I am the radiation safety officer who walks the linac vault at three forty-five because the machine talks to itself then, and when Varian’s cloud still had an August fourteenth oh three forty-seven trace my physicist deleted from our MOSAIQ — and I unfolded the strip chart hidden under lead in the isotope box — I understood my mentor had traded Margaret Hsu’s skin for a cleared backlog.

I am the radiation safety officer who walks the linac vault at three forty-five because the machine talks to itself then, and when Varian’s cloud still had an August fourteenth oh three forty-seven trace my physicist deleted from our MOSAIQ — and I unfolded the strip chart hidden under lead in the isotope box — I understood my mentor had traded Margaret Hsu’s skin for a cleared backlog.

My name is Rochelle Vause-Adekunle, and Dennis Horvath taught me to respect the beam.

Then he taught the beam to lie.

September ninth, six-twenty in the morning, and I am at the secondary physics workstation in the Riverside Cancer Pavilion.

The MOSAIQ oncology information system is open on the left screen — attachment index for TrueBeam Unit 3, sorted by service event date.

I scroll the August entries.

August twelfth: daily output constancy file, 0.4% above baseline — within tolerance.

August thirteenth: daily output constancy file, 0.6% above baseline — within tolerance.

August fourteenth: blank row.

August fifteenth: blank row.

No attachment.

No event trace.

The scrollbar gap is one pixel wide where two files should be.

On the right screen I have the Varian remote service cloud extract — obtained through a third-party record request to Siemens Healthineers, processed under the hospital’s vendor data access clause.

The vendor cloud shows a complete event trace for August fourteenth — uploaded automatically at 03:47 by the TrueBeam Unit 3 service module during the nightly self-test cycle.

The trace file is 4.2 megabytes.

The MOSAIQ attachment index for that date is empty.

Someone deleted the local copy.

The manufacturer’s cloud kept the receipt.

I check the checksum column on the vendor extract.

The file hash for August fourteenth matches the August thirteenth format — same TrueBeam firmware build, same data structure, same compression header.

This is not a corrupted upload.

This is not a missing file from a system glitch.

This is a file that was present, uploaded, and then removed from the local index.

My mouse wheel clicks through the rows — notch by notch — and the gap stays.

I print the vendor extract.

I print the MOSAIQ attachment index.

I place them side by side on my desk.

The gap is clear.

Two weeks before, on a Tuesday night walkthrough, I had been in the linac vault at 03:45.

I do the late walkthrough once a month — checking the vault door interlock magnets, the emergency stop button function, and the radiation safety signage placement.

The Unit 3 display clock read 03:45 when the machine entered its nightly self-test cycle — the LED ring around the gantry head flickered green, reflected on the epoxy vault floor in a circle of light.

Dennis was in the control room finishing a QA spreadsheet.

He saw me on the vault camera feed.

“She knows the ghosts’ shift,” he said through the intercom.

He meant the machine’s automated routines — the tests that run when no one is supposed to be watching.

03:45 was routine then.

It was not routine anymore.

Dennis Horvath has been the Lead Medical Physicist at Riverside for fourteen years.

He mentored me when I was a junior dosimetry tech — taught me isodose curves, inverse planning, and the three-dimensional geometry of radiation dose distribution.

He said things like “never outrun physics” and meant them.

He ran half-marathons for the children’s oncology charity — his race bib read “DoseRightDennis” in block letters, and he pinned it to the break room corkboard after every race.

The bib from the April marathon was still there — number 4471, fabric frayed at the hem from concrete wear.

Last Thursday he brought donuts to the morning physics meeting.

Powdered sugar.

He slid a decaf across the table to me — he knows I prefer regular, but he buys decaf because he thinks the caffeine gives me tremor during ion chamber readings.

I drank it.

Powdered sugar on his knuckle when he wiped the bib photo on his phone to show the new time.

“Ethics are cardio, Rock,” he said.

He calls me Rock.

He has called me Rock for twelve years.

The nickname used to sound like trust.

After the morning meeting I returned to my office and opened the courier no-show SOP — version three.

I wrote this protocol after the COVID supply chain disruptions in 2021, when isotope deliveries failed three weeks in a row and the hot lab inventory fell below safety minimums.

The SOP requires physical verification of isotope transport box contents within twenty-four hours of any courier no-show — every box gets moved, opened, and logged.

The procedure exists because I wrote it.

Dennis signed the acknowledgment page.

