How I Plan a Trip

1. Why this page exists

We met a couple on holiday who watched us navigate a city off a single file on my phone, and asked how it was made. Not where we’d been, but how the thing was made. That was the second time somebody had asked, and it is the reason this page is here rather than in a folder. Lived

The honest answer is longer than it should be, so here is the short version first.

An AI will write you an itinerary in thirty seconds. It will be plausible, well-organised, and it will name real places. This is not a hard problem any more and it has not been for a while. Every large travel company has shipped a version of it, and the free models are good enough that paying for one is difficult to justify. Researched

That was never the hard part. The hard part is everything the itinerary does not know it is wrong about (the time you don’t actually have, the constraint you stated too vaguely for anyone to act on, the local knowledge no amount of searching surfaces) and then the entirely separate problem of getting from a good plan to a booked trip.

I know this because I built a planning system, ran it on real trips, and watched it fail. Nine versions later, the corrections cluster in a few specific places, and none of them are “it didn’t suggest enough restaurants.” What follows is those places.

One thing to be straight about before you read on: this is a system built by one person for two travellers, and it is calibrated to us. The mechanisms generalise. The settings do not. Where something is a preference rather than a principle, I have tried to say so. Lived

2. What it actually produces

Worth seeing before you decide whether the rest of this is worth your time. Three things come out the other end.

A day you can follow without thinking. Not a description of a day. A list of timed rows you read one-handed, while walking, on a phone with eleven percent battery.

Illustrative. The shape of a day, not a day from a trip I took.

Read it again and notice what is not there. No paragraphs. No “you might also consider”. No three-way choice inside a row: one default per slot, with the alternates at the back where they are available and not in the way. And every tip carries a reason rather than an adjective: “wholesale hour, before the stalls turn touristy” tells you why you are standing there at eleven in a way that “vibrant local market” never does. Lived

The tags are the system showing through. A sit-down row is there because somebody in the party cannot walk twenty minutes without one. The meal row is fixed because two people in the group degrade when food is late, for unrelated reasons that happen to converge. The bail-out exists because a day with a child in it needs one that does not feel like a failure when you use it. None of those rows were chosen to fill the day. They were chosen by constraints somebody wrote down before any day existed, which is most of what the rest of this page is about.

A map that works offline. Every place from the grid, pinned, coloured by day for a single city or by leg for a multi-city trip, each pin carrying its own time and tip so it still makes sense when you tap it in the street three days later.

And one file that opens with no signal. The whole trip (days, food, transport, the booking deadlines) as a single page you add to your home screen. The test it has to pass is turning the network off and reloading. That is not a purity exercise: underground, abroad, roaming disabled, standing outside a restaurant that may or may not be shut on Mondays is exactly when you need it, and exactly when a website fails you. Lived

What it does not produce is a decision. Everything above is a proposal you are free to ignore, which is the argument of the next section and the reason any of this is bearable on a holiday.

3. The plan is a set of options, not a schedule

The obvious objection to a holiday with times on it is that it sounds exhausting: a schedule, a calendar, the exact thing you went away to escape. That objection is right about the failure mode and wrong about what the document is.

None of this is a schedule, and there is no obligation to follow any of it. The times are not appointments; they are proof that things fit. The market is open when you would arrive, the drive is forty minutes and not two hours, the museum is not shut that day. Once a day is known to be feasible you can ignore its order, or the whole day. What the work buys is narrower and better than a plan: on the day, you never have to decide from scratch.

There are two bad trips and this sits between them.

• The over-planned trip, resented by Wednesday, where the schedule is a debt you are servicing and something good happening unexpectedly reads as an inconvenience.
• The unplanned trip, standing in a street at eleven, reading reviews you could have read three weeks ago, ending up at the nearest acceptable place because everyone is hungry.

The second is far more common, and almost nobody files it as a failure. It gets remembered as “we didn’t really do much.”

The way between them: research in advance, choose on the day. These have opposite requirements. Research does not benefit from time pressure in an unfamiliar city. Choosing does not benefit from happening three weeks early, when you do not yet know the weather, whether you slept, or whether somebody told you about somewhere better last night. So the scarce and expensive things get decided ahead, because they must be, and everything else gets researched ahead and chosen in the morning, from a shortlist that already has the reasoning attached to each option. “Highly rated” is not a reason, and you cannot confidently overrule a recommendation whose logic you cannot see.

