See the connections between your notes.

Every August the garden behind Ines's house on Alder Lane smelled like tomato leaves and hot brick. She grew the same six varieties for forty years, and she never bought seed. At the end of each summer she let the best fruit overripen on the windowsill, scooped the seeds onto paper towels, and wrote the variety and the year in pencil. The envelopes lived in a biscuit tin in the pantry.

What I didn't understand as a kid: she wasn't just saving seeds, she was saving the garden. If a blight took one variety, the others carried on. If a summer was bad, the tin still held three or four years of backup. The garden was never in one place. It was spread across the beds, the tin, and the neighbours she gave seedlings to.

She used to say the compost heap was the most important bed in the garden, because everything else came out of it.

I still have two of the envelopes. "Coração de Boi, 1994." I have no idea if the seeds would still germinate. I haven't dared to try.

Looked up the old address today. The house on Alder Lane was demolished and the lot is now four townhouses with gravel front yards. The pear tree is gone. The garden is gone. An old Street View capture still shows the corner of the greenhouse, blurry, behind a parked van.

It is strange how a place can vanish while the memory stays so detailed: the loose paving stone by the gate, the sound the screen door made. But the memory is also getting simpler each time I recall it. I'm not remembering the house anymore; I'm remembering the last time I remembered it.

Places change whether we record them or not. Photographs freeze a moment, but they don't hold the layout, the smell, the route from the kitchen to the beds. I want to sketch the floor plan from memory before that goes too.

Question for later: is a memory that has been retold a hundred times more true or less true than one I haven't touched in years?

Found the mixtape Dad made for the drive to the coast in 1998. Ninety minutes, his handwriting on the insert: "Summer Roads, side A for going, side B for coming back." I don't own anything that plays it anymore.

I remember almost every song in order, which is ridiculous, because I can't remember what I did last Tuesday. Music seems to store memory differently, not as facts but as sequences. Hear the first bar of one track and the next one starts playing in my head before I've decided to think of it.

That's the thing about a mixtape: the order is the message. Anyone could list those songs. Only he would have put them in that order, with that gap of silence before the last one.

Nostalgia gets a bad reputation, but I think part of it is just noticing that someone once paid close attention to you.

Same feeling as Alder Lane doesn't exist anymore: the thing itself is gone, but the sequence survives.

Mum's shoebox arrived: roughly six hundred prints, most unlabelled. Most of them were taken by Dad, on a camera he still talks about. I spent the weekend scanning. About a third of the faces nobody in the family can name anymore. Someone knew every one of these people well enough to photograph them at a birthday. One print shows a corner shop I eventually found on Street View, repainted but unmistakable.

The prints survived forty years in a cupboard. My own photos from 2009 are on a hard drive I can't open. Paper is a surprisingly good format; formats that need a machine to read them have a much shorter memory. There is a name for this worry, the digital dark age: en.wikipedia.org/wiki/Digital_dark_age

What makes a photograph useful later isn't resolution, it's the caption. A picture with a name and a year written on the back is a memory. The same picture without it is just a pleasant stranger.

Plan: scan everything, write down what my aunt remembers while she still remembers it, and keep copies in more than one place.

Ines made dumplings every Sunday and never wrote the recipe down. When she died, we realised nobody knew the ratio. My aunt and I spent three Sundays getting it wrong before it tasted right, and we still argue about the pepper.

What survived wasn't a document. It was the practice: the flour on the counter, who folds and who boils, the rule that the first batch is always for the kids. The recipe lived in her hands and in our habit of showing up.

Now we do it on the first Sunday of every month and my niece folds them badly and proudly. I wrote the recipe down this time, with my aunt's corrections in the margin, but I've started to think the page is the least important part.

Traditions are kept alive by being repeated, not by being recorded. The record helps you restart. It doesn't keep anything alive on its own.

See also My grandmother's tomato garden.

Someone asked why I have eleven years of notes I never reread. Fair question.

Part of it is fear: I've lost things before. A dead laptop took two years of journals with it, and I still feel the gap. But mostly I don't know in advance which thoughts will matter. The note I dismissed as trivial was the one I needed during the move.

Keeping isn't the same as hoarding. Hoarding is keeping without returning. An archive is only valuable if something flows back out of it. Andy Clark and David Chalmers argued that a notebook can be part of your mind if you rely on it the way you rely on memory. By that standard most of my archive isn't part of my mind at all. It's a storage unit I pay for with guilt.

Reading about the digital dark age (en.wikipedia.org/wiki/Digital_dark_age) made me realise format matters too. Plain text will outlive every app I use.

Goal: keep everything, but build the habit of going back.

Fourteen drafts of the same novel. None finished. The folder is called "lighthouse" after the working title and I open it about twice a year, feel bad, and close it.

Looking at it honestly today: every draft tried to build the whole book at once. Full outline, a cast of twelve, three timelines. Each one collapsed under its own structure around chapter four. It's Gall's law, basically: a complex system that works grew out of a simple system that worked, and I keep trying to skip the simple one.

But the drafts aren't wasted. Reading them back, the good scenes are all small: the keeper mending a net, the argument about the fog signal. They're compost. The next attempt should start from one of those scenes and let it grow.

