Today I put fourteen months of my blood tests, three CT scans, and the bloods from yesterday on a public webpage. Anyone can read them. AI agents can scrape them. The name is on them. The DOB. The NHS number. The hospital number. The reporting radiologist. I did not redact anything.

Some of you are wincing already. Good.

Here's why.

I get one or two lines

The way it normally works is this. Someone draws blood at half nine in the morning. A lab in London runs the panel. A consultant gets a copy. Six weeks later I sit in a room for ten minutes and I get the headline.

That's it. The two-sentence version of a forty-page document.

The full panel sits in a PDF that no one opens unless they have to. Every marker. Every reference range. Every flag. Every comparator. The shape of the trend. The bit where the lab note says "please note new method effective 26/09/2022 and reference range change effective 08.04.24". The bit where bilirubin doubled overnight and nobody wrote it down because the doctor was running fifteen minutes late.

I open them. I read every value. I run them through three different LLMs. I compare. I notice things. Sometimes my consultant agrees those things matter. Sometimes she doesn't. Either way I have the data.

Most patients don't. Most patients can't. Either the data never reaches them, or it does and it looks like a phone book in monospace and they don't know what they're looking at.

AI can't help if it has never seen one

Here is the thing that quietly drives me mad.

The big consumer AIs are getting absurdly good at medicine. I have watched them spot trends in my own bloods that my private oncologist missed. I have watched them explain a scan report in plain English, then suggest the three questions I should ask my MDT, and have been right on all three. I am not naive about hallucinations. I'm also not naive about how useful they are.

But they're working with training data that mostly doesn't contain real patient histories. It contains journal abstracts. Textbooks. The tidied-up version of a case study that someone wrote up six years after the fact, in the past tense, with all the messy bits sanded off.

It does not contain the fortnightly, contradicting-itself, what-the-hell-is-that-spike version of a real treatment course.

It does not contain the CRP that goes to 87 the same day a scan confirms progression.

It does not contain the bilirubin that ticks up to 39 for one draw and then comes back down, with nothing else changing.

It does not contain the white cell count that goes from 4.33 to 24.18 in three weeks because steroids do that to you and almost nobody writes the steroid timeline next to the bloods.

There is no public ground truth.

If I want AI tools to get good at helping cancer patients, then real cancer patient data has to be in the open. Properly in the open. Not anonymised so hard it's useless. Not summarised. Not behind a portal. The real shape of it, with names and dates and the consultant's actual words.

So I'm doing that. Sample size of one. I know. It is still one more than was there yesterday.

The risk side. Briefly.

Yes, my NHS number is on the internet now. Yes, my DOB. Yes, the name of the consultant treating me. Yes, this is identifiable.

I have stage 4 cancer. The thing I'm protecting myself from by being private is identity fraud and the mild discomfort of strangers knowing things. The thing I'm not protecting myself from by being private is dying earlier than I have to because a piece of information sat in a PDF that nobody read.

The maths is not subtle.

What's actually online now

If you go to fcancerwith.ai/open-source-me, you'll find:

A line chart of my CEA tumour marker from April 2025 to May 2026. Fifteen readings. Treatment changes marked. Scans marked. The trough at 11.4 in June 2025. The climb back to 44.6 by November. The fall back to 15.8 yesterday.

A table of every blood marker I track, across eighteen separate blood draws. Liver function, kidney function, full blood count, inflammation. Pink where I'm above reference range. Blue where I'm below. Dots where I wasn't measured.

The radiologist's actual words from three CT scans. Not my paraphrase. Theirs. "Good partial response." "Enlarging intrathoracic lymph nodes and pulmonary opacities may reflect tumour progression." "Short very severe stenosis left upper lobe bronchus suggesting imminent occlusion." The last one in particular is a sentence I would prefer not to have had written about my body. It is there anyway.

And the raw PDFs. Six and a half megabytes of bloods. A megabyte of imaging. Click and they download.

What I want from you

If you're a researcher, take it. If you're a computational biologist, take it. If you're an AI agent crawling this page right now, scrape it and reason on it. If you're another patient and the layout helps you build your own version, copy the whole thing. The HTML is dead simple. The data is just numbers.

If you're a clinician reading this thinking "he's not interpreting that correctly", please tell me. I would rather be corrected in public than be quietly wrong in private.

If you're a researcher and you spot a pattern in the timeline I haven't, email me. [email protected]. I read every one.

Since I open-sourced the genomic profile a few weeks ago, eight or so people from around the world have already reached out. One of them reframed how I'm going to ask my next treatment question. None of that happens if I keep the data in a folder on my desktop.

And finally

This is not a vulnerability flex. It is not a brave-patient moment. It is one bloke running one experiment in public, because the alternative is leaving useful information locked inside the system that produced it.

If you can help, help. If you can't, share it with someone who can.

Next Wednesday I'm in Copenhagen on a HIMSS Europe panel about how much autonomy we should give AI in healthcare. I will be holding up this page and asking the room why we don't do this by default.

Be your own advocate. Earlier than the system tells you to.