Mood is what tunes this. Set per leg rather than per trip: the same city in the same week is a different trip depending on whether you came to decompress, to see things, to celebrate, or to live there a while. Mood does not change what is available (the researched options do not shrink), only how much is expected to be used. A decompression leg with three anchors on the grid has not been planned badly. It has been planned correctly, and then mostly declined. Lived

One finding runs against the obvious conclusion, and it is why “just leave it open” is not the answer: unstructured time on a holiday does not reliably become rest. It dissolves. A day with nothing on it turns into a slow morning, a long lunch and a vague afternoon nobody remembers. So even a deliberately lazy day gets a light rail (a walk loop, a swim, a café), not as a schedule, but so there is something to say yes to when nobody wants to invent an idea from nothing. Lived

That is the whole design goal: maximum clarity about what is possible, minimum pressure to do any of it.

4. Count the time before you plan any of it

This is the correction that matters most, and it came from the version that got it wrong.

Some trips are built around a place. Many are not. They are built around a fixed commitment, and the travel is what happens in the gaps. A conference. A wedding. A medical appointment. A family obligation. The planning instinct is to look at the dates, see five days, and start filling five days.

My system did exactly that, and would have produced a four-day city break for a trip that had one genuinely free day. Lived

So the first thing that happens now, before anything else, is a table. Every day, what is committed, what is actually left. And then two rules that sound obvious and are not:

Half-days are not days. A morning appointment plus a free afternoon is one usable half-day. It is not a day, and if you write it down as a day you will plan a day into it and then spend the trip running late.

Someone recovering from something is not free time at all. A companion who needs to rest is a hard subtraction from the trip, not a soft one.

Then say the real number out loud. “You have one free day, not two.” Writing this section, I am fairly sure that single sentence is the most valuable output the whole system produces, and it takes ten seconds.

There is a related habit worth stealing even if you ignore everything else here: find the immovable hour (the ceremony, the check-in, the appointment with a hard window), name it, and then walk the route to it the day before and time it. The classic failure around a fixed commitment is never the plan. It is traffic. Lived

5. Write your constraints as a list of what they rule out

“Take it easy” is not a constraint. It is a mood. It cannot be acted on, and, this is the part that surprised me, it will not stop a planner, human or otherwise, from proposing exactly the wrong thing, because nothing about the proposal visibly contradicts it.

If anyone on the trip has a physical restriction, the useful form is not a description of the restriction. It is an enumerated list of what it rules out, by category, written before any day plan exists.

The reason that works is not diligence. It is that the cases which actually catch you out are the ones where the entire point of an activity is the thing being ruled out, where the contraindication is not an incidental feature of the experience but the experience itself. Those never look like exceptions from the outside, and they are frequently the local speciality, because the local speciality is what you go and do. A restriction held as a sentence lets them through. The same restriction expanded into ruled-out categories catches them, because you are matching against categories instead of against a feeling.

Two practical notes, from having done this the wrong way round first. Lived

A restriction on one person often removes a category for everybody. It is easy to plan carefully around the affected traveller and still lose a whole class of activity for the group without noticing, because nobody wrote the class down.

Do it before you are attached to anything. Writing the list after the days exist means arguing with your own plan, and in that argument the plan generally wins.

The general form: do not record the restriction. Record its consequences, as a category list, in advance. It takes about five minutes, and it is the difference between a constraint you have respected and a constraint you discover at the counter.

6. Plan first. Build second. Do not skip the gap between them

My system produces two artifacts: one file you carry, and a map. Both of them are good. Neither is allowed to exist until the content is locked.

This sounds like process for its own sake. It is not. It is a straightforward observation about where the work actually is. Rendering a beautiful document around content that is still moving means re-rendering it on every change, and the effort you spend on formatting is effort you did not spend deciding whether the third day is any good. Worse, a rendered artifact looks finished, which makes you less likely to tear up the day that needs tearing up. Lived

So: plan in conversation, argue with it, change the order twice, and only then build.

