After a year of using the lab-decoder prompt on individual panels, I realized I'd been doing something dumb: treating every blood draw like the first one that ever existed. A cholesterol number on its own is just a number. A cholesterol number that's been creeping up 8 points a year for three years is a story. The doctor sees the trend in their system; I only ever saw the snapshot in my portal, reacting to each year fresh like a goldfish discovering its own bowl for the hundredth time.
The thing that finally bugged me into changing was a 'borderline' flag that had been 'borderline' for three years running. Every year I'd see it, mildly panic, get reassured on the call, and forget about it. Nobody — including me — was looking at the fact that 'borderline and stable' and 'borderline and climbing' are completely different situations, and I had no way to tell which one I was in because I never put the years side by side.
The workflow, end to end
It's almost embarrassingly simple, and the Decode My Lab Results Into Plain English Before the Follow-Up Call prompt needs no modification to do it. In the lab_panel variable, instead of pasting one year, I paste two — clearly labeled '2024' and '2025' for each test — and in patient_context I add one line: 'Please compare year over year and note direction of change.' Gemini 3 Pro Preview handles this beautifully because it's a long-context, table-heavy task, which is exactly what Gemini eats for breakfast. The model keeps the rows aligned across years without me having to babysit it, and it adds a column showing the direction each value moved.
- Paste both years into lab_panel with year labels on every single row
- Add 'compare year over year, note direction of change' to patient_context
- Read the 'Worth asking about first' section as a trend list, not a snapshot list
- Save the output so next year you're comparing three points, not two
- Keep last year's PDF in the same folder so future-you isn't hunting for it at 11pm
What the trend view caught that a snapshot missed
The first time I ran it this way, every single one of my values was technically 'in range' — so a snapshot would have said nothing, shrugged, and moved on. But the comparison showed my fasting glucose had moved from 88 to 94 to 99 over three draws. Still in range. Also clearly walking toward the door marked 'prediabetes' with its coat already on. The prompt didn't diagnose that — it's strict about never doing that — but it flagged the upward direction and gave me a clean question: 'My fasting glucose has risen three years running while staying in range; is this a trend worth getting ahead of?'
That is a far better question than 'is 99 okay?' Because '99 okay?' gets you a true-but-useless 'yes, that's within range,' and the conversation ends. The trend question gets you an actual plan. My doctor's answer was 'good catch, let's add an A1C next time and tighten up the evening carbs.' We made a small change a year or two before it would have become a flagged number. That's the entire point of trends — they let you act while the change is still cheap.
The honest caveat
This only works if you actually keep last year's PDF, and I will admit I failed at that for the first two years and had to email the lab to get them resent, which involved a phone tree and the kind of hold music that makes you reconsider your life. And the prompt is only as good as the labels you give it — be obsessive about marking which value is which year, or the model will line them up in a perfectly confident, perfectly wrong way. It won't catch a labeling mistake; it trusts your input completely.
Why 'in range' is not the same as 'no news'
The mental shift this produced is bigger than the prompt itself. I used to treat 'in range' as a green light that ended the conversation, and 'out of range' as a red light that started a panic. The trend view taught me there's a whole third state that the single-snapshot view literally cannot show you: in range but moving. That's the state where the cheap, easy interventions live — a tweak to dinner, a bit more walking, an extra test next year — because the body is drifting but hasn't arrived anywhere yet. Catching the drift is the entire value, and you can only see drift across time.
Grab the Decode My Lab Results Into Plain English Before the Follow-Up Call prompt on Prompt Dock, feed it two years instead of one, and start reading your bloodwork like a chart instead of a jump-scare. I keep mine in the same folder as my Organize My Messy Symptom Notes Into a Journal outputs, so I walk into every physical with both a trend and a story — which is a genuinely powerful combination to hand a doctor who has eleven minutes and a full waiting room.