Related: Notes are compost, not storage

A note that is only stored is dead weight. A note that is revisited breaks down into something the next idea can grow in.

The compost heap is the right model. You throw in scraps you can't use yet: a quote, a half-formed question, the good scene from a failed draft. You turn it now and then. Given time and a bit of heat, it becomes soil. Not the original material, but the thing that makes new growth possible.

This changes what a good note is for. It isn't a finished record; it's material for later thinking. Andy Matuschak calls the durable version an evergreen note: written to be returned to, refined and linked, so it accumulates value instead of decaying.

Practical rules I'm trying:

• Write notes for my future self, not for filing.
• Turn the heap: revisit a few old notes every week.
• Let unrelated scraps sit next to each other. Heat comes from mixing.

See Feedback loops in the compost bin for why mixing matters.

Spent the afternoon drafting the essay on attention with an AI model as a sparring partner. Useful in ways I didn't expect. It was good at structure (it spotted that my third section was really my introduction) and bad at voice. Every sentence it offered was smooth and slightly nobody's.

The trick was to treat it like a collaborator who has read everything and remembered nothing about me. When I pasted in my own old notes it suddenly got sharper, because now it was arguing with my thinking instead of the average of everyone's.

Typing on his writing ball, Nietzsche said our writing tools are also working on our thoughts. That seems even more true of a tool that talks back.

Rule for myself: it can propose, I decide. Nothing goes in the essay that I wouldn't defend in my own words, and the final draft is typed by me.

Andy Clark and David Chalmers' extended mind paper uses the example of Otto, who has memory loss and relies on a notebook to find his way to a museum. They argue the notebook is part of his mind because he trusts it, consults it automatically and keeps it with him.

What I notice: Otto's notebook works because he reads it constantly. External memory isn't a place you put things; it's a conversation you keep having with your past self. The moment you stop consulting it, it stops being memory and becomes an archive.

This is where software could help more than it does. A good tool wouldn't just store what I wrote. It would bring things back: the right old note at the right moment, the one I'd have forgotten to look for.

Otto didn't need a better notebook. He needed a notebook that remembered to open itself.

Late in life, as his eyesight failed, Nietzsche switched to a Malling-Hansen writing ball, and a friend noticed his prose got terser. The reply from Nietzsche was that our writing tools are also working on our thoughts.

I see it in myself. In a word processor I write long, careful paragraphs. On my phone I write fragments. In an outliner I think in hierarchies, and ideas that don't fit a hierarchy quietly disappear. Folders ask "where does this go?" when the better question is usually "what does this remind me of?"

A tool that shows relationships instead of locations would change what I notice. Not because it's smarter than me, but because it would stop asking me to decide where a thought lives before I understand it. That is the notebook in External memory is a conversation, opening itself.

The medium isn't neutral. Choose the one whose shape matches how you want to think.

For years I thought my notes needed a better filing system. They didn't. They needed attention, the way unfinished projects do.

Organizing is a one-time act that feels productive. Tending is small and repeated: open three old notes, fix a sentence, link one to something newer, promote the one that has become an idea. Andy Matuschak's point about evergreen notes is really a point about maintenance.

A garden isn't organized once. It's weeded, watered, pruned and occasionally left alone. Creative projects are the same. What survives is whatever I keep returning to for ten minutes at a time, not whatever I planned most carefully.

Ten minutes a morning. Three notes. One new link. That's the whole system.

Builds on Notes are compost, not storage.

Noticed a pattern: if I don't write down a question within about a day, I lose interest in it. Not the answer. The question itself. The itch fades.

Curiosity seems to decay on a curve. Right after reading something surprising I have ten follow-up questions. A week later I have one. A month later I can't remember why it mattered.

So the job of a capture is not to answer anything. It's to preserve the question at its original temperature, so that when I come back I can feel the itch again. A quick note that just says "why do starlings flock in those shapes?" is enough. (Years later that one grew into Patterns: the same shape everywhere.)

The questions I did write down years ago are strangely alive when I reread them. The curiosity was stored, not lost.

Last week I reread a note from 2019 about my grandmother's seed tin (My grandmother's tomato garden) and realised it was the answer to a problem I'd been stuck on for a month: how to keep a long creative project alive. Redundancy, small batches, give copies away.

I hadn't looked at that note in years. It was useful precisely because I had forgotten it. It came back without the baggage of the context I wrote it in, so I could read it as if someone else had.

Forgetting isn't the enemy of a second brain. It's part of the cycle. Ideas go dormant, and when they come back into contact with newer thinking they can combine in ways they couldn't at the time.

It is the lesson of Otto and his notebook, turned around: the notebook only counts as memory if you keep reading it.

The failure mode is not forgetting. It's never coming back. That's the part a system has to solve.

Related: Curiosity has a half-life

Killed my sourdough starter after eight months. Diagnosis in hindsight: I "optimised" the feeding schedule to once a week because it seemed fine, and it was fine, right up until it wasn't. The yeast had been losing ground to the bacteria for weeks and the signals were there: slower rise, sharper smell. I just wasn't paying attention to them.

