The sleep test

The test name is polysomnography.

I learned the word at a hospital billing counter in Dhaka. I said I needed a sleep test. The woman paused. It had another name, she said, but she could not remember it.

I searched online and pronounced the word from my phone.

Polysomnography.

She entered it into the billing system. The test appeared.

Before the word, my request was informal. After the word, it became a service the institution could recognize, price and schedule.

Language was the first interface.

Method and limit

This field note uses six materials:

• notes written within twenty-four hours of the test;
• appointment messages and time stamps;
• the fourteen-page diagnosis and titration report;
• remembered conversations from the night and the following consultation;
• subsequent product research and visits to medical-equipment stores in Dhaka and Bangkok;
• twenty-four weeks of Garmin sleep records from 1 February to 18 July 2026.

I later returned to these materials through my unlearning framework. The framework asks what route an interface places first, what reward holds it there, and what artifact carries that order.^[1]

The report contains my own medical information. Other people are described by role because I did not meet them as research participants or obtain consent to publish their identities. Conversations are paraphrased where exact wording is unnecessary.

This is not a clinical evaluation of my treatment. It is an interface audit of how a body, a protocol and a report produced a medically legible sleeper.

The room that became a lab

Before payment, the counter sent me to a respiratory room beside the outpatient consultation area. The room handled availability, preparation, scheduling and report collection. The actual night study took place elsewhere, in a hospital cabin temporarily arranged around the equipment.

The service was new. I was later told that I was its twenty-seventh patient.

Its institutional presence was small:

one room
one recording system
one night technologist
one patient at a time

The hospital was large. The sleep service depended on a narrow human and technical arrangement inside it.

That dependence became visible when my original appointment was cancelled. A relative of the technologist had died. The test had to move to another date and then into an available institutional slot.

The preparation sheet instructed me to wake early and avoid sleeping during the day. I followed it. I was physically and mentally prepared for the original night. When the appointment moved, that preparation became unusable.

On the rescheduled date, I was told to report at 9 pm.

I do not normally sleep at 9 pm. My dinner is also later. I tried to negotiate the arrival time, but the available slot came first. I ate quickly and reached the hospital with my partner and a friend.

By then, the test had already begun shaping the body it expected to measure.

A fresh patient

The cabin resembled a hotel room with medical equipment added to it. I changed into a cotton hospital gown and completed an intake form.

The form was in Bangla.

I asked what would happen if a non-Bangla-speaking patient came for the test. I received no answer.

I had used the same hospital for several years. It already had my patient number, insurance records, consultations and recent test results. During that visit, however, staff repeatedly said that an information-system upgrade had made older records unavailable.

The sleep form therefore began from zero.

The institution knew me as a returning patient and received me as a blank one.

This was not only a data problem. It established the sequence of knowledge. The standard form arrived before the medical history the hospital already held. What fitted the form became immediately available. What remained elsewhere had to be remembered and narrated again.

Producing the measurable sleeper

The technologist attached sensors to my scalp, face, chin, chest, abdomen and legs. Bands measured respiratory effort. Other sensors recorded airflow, oxygen saturation, heart activity, movement, position and snoring.

My partner photographed the wired body. We made jokes about familiar film and internet tropes to reduce the tension.

I was both the subject and a prop inside the room.

The report later described the arrangement in technical language:

Three EEG channels, two EOG channels, one ECG channel, chin EMG, left and right leg EMG, a snore microphone, airflow sensors, respiratory effort belts, pulse oximetry and a body-position sensor.

At 11:51 pm, the recording began.

The report calls this moment “lights out.”

I call it the beginning of trying to sleep.

For what felt like several hours, I moved between positions. Sensors loosened. The technologist returned to adjust them. I felt cold in the hospital gown and requested another blanket. I turned the light on and off. I went to the toilet repeatedly.

The report later stated:

The patient had no episodes of waking to urinate during the night.

My field note records more than five toilet visits.

The difference is small clinically unless those visits affected the recording. Analytically, it matters. My body remembers repeated interruption. The summary remembers none.

What counts as sleep

I told the technologist that I had self-identified ADHD, anxiety and a history of panic attacks. I also explained that I normally use sound to settle into sleep.

She needed sufficient data before the test could move to titration.

I needed conditions under which sleep might occur.

At first, I was told not to use my phone or play sound because audio recording formed part of the study. I asked whether my usual sleep sound could be permitted at low volume. The technologist consulted the supervising doctor and returned with approval.

I played a familiar soundscape. When that did not help, I changed to a downloaded recording of rain with quiet piano.

Eventually, I slept.

This was the most important alteration of the night. The protocol did not disappear. Its sequence changed slightly.

