We show the work. Including the work in progress.
This page is the production line behind the advice: what we're building, what's been requested, what we're reading now, what just got graded, and what shipped. The same honesty we apply to the evidence, applied to how the evidence gets made.
Building next
Forward-looking. Intent, not commitment.Direction, not promises. This is what we are building and the order we are building it in. Dates move and some of it will change. When a thing ships it drops into What shipped below, with the detail. Martin-edited.
The authoring queue
Illustrative pre-launch pipeline. The live queue fills from your suggestions.Every entry moves left to right. Nothing reaches the library until it's graded and receipted, and you can watch it get there.
Tell us what to grade next
Unsure about a supplement, a protocol, a claim you keep seeing? Send it. Requests shape the queue, and tell us where the confusion is thickest.
What shipped
Knowledge base, stories, product. New entries, regrades, and what changed our mind.
When a log fell outside what the app accepts, it used to say it could not save and leave you guessing. Now it names the answer that is off and the range it will take, so you can fix it in one go.
Each pillar page now shows its score out of what is open to you right now, and explains line by line where the number comes from. It keeps a record of what each log moved, and you choose which cards sit on the page.
From Your practices on your day you can change what a practice is called or how often you do it, and stop it with an undo. A stopped practice from the app's own list can be picked up again later, and setting a suggestion aside can be taken back too. Change how often you said you do something, and the app stops measuring you against the old answer.
A morning energising breath practice now climbs in four steps, the first three with a counter to follow along. Every step stays held back until you have answered a short health question, because for some people fast, forceful breathing is the wrong tool.
When your logs show a pattern worth your attention, like several nights of ten hours or more in a week, or a fortnight of low mood ratings, the check-in raises it as something you can answer or mute, comes back to it at most once, and tells you when the pattern has cleared. It says what your logs show and asks you about it. It does not diagnose anything.
Getting started now opens on a few of the app's own logs, laid over a picture of your day. What you answer becomes your real day, and before you make your account it shows where you stand on the spine. A short tour of the real pages follows, and you can replay it from Settings.
A practice on your day can now be started right where it sits: a breath pacer for box, resonance and alternate-nostril breathing, a journal page that keeps your words if the app closes, a stillness timer, a timer for the cold practices that keeps its safety note on screen, and guided rounds for balance work with a stop button always in reach.
The daily check-in now rests as a sheet over your day instead of a screen of its own. Sleep, meal and movement logs use the same form wherever you open them, and a log you mention in chat can have its missing details filled in right there.
Before chat could draw on them, three of the sweetener entries were attacked by a review whose only job was to find where we were wrong. The sucralose entry had called its insulin finding unreplicated when two other trials had in fact found the same thing, and it now says so. The sweeteners overview now marks water as where we point you by preference rather than by trial finding, because a year-long trial against plain water found sweetened drinks slightly better for keeping weight off.
Read the entry →About thirty new entries arrived, on sweeteners, hormesis, methylene blue, biphasic sleep, neck training, spicy food and more. Every one of them was graded, with its sources, before chat could draw on it. They reach the library with this update.
Read the entry →A new Day page lays your day out as blocks around a ring. Tick a block or mark it missed, drag blocks into a new order, get a reminder at the time you set, and let chat suggest changes that only land when you confirm them.
Stepping up a level, crossing into a new region of the spine, or being placed for the first time now gets a full screen of its own, saying what changed and how your suggestions shift from here.
We reread the prescription stack, the optimiser and the 4am exec against the knowledge base and corrected what overreached. Each ending is now stated as a bet, and the 4am exec now says why his steps run in the order he can manage rather than by strength of evidence.
When a chat works something out, it can offer to save it as a plan tied to where you are now. Nothing is saved until you say so. The plan lasts fourteen days, then asks you to look at it again.
Each pillar score is now built from the practices you keep. A practice you let lapse fades gently over days and settles rather than vanishing, and adding new practices to the app cannot shift your number by accident.
A few safety questions can no longer be skipped. Until they are answered, the practices they guard stay held back, and where a practice does not suit what you have told us, you are offered a safer version or pointed to the right kind of help.
