I Believed a Diagnosis for Fifteen Years. Then I Tested It.

For about fifteen years I believed I was hypoglycaemic.

It was not a casual belief. It explained a cluster of symptoms I genuinely had: the shakiness, the irritability that arrived out of nowhere, the sense of needing to eat right now. I had built habits around it. I carried glucose powder. I ate on a schedule designed around it. When the symptoms came, I treated them, and the treatment worked, which confirmed the whole thing every single time.

In fifteen years I never tested it once.

Then, at thirty-three, I put a continuous glucose monitor on my arm and ate normally for a fortnight.

Where to start, depending on who you are

This page is a case study in testing a belief, and the belief happens to be about glucose. Read it for whichever part is yours.

Nothing here is age-specific. The habit of treating an old explanation as a settled fact does not have an age bracket.

What the numbers said

Lived Eleven days of continuous readings, eating normally, travelling for part of it, logging every symptomatic episode as it happened.

Hypoglycaemia has a definition, and it is not “feeling like this.” Researched · Whipple’s triad, the standard clinical criterion The clinical standard is Whipple’s triad, and it requires all three: symptoms consistent with low glucose, a documented low reading at the time of the symptoms, and relief when glucose is raised.

I had one and three. I had never had two. Two out of three is not a diagnosis, and eleven days of measurement, including during the symptomatic episodes themselves, produced not one reading below the threshold. My lowest ever was 81, comfortably normal.

Lived There is also a cross-check that matters: my long-term blood-sugar average, measured independently by a lab, predicted almost exactly the average the sensor recorded. Two unrelated methods agreeing means the sensor was not producing a flattering artefact. The result is real.

The fallback theory, which was also wrong

When the absolute numbers came back clean, I did what everybody does: I moved the goalposts. Perhaps it was not the level but the speed. A fast drop, even within the normal range, causing the symptoms.

It is a reasonable hypothesis. It is also testable, and the data killed it immediately.

The single steepest glucose drop in the entire dataset, a rise to 143 and back down to 82 inside fifty minutes, produced no symptoms whatsoever. I felt completely fine and did not notice a thing.

Meanwhile, a day with a gentler decline to a nearly identical low point was strongly symptomatic.

Same destination. Opposite experience. The difference was not in my blood.

What actually predicts an episode

Once glucose was ruled out twice, the pattern in the log was not subtle. The symptomatic days had four things in common, and the asymptomatic days did not:

1. How long I had gone without eating
2. How badly I had slept
3. How dehydrated I was
4. How much stress I was carrying

Stack two or three of those and I get an episode. Stack none, and a sixty-one-point glucose drop goes completely unnoticed.

The symptoms were always real. Nothing here says it was in my head. What it says is that I had attached real symptoms to the wrong cause for fifteen years, and every one of the four actual drivers is something I can influence, which the imaginary one was not.

The loop: why the belief survived so long

This is the part I find most interesting, and the reason this page exists.

A false belief about your body does not survive by being unchallenged. It survives because it generates its own evidence.

Here is the loop I was running. I would feel an episode coming. I would take fast sugar, glucose powder or a sweet drink. I would feel better twenty minutes later. Confirmed again.

Except: at a normal reading of, say, 88, taking fast sugar produces a sharp rise and then a compensating dip, a mild undershoot an hour or so later. Which feels like the thing I was treating. So I would treat it again.

The treatment was manufacturing the next episode, and the next episode was the evidence for the treatment. Fifteen years of that is not stupidity; it is a closed loop with no exit built into it. The only thing that opens it is a measurement from outside.

I found a second, smaller version of the same shape. I had also believed that needing to urinate frequently was a signal of low blood sugar. The physiology does not permit it; that mechanism requires glucose above about 180, roughly twice what I was claiming to detect. When I checked the log, I found three of those episodes at readings of 153, 105 and 85. Same sensation, three completely different states.

The mechanism was reverse causation: notice the sensation → open the app → whatever number is on the screen becomes the explanation. The app supplied a number every single time, so the belief was confirmed every single time. It would have gone on being confirmed forever.

Then I took the sensor off

This is the part I would most like people to copy.

I had answered the question, so I stopped measuring. Glucose was a question, not a lever. Continuing to wear the monitor would have converted an investigation into a hobby: the accumulation of data I had no intention of acting on, which feels like diligence and is actually just collecting.

Instrument, answer, retire.

I will also say what I did not do. I did not build software around it. The obvious next move for someone who builds things was a pipeline: pull the readings automatically, store them, chart them, watch them forever. I would have enjoyed that a great deal, and it would have been a displacement activity dressed as rigour. The finding was already in.