The initial route was:

protocol first
familiar sleep practice removed
body expected to adapt
data expected to follow

The negotiated route became:

protocol
patient describes an unmet condition
technologist seeks authorization
familiar sound enters the room
sleep becomes possible

The adjustment did not invalidate the test. It helped the test continue.

The technologist later told me that I was the first patient who had requested sleep audio. The encounter made that need visible to the service. I learned that a rule presented as fixed could contain room for judgment.

The split night

At 3:45 am, the diagnostic phase ended.

The report recorded 164 minutes of sleep during 234 minutes of diagnostic recording. It identified 79 apnea and hypopnea events, an apnea-hypopnea index of 28.7 per hour, severe snoring and a lowest oxygen saturation of 88 percent.

An AHI between 15 and 30 is conventionally classified as moderate obstructive sleep apnea.^[2]

The diagnosis therefore did not depend on one isolated event. The recording documented repeated obstruction across more than two and a half hours of scored sleep.

The night then changed tasks.

Staff woke me, fitted a nasal mask and started positive airway pressure titration. I initially felt claustrophobic. I was told the sensation should improve once the airflow stabilized.

It did become easier. I slept.

The titration section records nearly three hours of sleep. Across the full titration period, AHI fell from 28.7 to 1 event per hour. Average oxygen saturation increased from 94 to 96 percent. The snore index fell from 469.7 to 40.7.

The central clinical result is difficult to dismiss. The study recorded obstructive sleep apnea, and positive airway pressure greatly reduced the measured respiratory events during that night.

But the report moves from this strong finding to a more specific recommendation without showing every step.

The report as interface

The report did not simply preserve the night. It reorganized it.

The lived sequence contained preparation, rescheduling, unfamiliarity, cold, repeated toilet visits, surveillance, negotiation, claustrophobia and eventual sleep.

The report produced another sequence:

recording
diagnosis
titration
normalization
recommendation

Both sequences are real. Only one enters the formal medical record.

This is sequence authority.

The report arrives first when another clinician, insurer or supplier encounters the night. My field note arrives later, if I remember to introduce it.

The report also contains internal inconsistencies.

Sleep latency

The diagnostic table records sleep latency as zero minutes. I experienced a long struggle to sleep.

This does not necessarily mean the machine was wrong. Sleep onset is scored through physiological signals, not through the subject’s memory of falling asleep. I may have entered brief stage N1 sleep while continuing to experience myself as awake.

The difference reveals two definitions rather than automatically proving an error:

machine sleep: a scored physiological state
lived sleep: the remembered experience of having slept

The report controls the clinical definition.

REM sleep

The diagnostic tables record zero minutes of REM sleep. Another summary sentence says that the lowest oxygen saturation occurred “during an apnea and R sleep.”

Both statements cannot describe the same diagnostic record without further explanation.

The absence of diagnostic REM also matters because obstructive events can vary by sleep stage. A first night in an unfamiliar laboratory can alter sleep continuity and architecture, including REM sleep.^[3]

This does not erase the 79 respiratory events that were recorded. It limits what this single night can say about how my apnea varies across a more ordinary night.

Toilet visits

My notes record repeated toilet visits. The report states that there were none.

Here the machine-readable night does not merely differ from my perception. The written summary excludes observable events that required leaving or interrupting the bed.

Sleep-stage arithmetic

The titration report gives a total sleep time of 177 minutes. It separately lists:

N1, N2 and REM already total 177 minutes. Adding the reported N3 produces 226 minutes.

At least one value or label is wrong.

The report nevertheless describes deep sleep as significantly improved. That conclusion may be correct, but the printed figures do not allow the reader to verify it.

Pressure and comfort

One page says pressure was increased from 4 to 15 cmH₂O and that I felt comfortable at 11. Another says I felt comfortable at 15.

I do not remember selecting either number. I remember initial claustrophobia, gradual adjustment and then sleep.

The report converts successful continuation into a statement about comfort.

Its pressure table creates another question. It records:

The final recommendation says:

Use Auto CPAP with nasal mask at 15 cmH₂O pressure during sleep time regularly.

The table shows effective control across several pressures. It does not show, at least in the printed pages available to me, which pressures included supine REM sleep, acceptable mask leak, stable oxygenation or sustained control long enough to select 15.

Clinical titration guidance does not choose pressure from AHI alone. The selected pressure should control obstructive breathing, maintain adequate oxygenation, have acceptable leak and, for an optimal titration, include sufficient supine REM sleep at the chosen pressure.^[4]

The report may contain enough underlying signal data for a clinician to justify the choice. The printed recommendation does not make that reasoning visible.

Can one night justify one machine?

A single polysomnography night can provide clinically meaningful evidence. My diagnostic phase crossed the conventional threshold for moderate OSA. The titration phase showed a major reduction in respiratory events with positive airway pressure. Split-night studies are an established clinical route when sufficient moderate or severe OSA appears during the diagnostic portion and enough time remains for titration.^[5]

The first-night effect does not make the diagnosis imaginary.