Swipe the Spine page and it turns into a read-up: how much of your picture the system has so far, and the practices ahead with what each one is for. It opens from the Library too.
The library gained a genetics section of 115 entries on gene variants, carrier status and the pathways they sit on, and what each one is known to affect. Eight new entries joined the knowledge base chat draws on, including statins, menopausal hormone therapy and psyllium, and reach the library with this update. Uploading your own genetic data, which went in back in March, is switched off for now.
Read the entry →We found and fixed a set of quiet wearable bugs. Sleep that ran across midnight was being counted against the wrong day, and two different sleep-stage formats were being averaged into a number that meant nothing. It is one night, one window, one format now.
A deletion request was stopping part way through and leaving fragments behind. It now runs to the end, and the copy of your data you can ask for is complete and bounded. Privacy is not a line we write, it is a path we test.
The daily surface is now a single calm instrument. It opens with what the system noticed overnight, stands the one thing worth doing in the light, and asks only what it needs to. On a day that calls for rest, it says so and gets out of the way.
A measurement-first sweep of the parts that felt slow. Signing in no longer storms the server, the morning snapshot is fetched once and shared rather than fetched three times over, and the surfaces that stuttered now settle. Faster because it does less, not because it skips the work.
When a goal has done its work, the system retires it instead of asking you to keep proving something you have already fixed. The moment is marked, and it shows you what opens up next rather than leaving a blank where the habit used to be.
The library roughly doubled. New conditions, supplements, and diagnostics across all four pillars, plus a category and cross-reference layer so each entry links to what it touches. Every one still carries its confidence tier.
Read the entry →Three patterns went up: the prescription stack, the optimiser, the 4am exec. Each one shows its loop, the order it comes apart in, and a graded entry behind every step. They stay marked as composites until a real tester walks one.
Four hubs, one per pillar. Each shows what that pillar does to the other three. No pillar stands alone.
The daily check-in now opens with what the system noticed overnight. One line, the most relevant signal, confirm or dismiss. Proactive without the noise.
Every entry carries its confidence tier. Tiers run Foundational through Experimental.
Read the entry →Added at Foundational. Attentional clearing as substrate work, and the receipt behind the optimiser story.
Read the entry →Every story on the site is a marked composite: mechanism and path, no outcome claims. When a tester walks a pattern with logged data showing the predicted resolution, the composite retires and the real story takes its place. That change shows up here.
Proactive nudges, filtered through your preferences. Every category can be switched off. The system earns the right to interrupt, or it stays quiet.
Twenty-one pattern detectors, each held to one rule: it only fires when the data is there. No calling a trend from three days and a hunch.
The architecture the whole system runs on. 33 vertebrae across five regions, each one a state with its prerequisites, grounded in Reflex Congestion Theory. It decides what comes next.
The first conversation places you on the spine from what you say. Back-to-baseline framing, calculated targets, no generic intake form.
Upload raw data from any major provider. It gets parsed once, then the file is deleted. After that, chat could read what your genome says about how your body responds, held to the same evidence bar as everything else. Upload was later switched off, on 11 July.
Health Connect integration: real sleep stages, overnight HRV, resting heart rate, step rhythm. The system reads what your body did, not what you remember.
The AI never freelances. Every answer is assembled from graded knowledge-base entries, tier by tier. It interprets the evidence. It does not invent it.
Every logged meal, workout and night passes through one validation gate: range checks, cross-field consistency, duplicate detection. No silent bad data reaches the scores.
Tell the app what you already do and it tracks it, while staying honest about confidence. A self-reported habit stays marked that way until logged data or time confirms it.
The first version of the morning read that connects the pillars. It weighs sleep debt against training load against stress, then surfaces the one thing worth doing, with the reasoning.
The onboarding that reframes the whole thing: back to baseline first, then above it. Goals carry a lifecycle. They adapt when you stall, and graduate when you no longer need them.
One thesis, fixed in the first commit: restoration, not optimisation. Everything since has been building toward it.
The changelog, with the reasoning left in.
A letter when there is something worth your time: one library entry read properly, what changed in the app, and what changed our mind. Sent from Realised.