Site, Climate & Context

Start by drawing the line clearly, because site work is where people most often ask Claude to do the one thing it fundamentally cannot. Claude has no senses and no live data. It cannot measure your plot, read your contour survey unless you give it the numbers, know your neighbour's window heights, or feel your site's afternoon heat. If you ask it "what is the climate of my site" with only a place name, it answers from training-data generalisations that may be dated, coarse, or simply wrong for a microclimate - a valley, a coast, a dense city block behaves nothing like the regional average. And without web access it does not know this year's data at all. Treat any place-fact it volunteers from memory the way you treated a precedent from memory: a lead to confirm, never a finding.

What Claude can do, superbly, is reason over data you provide. Give it the actual figures - latitude, monthly temperature and humidity normals, prevailing wind direction, rainfall, sun angles, the plot dimensions and levels, road and access details, adjacency and overlooking notes - and it becomes a fast, structured analyst of exactly that material. It will organise it, spot tensions ("your best view is west, but that is also your worst afternoon heat gain - these fight"), connect it to design implications, and hold far more variables in play at once than you comfortably can at the end of a site visit. This is the retrieval-versus-generation distinction again, moved outdoors: grounded on your site data it reads and reasons; ungrounded it guesses.

So the practical rule for the whole lesson is: Claude is a context analyst, not a site surveyor. Your job is to do the seeing, measuring and sourcing - stand on the site, commission the survey, pull the climate normals from a real dataset, read the byelaws - and Claude's job is to help you make sense of it. When you keep that division, Claude is genuinely valuable from the first site visit onward. When you blur it and let Claude 'tell you about the site,' you get confident, generic, sometimes wrong context that quietly steers the design off the actual place.

The highest-value move in this lesson is turning the chaos of a site visit into a clean context brief. After a visit you usually have a mess: a scribbled sketch, a voice-note, thirty phone photos, some measurements, a climate table you downloaded, the byelaw constraints from Lesson 2.2, a screenshot of the survey. Individually useful, collectively unstructured. Dump the lot into Claude and ask it to organise, and you get in minutes what used to be an evening's work.

A context-brief prompt looks like this:

Here is everything from my site visit and desk research (pasted / attached):
- site dimensions and levels
- monthly climate normals (temp, humidity, rainfall, prevailing wind)
- my voice-note observations (transcribed below)
- adjacency, access and overlooking notes
- the byelaw constraints (setbacks, FAR, height)

Structure this into a context brief with sections: Location & climate,
Orientation & sun, Access & circulation, Views & overlooking, Topography
& drainage, Neighbours & context, Constraints, and - separately - the
key Opportunities and Tensions you see for the design.
Use ONLY the data I gave you. Where something is missing, list it under
"Still to confirm on site" rather than assuming.

The payoff is threefold. First, completeness: the section headings act as a checklist, and the "still to confirm" list is gold - it catches what you forgot to record before you leave the area, turning Claude into a pre-emptive snag-list for your next visit. Second, structure: your rambling notes become a document you can hand to a client or a collaborator. Third - and this is where Claude adds real thinking - the Opportunities and Tensions section. A good analyst does not just list facts; it notices where they pull against each other. Claude, holding all your variables at once, is good at surfacing these: the west view versus west heat, the quiet rear versus the poor access, the great south light versus the neighbour who overlooks exactly there. You will not agree with all of them, and some will be naive - but as a foil that forces you to confront every tension in your own data, it is genuinely useful, and it is grounded, because every tension traces back to numbers you supplied.

Once the brief exists, Claude becomes a strong partner for the reasoning that follows - climate-responsive strategy - provided you keep feeding it real numbers and keep the verification where it belongs. Give it your latitude and climate normals and it will reason about sun path, overheating risk, useful daylight, cross-ventilation potential and the passive moves that suit your climate type. For a hot-dry site it will push thermal mass, shaded courtyards, small openings, evaporative cooling; for warm-humid it will push cross-ventilation, shade, raised floors, lighter construction. This is sound climate-responsive thinking, and Claude articulates it well and connects it to your specific orientation and constraints.

Two cautions keep it honest. First, orientation reasoning is latitude- and hemisphere-specific, and it is a classic place for a confident slip - the "orient to the south" advice that is right for much of the northern hemisphere is wrong in the southern, and the specifics of shading angles depend on your exact latitude. So state your latitude and hemisphere explicitly, and sanity-check the directional logic against first principles or a real sun-path tool rather than trusting the prose. Second - and this is the hard limit - Claude reasons about performance; it does not simulate it. A remark that a room "will overheat in summer" or "gets good daylight" is qualitative reasoning from the data, not a computed result. It is excellent for shaping direction and asking the right questions early; it is not a substitute for a daylight simulation, a thermal model or a shadow study when numbers matter. Use Claude to decide what to test and why, then test it in a real tool.

The useful framing is Claude as the experienced design tutor walking your site data with you: quick, well-read, full of relevant strategy, able to connect climate to form - and confidently wrong just often enough that you check the directional and quantitative claims. Everything it says is a hypothesis grounded in your data and refined by your judgement and, where it counts, by proper analysis. That is a big step up from a blank page, and a long way short of a verified performance report - which is exactly the right place for it to sit.

Site is more than sun and wind. The best context briefs read the human and cultural setting too - the character of the neighbourhood, movement patterns, local building traditions, materials of the place, what the street feels like, how the community uses the surroundings - and here Claude helps in a more careful way. On the tangible, given data (traffic counts, adjacent uses, plot history, local material availability you researched) it structures and reasons well. On the intangible and local - the feel of a street, a region's vernacular, cultural sensitivities - it can offer useful prompts and general knowledge, but this is exactly where memory-generalisation is thinnest and where your own eyes, and local voices, matter most. Claude can remind you to consider the vernacular; it cannot tell you what your particular street actually is.

So use Claude to widen your questions, not to answer them for the place. "What contextual factors am I likely to be under-weighting for a small infill site in an older Indian neighbourhood?" is a great prompt - it surfaces a checklist (party walls, existing drainage, construction access down a narrow lane, dust and noise to neighbours, matching eave lines, local artisan availability) that sharpens your next visit. But whether any of it is true for your site is something only the site can tell you. This keeps you from the real failure mode of AI-assisted context work: a fluent, plausible, generic 'analysis' that could describe anywhere, quietly detached from the actual place. A context brief that could be about any site is not a context brief; it is filler.

Hold the whole lesson together with one discipline: every meaningful line in your context brief should trace back to something real - a measurement you took, a dataset you pulled, an observation you made, a local source you consulted. Claude's role is to gather those into structure, surface tensions and opportunities across them, and pressure-test your reading with questions - all of which makes you faster and more thorough. What it never does is replace the site visit, the survey, the local knowledge or the simulation. You bring the place; Claude helps you understand what you brought; you decide what it means for the design. Keep that order and the context brief is grounded, specific and genuinely about your site - which is the only kind worth writing.