It does, however, place limits around more specific claims.

My report supports these conclusions:

• obstructive respiratory events repeatedly occurred during recorded sleep;
• the measured AHI fell within the moderate OSA range;
• positive airway pressure substantially reduced those events during titration;
• oxygen saturation and snoring improved during the PAP phase.

The printed report does not transparently establish:

• why 15 cmH₂O was selected over the other effective recorded pressures;
• whether 15 is intended as a fixed pressure, an upper limit or part of an unstated automatic range;
• which pressure included adequate supine REM sleep;
• how mask fit, leak and long-term comfort affected the choice;
• why one commercial brand or supplier would be clinically necessary;
• how cost and my ability to continue treatment were weighed.

Recommending PAP after this study is not inherently excessive or unethical. Recommending a specific device may also be reasonable when reliability, data access, servicing or pressure performance differ materially.

The ethical problem begins when specificity exceeds the explanation available to the patient.

A durable recommendation should separate four decisions:

1. Does the evidence support an OSA diagnosis?
2. Does the evidence support PAP treatment?
3. What pressure range and mask are appropriate?
4. Which available device can deliver that treatment safely and sustainably?

My report answers the first two strongly.

It compresses the last two into one line.

Clinical guidance treats PAP as continuing care, not as a one-time hardware purchase. It emphasizes education, mask fitting, troubleshooting, monitoring of objective use and efficacy data, and follow-up after treatment begins.^[6]

A device that is technically excellent but financially unreachable produces no treatment. A cheaper device without trustworthy performance, service or monitoring may produce another risk. Cost is therefore not external to clinical effectiveness. It shapes whether the prescribed route can be entered and maintained.

From pressure to product

The report recommends a category of treatment: automatic positive airway pressure with a nasal mask.

The consultation moved from treatment category to product market.

The consultant explained that good machines were expensive. He warned against low-cost devices and referred me to a local supplier. He did not write a specific brand in the report, but described an international brand as preferable and told a story about a patient allegedly harmed by a cheaper device.

That story changed the decision environment.

The choice was no longer presented as:

different clinically suitable devices
different prices
different service arrangements
different forms of follow-up

It became:

expensive recognized device
or
unsafe uncertain device

I cannot verify the reported death or connect it to a particular product failure. I also cannot establish whether the referral involved any financial relationship. Treating either suspicion as fact would reproduce the same problem I am examining.

The observable issue is narrower.

A story about danger arrived before transparent comparison.

Country of manufacture became a shortcut for reliability. Brand recognition became a shortcut for safety. The supplier card made one purchasing route immediately available while alternatives remained abstract.

The treatment that did not begin

I did not reject positive airway pressure.

I could not convert the prescription into a confident purchase.

I searched the Dhaka market. During a later visit to Bangkok, I visited medical-equipment stores there as well. The same brand repeatedly arrived first. Sellers presented ResMed as the recognized international option and pushed me toward it.

The available machines did not form a simple price list. I encountered new machines, used machines, different models, uncertain warranties and cheaper alternatives whose reliability I could not independently judge.

The report had reduced the clinical decision to one line:

Use Auto CPAP with nasal mask at 15 cmH₂O pressure during sleep time regularly.

The market reopened everything that line had closed.

Which model?
New or used?
What pressure range?
What mask?
Whose warranty?
Who reads the device data?
Who adjusts treatment?
What happens if the machine becomes uncomfortable?

The supplier could sell a machine. The report could prescribe one. Neither interface gave me a clear method for comparing the available products.

The clinical sequence therefore stopped at the point of transaction:

diagnosis
titration
prescription
price
uncertainty
no device

The prescription produced medical certainty and practical hesitation.

The second machine

In early February 2026, I began wearing a Garmin Forerunner 165.

It was not a substitute for CPAP or polysomnography. The watch could not tell me how many times my airway closed. It could not reproduce the EEG, airflow, respiratory-effort or clinical oxygen measurements used in the hospital study. Garmin also states that the device is not intended to diagnose or monitor a medical condition.^[7]

It answered a different question.

The hospital had measured what happened to my breathing during one night. The watch began estimating when and how long I slept across ordinary life.

From 1 February to 18 July 2026, it produced twenty-four weeks of sleep records.

The averages flatten substantial variation. Weekly recorded sleep ranged from 3 hours 14 minutes to 5 hours 34 minutes. Average bedtimes moved across late night, early morning and, in some weeks, daytime.

The watch made one pattern difficult to ignore. Whatever happened to my airway during sleep, I was also recording very little sleep and entering it at unstable times.