The gate between the phases is a single word. I say “locked”, and not before. If you are doing this with an AI, tell it that explicitly at the start, because the default behaviour of every model I have used is to render you something impressive on the first turn.

The second half of this rule is about how the plan itself gets presented while it is still moving, which turns out to be a different problem, and it is the next section.

7. Options where it costs money. A recommendation everywhere else

Everything above says have options. This is the one section that pulls the other way, and the line between them is worth getting exactly right, because I had it drawn in the wrong place for several versions.

The distinction is not how many options. It is whether the decision locks you in.

Decisions that lock in: shortlist them, then close them. Hotels, the order you visit cities in, which operator, anything that takes a deposit or fixes the route. Here you do want two to four live options with the reasoning attached, chosen on the total picture rather than item by item, because a per-item choice made four times in a row produces a combination nobody would have picked. And then you want to stop. Past a point, more options here is not more freedom. It is just more searching. I once looked at twenty London hotels across seven separate searches, long after the answer had stopped changing, and nothing improved. Lived

So the system carries a running list of what is already closed, and a rule I need more than most people do: on the third return to a settled, reversible decision, name it as a lap and close it. Reassure; don’t re-option. And never re-open something already booked. That is pure downside, and it is a distinct habit from wanting choice.

Decisions that are free: keep every option, and add a recommendation. Breakfast, which viewpoint, what to do with a spare afternoon, which of the six cafés. Nothing at all is lost by carrying all six. What is lost is when all six arrive undifferentiated, because then the comparison work lands at nine in the morning, hungry, on a phone, in a street, which is the one place it is guaranteed to go badly.

The fix is not to delete five of them. It is to say which one I would pick and why, and leave the other five right there. A default is not a restriction. It is what makes it possible to stop thinking when you want to, and to swap when you don’t. The reasoning is the part that makes swapping safe: “closest to the morning stop, opens at seven, the other five open at nine” tells you in one line whether it applies to you today, and lets you overrule it with confidence rather than a shrug.

One recommendation per slot on the day page. Every alternative one tap away at the back. That is the entire rule, and it is close to the opposite of a menu. A menu asks you to decide; a recommendation with visible alternatives lets you decide only if you feel like it.

8. What you came for decides what actually gets researched

Mood sets how full the days are. Purpose sets what goes in them, and this is where most planning, human or otherwise, quietly defaults to a generic trip: the landmarks, a museum, some well-reviewed restaurants. If you came for something specific, that list is close to useless.

The rule underneath all of the below is one line: research the named interest, not the category. Searching “what’s on in the city” returns nothing you can use. Searching for the specific competition, the specific artist, the specific brand, on the exact dates, either returns a trip anchor or returns nothing, and both are useful answers. Lived

If it’s a shopping trip. More money moves here than anywhere else on a trip, and almost none of it gets planned.

• What’s worth buying is what is cheaper or unavailable at home, not what the place is famous for. Those are different lists, and from India the gap is often large enough to pay for a flight.
• Outlet versus city store is a real decision with a travel cost. Outlets are usually an hour out of town and eat half a day. Above a certain basket that is obviously worth it; below it, you have spent a holiday afternoon to save less than the taxi.
• Ask for the tax-refund form at the counter, at the moment of purchase. Not at the airport. The paperwork generally has to be issued with the sale and cannot be reconstructed afterwards. I have lost a refund exactly this way, and there was no recovery path once the invoice had been issued without it. Lived
• Trade discounts and tourist refunds are often mutually exclusive. Shops selling at trade prices are frequently not registered in the tourist-refund scheme at all, so the two are close to alternatives rather than additive. Ask which one applies before you buy. Lived
• Net the refund against what you’ll owe at home, and treat it as one number. The refund frequently exceeds the duty, which means customs is rarely a good reason to buy less, but declare properly. The risk is walking through the green channel with a large basket, not the duty itself.
• Shopping changes the trip’s order and its baggage. Shop last and fly home heavy; book the bags for the leg that carries them, not for the whole trip. Lived
• ⚠ Honest limit: the mechanics above I would defend anywhere. Which brands, which streets, what is authentic versus marked up: I have no vetted list, and anything I told you there would be the same search you would run. I’d rather say that than dress it up. Reasoned, not tested

If it’s a music, nightlife or events trip. The scarce thing is not the hotel.