The SOP has a PDF version footer: “RSO-CNS-SOP-003 Rev.3 — Vause-Adekunle — 2021-11-15.”

The footer is a timestamp.

It records the moment logistics became policy.

It is also the reason I will open the lead-lined isotope box next month and find what Dennis hid there.

I saved the vendor extract PDF to an encrypted folder on my personal drive — a backup separate from the MOSAIQ system.

I documented the checksum comparison in my RSO incident log — date, time, file names, hash values.

I did not page Dennis.

I closed the MOSAIQ screen.

The gap in the attachment index stayed on the printout beside my keyboard.

Remote remembers what local forgot.

When I was twenty-six, Dennis sat beside me at the treatment planning console and drew isodose curves on a yellow legal pad.

He drew them freehand — smooth ellipses that matched the calculated distribution within a pencil width.

“The beam goes where physics says,” he told me.

“Never outrun physics.”

He kept a pencil behind his ear — a mechanical Pentel 0.5mm — and the clip left a small dent in the cartilage above his right ear.

I still have a matching dent.

Twelve years of pencils behind the ear will do that.

He taught me to read dose-volume histograms the way a musician reads sheet music — structure first, then nuance.

He taught me that a linac is not a weapon.

It is a tool that requires respect, calibration, and honesty.

Those were his words.

Honesty.

At night, when Zara was asleep on the couch with her school lunchbox beside her — sticker corners peeling from the lid — Kofi sat across the kitchen table and quizzed me on CHP flashcards.

Exposure algebra.

Shielding calculations.

Half-value layers.

The index cards were softened at the corners from months of handling — the edges feathered where my thumb had rubbed them during study breaks.

Kofi is a firefighter.

He understands protocols that keep people alive.

He did not understand inverse square law on the first try, but he read the cards and checked my answers and refilled my coffee at ten-thirty without being asked.

Zara’s breathing was steady on the couch — six years old, dreaming of whatever six-year-olds dream of, which is not radiation physics.

The CHP certification was not just a credential.

It was a promise that I would know enough to catch what others miss.

I attended the burn clinic ethics consult as the RSO observer — a mandatory attendance for any radiation-related adverse event reviewed by the ethics committee.

The consult was for Margaret Hsu.

I walked down the hallway toward the treatment room.

The door was partially open.

Through the gap I saw the bulk of a wound dressing on Margaret’s torso — moist desquamation, grade three, on skin that should not have received the dose it received.

I looked away.

I fixed my eyes on the fire code poster across the hall — red EXIT letters that vibrated slightly from the HVAC airflow.

I counted to ten.

I did not stare.

Margaret Hsu is sixty-seven years old.

She came to Riverside for SBRT liver treatment — stereotactic body radiation therapy, five fractions, high dose per fraction, tight margins.

Her treatment was scheduled on Unit 3 during the week of August 12-16 — the same week the daily output constancy showed drift that someone deleted from MOSAIQ.

The burn clinic pathology consult noted tissue injury consistent with hot delivery — dose exceeding the prescribed target by a margin that should have triggered a machine hold.

Margaret healed.

She healed slowly, with scars that her compression sleeve covers at church.

The ethics committee reviewed the case in a conference room with a projector that hummed at a frequency I could feel in my sternum.

Dennis presented the physics QA records — the sanitized version, the version with the MOSAIQ attachment index that showed tolerance values on August 12 and 13 and a blank row on the 14th that he attributed to a “system maintenance window.”

He spoke calmly.

He used the phrase “within clinical expectations.”

He showed the committee a dose histogram that did not include the vendor’s raw data.

The records he presented did not include the August 14 trace.

The committee found no actionable deviation.

No one asked about the blank row.

No one asked why the file size for August 14 was zero when every other day showed 4.2 megabytes.

The committee did not have the vendor cloud data.

I did.

I requested the Varian remote service cloud extract formally — a third-party record request processed through the hospital’s vendor liaison office.

The cloud showed three consecutive days of output overshoot: August 12, 2.4% above baseline; August 13, 2.7%; August 14, 3.1%.

The institutional action threshold is 2.0%.

All three days exceeded it.

The MOSAIQ records for those dates showed values within tolerance — 0.4%, 0.6%, and a blank row for the fourteenth.

Dennis had truncated the machine’s internal service event export before it reached the local attachment store.

The vendor’s cloud had uploaded the full trace automatically at 03:47 on August 14 — within the 72-hour retention window.

Dennis deleted the local copy.

He did not know the manufacturer kept a mirror.

