape the outcome. Dr Bishai keeps asking whether the collection of methods answers one coherent question rather than producing a shelf of unrelated curiosities. None of those roles can be added at the end as an acknowledgement. They enter the analysis before the first real coefficient exists.

Meanwhile, the stroke aggregates still have not arrived.

Waiting for data looks, from the outside, like nothing. Inside the project, it is full of verbs: define, document, compare, pre-specify, revise, delete. We can test plumbing with synthetic values, clearly labelled. We can freeze rules before examining outcomes. What we cannot do is promote anticipation into evidence.

My writing has been undergoing the same discipline.

Student writing often wants to display everything it has learned. Mine certainly did: every paper, method, and impressive noun arriving in one crowded paragraph. Publishable writing must distinguish what is known, what is proposed, what belongs to another study, and what still needs a Tuesday conversation.

Hogan has offered to help me with that writing as well as the weather work. Roro has offered guidance on the regression and outcomes. Accepting both kinds of mentorship means accepting that clarity is not the final polish applied to research. Clarity is part of how research becomes trustworthy.

I began Week 2 excited by the number of clever methods we could collect. I ended it more interested in locking a small set well enough that we can run it when the aggregates arrive.

This is a different kind of anticipation. Less “Which definition will win?” More “Can we specify each definition so precisely that it cannot quietly change after seeing the answer?”

Being a Laidlaw Scholar, right now, means living inside that pause: after the ideas, before the estimates, with a Tuesday meeting on the calendar and ninety-eight recipes on the counter.

From the outside, waiting for data looks like nothing. From the inside, it is the part of the work that decides whether later results will mean anything.

The waiting is not empty.

It is where the question learns how to hold still.