• Look up the dates before you fix the dates. A gig, a match or a festival is the one input that should set the calendar rather than fit into it.
• Tickets are the tightest lead-time item on any trip, tighter than flights and much tighter than hotels, and they are the thing most likely to be gone while you are still deciding.
• Which night the city actually goes out varies enormously, and getting it wrong costs you the entire point of the trip. So do the entry norms, whether it is ticketed ahead or on the door, and, the one everybody skips, how you get home at three in the morning. That last one is what turns a good night into a bad memory.
• ⚠ This is the machinery applied, not a trip I have run. Weigh it accordingly. Reasoned, not tested

If it’s a specialist hobby trip (diving, climbing, skiing, sailing) the research is mostly gear, certification, insurance and medical clearance, and the highest-value step is not research at all. See the expert brief in §13, and the scope fence in §15.

If it’s mostly about eating, pick by category mapped to where you will already be, rather than by rating. A famous restaurant across town costs you an afternoon; the second-best one near your morning stop costs you nothing. And flag the dietary constraint against the local default, not against the restaurant list. See §9.

9. Who is going changes every number

Same city, same dates, same purpose, and a completely different plan depending on who is in the party. Most planning takes this as a headcount and then ignores it. It is closer to being the whole brief.

• Two adults, similar pace. The baseline, and the one everything in this page is calibrated to. Say so plainly rather than pretending otherwise. Lived
• With a small child. The binding constraints are nap and meal timing rather than distance, the day has materially fewer usable hours than it looks like it has, and every anchor needs a bail-out that does not feel like a failure when you use it.
• With elderly parents. This is where the standing axis in §11 stops being a refinement and becomes the plan. Distance between places matters less than whether there is somewhere to sit at each of them, how far the toilets are, and how many stairs sit between the taxi and the table. Heat and queueing are the two things that end a day early, and neither shows up in a step count.
• A mixed group. The honest arithmetic is that a group moves at the pace of its slowest member while everyone politely pretends otherwise until about day two. Planning for the average satisfies nobody. Plan a spine everyone can actually do, and hang the harder optional things off it at points where people can split and rejoin at a named time and place. Named, because “we’ll find each other later” is how half the group ends up doing neither thing.

Two mechanisms from earlier apply directly here and do most of the work.

Enumerate what each person’s limits rule out, by category, before any day exists (§5). One list per person, written cold. This is dull and it is the single highest-return fifteen minutes in the whole process once the party is larger than two.

Put rest on the grid as a named block, per person who needs one. Scheduled downtime works; improvised withdrawal does not, because unallocated free time gets absorbed by whoever is most enthusiastic that morning. Also treat group format itself as an effort cost: a shared-table dinner or a large group tour is more tiring than the same hours spent privately, and for some people it is much more tiring. Lived

And dietary constraints bite in a specific place, which is not restaurants. Someone travelling with me is vegetarian, and restaurants are almost never the problem; every city has options and they are easy to find in advance. The problem is the local default in the settings you cannot pre-book: street food, a festival, a wedding, a long transit day. Flag the one context where the default has nothing, and name what they can eat there, in advance. Lived

⚠ The honest gap, stated because this section will look more authoritative than it is: I have not run this with a child or with elderly parents. My own trips are two adults; our son usually stays with his grandparents while we travel. The framework above is the framework I would use and it is reasoned from the parts that are tested; it is not lived, and I would rather flag that than have you discover it on day two of your trip. If you have run it, I would genuinely like to hear where it breaks. Reasoned, not tested

10. Sometimes the hotel is the destination

This is the most recent thing I changed my mind about. For years my hotel logic was half right: you are out all day, so buy location and do not overpay for the property. By that rule a good day was one where I barely saw the room.