Lessons:

• A system that looks stable may be slowly drifting. Stability is not the same as health.
• The feedback was available; I had stopped reading it.
• Small regular feeding beats heroic rescue.

I made a new starter from a friend's and I feed it every other day now. I also keep a spoonful dried in the freezer as a backup, which is exactly what my grandmother did with seeds. (See Resilience is redundancy.)

Afterwards I reread Meadows on feedback delays. The signal was there, just late, and a delayed signal is the easiest one to ignore.

Mistakes teach best when you write down what the signal was, not just what went wrong.

Reading about mycorrhizal networks: the fungal threads that link tree roots and move water, sugar and chemical signals between them. Suzanne Simard's research on this sits underneath a lot of The Overstory by Richard Powers.

The analogy writers reach for is "the wood wide web", which is cute but backwards. The internet is the newcomer. Networks that share resources and signals between nodes were solved underground long before we built cables.

I love this kind of cross-discipline rhyme. Ecology explains something about networks; networks explain something about how ideas travel between my own notes. The most useful connections are rarely inside one field. They're between two fields that don't read each other's journals.

Keep a running list of these rhymes, like Patterns: the same shape everywhere. They're where my best ideas have come from.

See Reading log: The Overstory.

Finished The Overstory by Richard Powers. Nine people whose lives are gradually pulled together by trees, told over decades. It's slow in the way a forest is slow, and I think that's the point.

The idea that stayed with me: trees in a forest are less a set of competing individuals than a community that shares through the soil. Older trees pass resources to their neighbours. Suzanne Simard's research sits under all of it.

It made me look at Balcony ecosystem, year two differently. The pots are separate, so none of that underground sharing can happen. Maybe that's why the whole thing feels so fragile.

Rating: five out of five, but give it a hundred pages before deciding.

Year two of the balcony garden. Last spring the aphids destroyed the chillies within a week. This year I did nothing clever. I planted marigolds and dill and waited. By June the ladybirds had arrived on their own and the aphids never got past a few colonies.

The lesson is that I didn't solve the problem; the ecosystem did, once there was enough of it. One plant is a target. Fifteen kinds of plant is a system with its own checks.

Inventory: chillies, tomatoes grown from a friend's seeds, dill, marigolds, basil, a fig that shouldn't survive up here but does, strawberries, and mint in its own pot for everyone's safety.

Things to try next year: a small worm bin for compost, and saving seeds from the best tomato the way my grandmother did: fruit left to overripen on the windowsill, seeds dried on paper towels, envelopes labelled in pencil.

Resilient systems rarely rely on one strong part. They rely on many overlapping parts, so that when one fails another carries on. A monoculture field is efficient and fragile; a meadow is messy and hard to kill. Even Darwin noted that a plot sown with several kinds of grass yields more than one sown with a single kind.

You can see it in a lot of places:

• A garden with diverse plants weathers a pest better than a single crop (Balcony ecosystem, year two).
• A seed library spread across several envelopes and several neighbours survives a bad year.
• Memories kept in more than one place (prints, scans, a relative who remembers) outlast any single copy.
• A creative project with several small working pieces survives losing one of them.

Redundancy looks like waste when things are going well. It's what you're grateful for when they aren't.

The design question is always: where is my single point of failure?

The compost finally got hot. Stuck a thermometer in and hit 58°C in the middle of the pile. The trick was mixing: greens and browns together, turned every few days.

It's a lovely feedback loop. Microbes break down material and release heat; heat speeds up the microbes; more activity releases more heat, until the food runs low and the pile cools. Turning it brings fresh material into the hot centre and the loop starts again.

It's the diagram from Meadows' systems primer made physical. Positive feedback builds; negative feedback stabilises. A pile that's too dry never starts. A pile that's too wet goes anaerobic and smells. The whole art is keeping the loop in the range where it feeds itself.

This is the physical version of an idea I keep returning to: mixing unrelated material creates the heat.

Branching shows up everywhere once you look: rivers, trees, lungs, lightning, the veins in a leaf, the citation graph of a good paper. The same shape solves the same problem: how to reach a lot of space from a single source cheaply.

Patterns like this are why cross-discipline reading pays off. If you learn how a river network balances flow, you've learned something about roads, blood vessels and org charts.

Even the flocking of starlings has a shape: thousands of birds each following a few neighbours, and a pattern no single bird intends. The tangled bank Darwin describes at the end of the Origin is the same idea in a hedgerow.

Other recurring shapes worth collecting: spirals (growth that keeps its proportions), cycles (compost, seasons, sleep), and networks with a few hubs and many small nodes.

My notes are starting to look like the last one.

Gall's law: a complex system that works is invariably found to have evolved from a simple system that worked. A complex system designed from scratch never works and cannot be patched up to make it work.

I keep rediscovering this in unrelated places:

• The balcony garden only worked once I stopped planning it and let it grow a plant at a time (Balcony ecosystem, year two).
• Every note-taking system I designed upfront collapsed. The one that stuck started as a single text file.
• Creative projects: the pieces that survive are small and complete, then grow.

This is Gall's law in practice. Start with the smallest version that actually works. Then let it grow, one working piece at a time.