What I would tell you before you buy one

Continuous glucose monitors are now easy to buy without a prescription in most places, and the marketing has moved well ahead of the evidence.

They are excellent for one thing: answering a specific question about you that you cannot answer any other way. Mine answered a fifteen-year-old question in eleven days, and it was worth every rupee.

They are poor at what most people use them for, which is optimisation: chasing a flatter line, treating every post-meal rise as damage. If your glucose regulation is normal, a rise after eating is not a problem; it is the system working.

The numbers usually quoted (stay between 70 and 180) come from the international consensus on time-in-range, and it is worth knowing that those targets were written for managing diabetes, not as a description of a healthy person. Researched · International Consensus on Time in Range, Diabetes Care 2019 For what a healthy non-diabetic actually looks like, the better reference is a study that put monitors on healthy people: they spent a median of about 96% of the day between 70 and 140, with average glucose around 99, and variability averaging about 17%. ⚠ So do not read a diabetes target as a health goal, and do not read a healthy-population average as something to hit. Between those bounds you are mostly watching normal physiology and, if you are anything like me, inventing a rule about it.

Three things I learned that are worth knowing:

• Liquid sugar spikes fastest, because nothing slows it down. Fine alongside food or after training; a different thing entirely on an empty stomach.
• Carbohydrates stacked forty-five minutes apart compound. My one high reading came from a late dinner followed by dessert. Either alone would have been unremarkable.
• Eating protein and vegetables before carbohydrates, and a short walk after a meal, both blunt the rise, and both are genuinely evidenced. Researched · Shukla et al., Diabetes Care 2015 · postprandial-walking meta-analysis, Sports Medicine 2022 Food order produced a substantially lower glucose response when carbohydrate came last; even a two-to-ten-minute walk after eating beat sitting, and sooner beat later. ⚠ The food-order studies are small (around 11–15 people) and were done in type 2 diabetes and prediabetes, so the size of the effect in a healthy person is unknown. Both are free and mildly useful, and both are far smaller levers than eating enough and sleeping properly. It is very easy to spend your attention on the small lever because optimising is more fun than sleeping.

What still isn’t answered

I want to be careful not to tidy this up more than the evidence allows.

There is a category of much more severe episode in my history (faintness, drenching sweat, extreme irritability) and nothing of that severity happened during the eleven days I was measuring. So it is not explained by this investigation. It is not hypoglycaemia, because the data rules that out. What it is remains an open question I am carrying to a doctor rather than closing myself.

That is the honest end of the story. The investigation answered exactly the question it was designed to answer, and it did not answer a neighbouring one that I would also like answered. Resisting the urge to stretch a clean finding over an untested question is most of what makes the clean finding worth anything.

What this is really about

I was confident, I was consistent, I was wrong, and I could have found out at any point in fifteen years for the price of a fortnight’s attention.

The most valuable thing you can do with a measuring device is point it at something you are already sure about. Testing what you doubt confirms nothing much. Testing what you know is where the fifteen-year errors are hiding, and everybody has some.

The evidence behind this page

Most of this page is n=1, my own eleven days of data, and that is the point of it rather than a weakness, because the question being tested was about me. But several general claims are load-bearing, so here they are.

⚠ The honest summary: this page is a method, not a study. What generalises is test the belief, not my belief was false so yours is too.

Cheat sheet for testing a belief about your body (🔧 method, not evidence)

• If you cannot name the measurement that would prove you wrong, it is a belief, not a finding.
• Check whether your treatment could be producing the symptom it treats. Closed loops are common.
• Log the context alongside the number: sleep, food timing, hydration, stress. The answer is often there and not in the reading.
• Test the thing you are most certain about, not the thing you doubt.
• When the question is answered, stop measuring.
• Do not stretch the finding over the next question. Leave it open and say so.

One-page version of the whole health series: the health cheat sheet.

Message me

If you are carrying a self-diagnosis you have never actually tested, and most people are, I would genuinely like to hear what it is. Mine took fifteen years and a fortnight’s data to overturn, and I am fairly sure it is not the only one I have.

Email: abhishek@terrak.ai LinkedIn: Abhishek Chaudhry

What’s changed

• 31 August 2026. Corrected a framing error: the 70–180 range I quoted is the diabetes time-in-range target, not a description of a healthy person, and the page now gives the healthy-population reference instead. Sourced the food-order and post-meal-walk claims, with their limits, and added an evidence table making clear which parts are n=1.
• 30 August 2026. Added provenance labels naming the source behind each claim.
• 30 August 2026. First published.