Then I spent a stay at the Okura in Bangkok mostly in the hotel, and it was one of the best parts of the trip. Lived The pool and the steam room, properly used rather than glanced at. A deep bath with salts, which is not a thing I have at home. A tea set in the room I actually made tea with, slowly, several times. Folding origami cranes (senbazuru, where a thousand of them carry a wish), which the property ran as a small evening ritual. Dinner ordered up rather than gone out for, which I would once have filed as a small defeat.

None of it was in any plan. All of it would have survived being planned, and some of it would have happened more if it had been.

So a hotel is now two different decisions, and the mistake was running one rule for both.

• A base. Out from nine until late, room is for sleeping. Buy location, quiet, and a bed that works; do not pay for a spa you will walk past twice a day. Most city legs.
• An anchor. The property is where part of the trip happens. You are buying the pool, the bath, the terrace, the thing in the room you would not otherwise do, and you will be inside it for hours, not minutes.

Decide which one each leg is before you book. A property chosen on location and then used as an anchor is the worst of both: you paid for a location you did not use and got a room with nothing in it.

Put it on the grid or it will not happen. “Pool and sauna, 16:00, before dinner” survives contact with the day. “We’ll use the spa at some point” does not, and you fly home having used it once.

Check what is actually there before you arrive. The best parts are often unadvertised and free, and pool and spa hours are much shorter than you assume. Finding that out at nine in the evening is how a planned evening becomes nothing.

11. Grade the day on two axes, because one of them lies

Most planning grades a day on effort: how far you walk, how many steps, how much elevation. That is the axis everyone tracks, and it is not the one that hurts you.

The system now carries two independent grades on every day:

• effort: light / medium / hard. Walking, steps, climbing.
• standing: low / high. Time spent locked in place, in heat or in a crowd, unable to sit down.

They come apart completely.

A festival night involves almost no walking and is the highest-risk night of the trip: you are standing still, in heat, in a crowd, probably drinking, finishing late, and under-hydrated. It grades light on effort. The day I have the most reason to remember graded light on effort. Lived

So the check that runs on any big day is a stack count rather than a grade: heat, prolonged standing, crowd density, alcohol, a late finish, low hydration. Three or more of those in one day and the day gets treated as a hard day, whatever the walking says. In practice that means a named place to sit down with a route to it, pinned on the map before you go, and everyone in the party knowing what the early warning signs look like so it isn’t only you watching for them.

Two smaller rules fall out of the same logic and are worth having:

Don’t stack two hard days. Don’t stack two high-standing days either. These are separate constraints and a plan can easily satisfy the first while violating the second.

Nights stack too. A hard day tomorrow caps dinner tonight; a late night forces a late start. One big night is fine. Back-to-back short nights compound, and by day four you are not enjoying the trip you paid for.

12. The artifact: one file, and it has to work in aeroplane mode

For a long time the output was a PDF. Then several PDFs. Then several PDFs re-rendered every time anything changed, which is how you discover that your planning system’s real bottleneck is document generation. Lived

It is now exactly one self-contained HTML file. Not a folder. Not a link. One file, opened in a browser and added to the home screen, where it behaves like an offline app for the trip.

The constraint that makes it work is worth stating precisely, because it is the one everything else follows from: it must render completely with the phone in aeroplane mode. Everything inline: the styling, any scripts, system fonts only, no fonts pulled from a CDN, no images unless they are embedded in the file itself.

This is not a purity exercise. A guide that needs signal is a guide that fails in exactly the circumstances you built it for: underground, abroad, on a dying battery, roaming disabled, standing outside a restaurant that may or may not be closed on Mondays. Test it by turning the network off and reloading. If it still works, you have a guide. If it doesn’t, you have a website. Lived

Three content rules that turned out to matter more than the format:

Every itinerary line is a timed row: one time, one place, one tip of about eight words. If a day reads as a paragraph, it has failed. You are going to consult this while walking.

The tip carries the reason, not an adjective. “How the lock lifts a 20m boat” is a better eight words than “beautiful canal views”. It tells you what you are going to look at and why you would bother.

Everything is tappable. Every place name opens in maps, every app name opens the app. The friction you are removing is not typing; it is the decision to bother.

And a navigation index at the top that jumps to each section, because the file is long and you are reading it one-handed.

13. Send it to somebody who has actually done the thing

This is the highest-return item on the page, and it is the one I resisted longest, because it feels like admitting the research was insufficient. It was insufficient. That is the finding.