But the watch introduced another clean result.

On recorded mornings, a proprietary system converted movement, heart rate, heart-rate variability and other signals into a number. The number often said Poor.

The hospital report had turned my breathing into AHI.

The watch turned my ordinary nights into a sleep score.

Neither number was meaningless. Neither contained the entire problem.

The hospital metric supported a respiratory treatment that I could not yet obtain. The consumer metric repeatedly described inadequate sleep without treating the obstruction already documented by the hospital.

The second machine continued measurement where the first machine’s recommendation had stopped.

The body improvises

From February through July 2026, after the sleep test and while I remained without a PAP device, I made smaller physical adjustments at home.

I began using two pillows. I also reminded myself to sleep on my side.

The two-pillow practice had not been tested during the hospital study. The report’s positional analysis, however, gave me some basis for trying side sleeping. During the diagnostic phase, it recorded an RDI of 46.3 events per hour while supine and 24 while non-supine.

The difference is meaningful, but its limit is equally important. Non-supine sleep was associated with a lower recorded RDI during that single study. It did not bring the result into the normal or mild range.

The Garmin could not establish whether either later practice reduced apnea.

These were interim counter-practices, not equivalent treatment. They changed what I did at home while the prescribed route remained financially and practically unresolved.

Prototype audit

Reflex. Comply with the test conditions, produce sufficient data and then follow the preferred purchasing route.

Reward. Diagnostic certainty, completion of diagnosis and titration in one night, efficient use of a scarce service, and rapid movement from report to treatment product.

Carrier. The billing term, preparation sheet, appointment slot, intake form, sensors, scoring rules, report tables, prescription, supplier card, product market and Garmin sleep score.

Sequence change. Introduce familiar audio during the test. Compare products rather than immediately purchasing the preferred brand. Track ordinary sleep timing and experiment with body position while treatment remains unresolved.

Return. The diagnosis retains clinical authority. The prescribed device remains inaccessible. The Garmin score risks becoming another result that arrives before felt experience.

Capture risk. A healthcare-access problem becomes a personal optimization project. Consumer measurement offers continuous legibility without treating the respiratory obstruction.

What changed

The technologist did not simply enforce the protocol. She interpreted it.

She adjusted sensors, found a warmer blanket, listened to my request, consulted the doctor and allowed an unfamiliar need into a new service. Her judgment made the protocol more workable.

That matters because the case is not a simple story of system against patient.

The interface contained several forms of intelligence:

• physiological measurement;
• clinical scoring;
• the technologist’s situated judgment;
• the patient’s knowledge of his own sleep practice;
• the consultant’s treatment experience;
• the report’s capacity to make the night portable.

The problem emerged when one form of intelligence was allowed to summarize the others without showing the cut.

The machine did not lie. The report did not tell the whole night. My memory does not replace the physiological record. The physiological record does not replace my account of how that record became possible.

Does this count as unlearning?

Partly, at the personal scale.

I did not unlearn obstructive sleep apnea. That is not what the framework proposes. What weakened was the authority of the prescription as a complete route from diagnosis to treatment.

The case now contains repetition. I researched machines in two countries, repeatedly encountered the same preferred brand, wore a consumer tracker for twenty-four weeks, used two pillows and reminded myself to return to side sleeping.

These practices did not establish an equivalent treatment or demonstrate reduced respiratory events. They changed the sequence through which I interpreted and responded to the recommendation.

Before
report
prescription
preferred product
purchase expected

After
report questioned
market compared
cost exposed
daily sleep observed
body position altered
treatment unresolved

A second capture also appeared.

When clinical treatment remained inaccessible, consumer measurement became available. The Garmin score offered daily legibility without respiratory treatment. It risked converting a healthcare gap into a personal optimization project.

For the watch to have sequence authority, however, the number must do more than appear. I would need to document how a low score changed my mood, schedule, behaviour or judgment about whether I had slept well.

The data establishes the carrier.

A future field note must establish its authority.

At the institutional scale, unlearning remains unproven. I do not know whether the service changed its preparation instructions, accommodation practices or reporting process after my encounter. A future patient may still have to negotiate the same exception from zero.

Individual accommodation weakened a default for one night.

Institutional memory determines whether the default learned anything.

After the test

In the morning, the technologist removed the sensors. Gel remained across my scalp and body. I thanked her and apologized for not giving her enough sleep.

The report later showed that I had given the system 164 diagnostic minutes and 177 titration minutes.

Enough to diagnose.

Enough to prescribe.

Enough to turn one difficult night into a portable medical fact.

Before leaving, I photographed the room and the equipment beside the bed. I wanted a memory of the place before the report became the night.

The report says the machine worked.

It did.

The remaining question is what else had to work around it, and what disappeared when the result became clean.

Endnotes