Before locking anything on a trip with a specialist component (diving, climbing, ski touring, sailing, anything with gear and consequences) write a brief and send it to a domain expert. Not a chat summary. A document they can read cold: who is travelling, the hard constraints, the shortlists with your reasoning attached, and a numbered list of questions.

I did this for a dive trip. The reply contained seven corrections. Six of them were things desk research had got wrong: a layering system specified backwards, a purchase order that should have been reversed, two “required” items that were merely optional, weight arithmetic that was inflated, an entire product category I had not considered, and a cheaper way to source locally that I would never have found. Lived

Ninety minutes of an experienced person’s attention beat several days of mine, and it was not close.

Two rules for handling the reply, both learned by getting them wrong:

Put the constraint that eliminates options at the top of the brief, so they don’t spend their answer on things that were never available to you.

Read what they actually said, not the momentum of it. A reply that corrects six things and endorses the seventh has not rejected all seven. In that same exchange I read a clear “why not, if you want to” as a hedge and demoted an option the coach had just approved. A correction-heavy reply creates a pull toward over-correcting, and you have to consciously resist it.

And one caveat on the advice itself: an expert’s advice is calibrated to their own profile. Someone who does this four times a year can improvise around a missing part. Someone who does it once a year is better off carrying their own. Take the correction, then check it against how often you actually do this.

The single most valuable question, and it goes last: “Anything obviously missing, or anything you’d drop?” An experienced person finds a gap in ninety seconds that research cannot surface at all, because they are not searching. They are remembering.

14. The small operational things that keep paying

None of these are clever. All of them have saved a real day.

Find out which app local businesses actually answer on. This is not the same question as which maps app to use. In Thailand it is LINE. Over ninety percent of business messages get opened there, and WhatsApp exists mostly at tourist-facing places because inbound tourism put it there. Roughly: LINE in Thailand, Japan and Taiwan; WeChat in China; KakaoTalk in Korea; WhatsApp across India, the Middle East, Brazil and most of Europe; Facebook Messenger as the near-universal fallback for anywhere with a page. Install the local one before you fly. Two minutes, and it is the difference between a held item and a wasted trip across town. Researched

Cluster errands by postcode before you schedule any of them. Obvious, universally skipped. Eight separate errands on one trip collapsed into three outings once I pulled the postcode for each and found two of the key shops shared one, and one was walkable from the hotel. That changed which day the shopping happened on, which changed the shape of the week. Lived

Convert every “book three months ahead” into an actual date. A relative interval asks you to do arithmetic at precisely the moment you are least likely to: the booking step, which for me is where everything stalls. Write book by 14 March. It is a small thing and it is the difference between a decision and a task.

Mark every price as verified or estimated, and make it visible. A planning number that looks like a quote gets trusted like one. One line I carried as an estimate came in thirteen percent higher once it was itemised at the till. Marked, that is a correction. Unmarked, it is a surprise, and surprises at the till are how budgets stop being believed. Lived

Separate the fitting from the purchase. For anything fit-critical (wetsuits, boots, packs, harnesses, tailoring) the fitting is the scarce thing and the transaction is not. Go and try things on, write down what fits, then buy wherever is cheapest, which may well be online from home three weeks later. Never buy a fit-critical item because it happened to be the one in stock. Badly-fitting kit is worse than the right kit arriving late. Lived

15. What the system deliberately refuses to do

A system that plans everything plans nothing well, and two of these fences were added after the system overreached.

It does not plan expedition trips. A dive liveaboard has no city geography, no transit pass, no restaurants sorted by cuisine and no useful map. A dive day is four dives, not a timed list of places. Its real deliverables are a completely different document: certification, insurance, medical clearance, a packing list. Forcing a city-guide system onto a boat produces a city guide for a boat. Those trips get their own process, and a trip that is half boat and half city gets both. Lived

Nothing in the recovery section is measured. There is a per-trip note about a walking loop near the hotel, where the nearest pharmacy is, and how to handle a time-zone shift on a dosing schedule. None of it is tracked, scored, or turned into a streak. This is a deliberate fence: the failure mode it exists to prevent is a holiday quietly becoming a metrics project, which is a thing I would absolutely do if nothing stopped me. The trip is a deload by default. Lived

It does not pretend the plan is the trip. Every day ends with an optional unhurried walk from the hotel with no phone and no podcast, the one block on the grid whose entire purpose is that nothing is scheduled in it.

16. If you want to run this yourself

The system itself is a plain markdown file: a few thousand words of standing instructions, a traveller profile, and a paste-in prompt. It is not software. It runs in whatever AI assistant you already use, and the reason it works is not the model, it is that the accumulated corrections live outside the conversation and get loaded every time.

That is the actual transferable idea on this page, and it is worth separating from everything above: the value is not in my rules, most of which are calibrated to two specific people. The value is in having a file at all, one that gets edited every time a trip teaches you something, so that the ninth trip does not repeat the second trip’s mistake.

Start it after your next trip, while you still remember what went wrong. One page. Three headings: what worked, what I’d never do again, what I wish I’d booked earlier.

So here is mine, cleaned up and made general: Carmen v10. One markdown file, free, and yours to edit. Paste it into whatever assistant you already use, work through the interview in §3 once, then hand it a trip brief when you need one.

There are two ways to make it yours, and both are here. The better one is to paste the file into your assistant and tell it to run the interview. It asks follow-ups, and it can tell that “take it easy” is not a constraint, which is the difference between a profile and a paragraph. The faster one is a form: two minutes, no conversation, answers written straight into the file along with the instructions each one triggers.

I would use the interview. I have also watched enough people never get round to a twenty-minute anything to think the form should exist. A filled-in file somebody actually made beats a better one they didn’t.

It carries everything on this page as standing instructions, with two differences from what I run privately. The personal half is replaced by an interview, so it builds your profile instead of assuming mine. And every health specific is stripped out, because that part should never be templated from somebody else’s document. It goes stale faster than anything else in a travel file, which is a mistake I have made and corrected more than once.

It is version ten because the first nine were private, and each one got rewritten after a trip went wrong in a particular way. Version eleven should come from somebody else’s trip. If you run it and it breaks somewhere I have not listed, tell me where. §13 of that file is the whole engine, and it works the same way for you as it does for me.

17. What I’d still get wrong

Four honest limits, because a system page that claims no failures is selling something.

The mood axis carries more weight than its evidence does. It is doing real structural work in §3: it is the thing that decides how full a day gets, and therefore how much pressure the plan applies. It is also the input I can least defend: four labels, useful across my own trips, and no evidence beyond that. I have deliberately not softened its role to match its evidence, because in practice it is the most useful single question I ask. But you should know those two facts sit next to each other. Reasoned, not tested

The party-composition section is a framework, not a report. §9 is the most confident-sounding thing here and the least tested. Two adults is what I have actually run. Everything about children, elderly parents and mixed groups is derived from mechanisms that are tested, which is not the same as having done it.

Which things to buy where rests on nothing. The shopping mechanics in §8 (refund paperwork, trade-versus-tourist schemes, order and baggage) are lived and I would defend them. Brand and street-level picks are the same search you would run, and I have no vetted list behind them.

Nine versions is not a lot of trips. Every rule here comes from a small number of holidays taken by one couple from Mumbai. The obligation audit, the restriction list, the artifact constraints and the expert brief I would defend anywhere. The rest is a working draft that happens to be written down, and the parts marked as reasoned rather than lived are marked that way on purpose.

18. The open door

If you are planning something and want a second pair of eyes on the shape of it (not the restaurants, the shape, the thing that decides whether the trip works) write to me. That is the part I enjoy, and it is the part the models are still worst at.

And if you run any of this and it fails somewhere I haven’t listed, I would like to know, because that is how the file got to version nine.

What’s changed

• 2026-08-25. Added §2, which shows what the system actually produces (a real timed day-grid, the map and the offline file) before asking anyone to read six thousand words about why. Added a diagram to §11, because effort-versus-standing is the one idea here that is easier to see than to read. Cut roughly a fifth of the prose out of §3 and §10 without losing anything they were carrying.
• 2026-08-21. First published.