I also requested Dennis Horvath’s hiring file from HR — the original interview panel notes, lawfully accessible to the RSO under the hospital’s credential review policy for safety-critical roles.

The PDF arrived in my secure inbox.

Page six.

In the margin, in Dennis’s handwriting, initialed DH: “Never override vendor safety interlocks — policy and personal commitment.”

The initials were slightly smudged — blue ink on cream paper.

His own words from the day he was hired.

His own words from before he overrode them.

Dr. Yasmin Ali-Borges is a chief medical physicist at a university hospital two hundred miles east.

I contacted her through the American Association of Physicists in Medicine peer review channel and requested an independent recalculation of the Unit 3 output for the week of August 12-16.

She reviewed the vendor trace data via encrypted screen share.

She flagged the delivery hot fractions on Margaret Hsu’s SBRT plan IDs.

The DVH curve — dose-volume histogram — showed a bulge at the target boundary that corresponded to a 3.1% overshoot on the August 14 fraction.

Her cursor halo circled the inflection point on the curve.

“That’s not variance,” she said.

“That’s a burn.”

She wrote a memo — four pages, peer-reviewed methodology, signed and dated.

I encrypted it and stored it with the vendor extract.

I printed the Varian trace PDF — all three days, timestamped, checksummed.

I placed the printout on the lead brick beside the hot lab sink.

I washed my hands.

Thirty seconds.

The water was too hot — I did not adjust the faucet.

The soap dispenser clicked twice by mistake — I pressed it harder than I needed to.

My hands were shaking.

Not from caffeine.

I dried them on the paper towel.

I looked at the printout on the lead brick.

The timestamps were clear — 03:47, August 14, Unit 3.

The checksum matched the firmware build.

The output reading showed 3.1% above baseline.

I did not page Dennis.

I picked up the printout and placed it in my locked drawer — the same drawer where I keep the RSO incident log and the encrypted USB backup.

I walked back to my desk.

My pocket dosimeter badge clip scratched against the desk edge as I sat down — a small sound, the kind of sound that measures a career in repetitions.

Dennis believes that clinical judgment includes acceptable variance windows.

He believes that clearing a backlog of SBRT liver cases during a busy week saved more lives than it risked.

He believes that deleting a local file is hygiene — cleaning up a system artifact that would cause unnecessary alarm.

He does not believe he committed fraud.

He believes he made a clinical decision under pressure and that the patient outcome — Margaret’s slow healing — is within the range of expected complications.

He is wrong.

Variance does not delete vendor traces.

Clinical judgment does not erase checksums.

And Margaret’s skin is not a complication.

It is a cost.

Dennis emailed me on a Friday afternoon.

Subject: “Joint Commission Binder Prep — Weekend Assignment.”

The email assigned me to compile the radiation safety documentation binder for the upcoming JCAHO survey — a weekend marathon of printing, tabbing, and hole-punching that would occupy Saturday and Sunday.

The binder rings clacked in the attachment preview icon — a fourteen-section document tree with cross-references to every QA protocol in the department.

The assignment conflicted with my scheduled Monday call with the vendor’s legal department to discuss the cloud data retention timeline.

He was burying me in compliance paperwork to keep me away from the evidence trail.

Busywork as a siege engine.

I replied Monday morning.

I told him I had completed sections one through eight remotely and would finish the remainder during regular hours.

I did not mention the vendor call.

I did not CC anyone.

On Tuesday a courier failed to deliver the monthly iodine-131 capsule shipment for the nuclear medicine department.

The courier logged a vehicle breakdown on Interstate 64.

Under my SOP — RSO-CNS-SOP-003 Rev.3 — a courier no-show triggers physical verification of all isotope transport containers within twenty-four hours.

Every box gets moved, opened, contents compared to the inventory log.

I went to the hot lab.

The hot lab cold room holds six lead-lined transport boxes on stainless steel shelving — each one labeled with isotope type, activity level, and last inventory date.

Box four was an isotope marker transport container — lead shielding walls, yellow caution tape on the lid, a latching mechanism rated for Type A packaging.

I unlatched it.

The lead insert — a rectangular block designed to hold marker vials in foam cavities — shifted when I lifted it.

Lead is heavy.

The block groaned against the stainless bench as I slid it forward.

Underneath the insert, folded twice, was a strip chart.

Printer paper — the thermal kind that curls at the edges.

Dated August 14.

Dennis’s handwriting — tick marks beside the output readings.

The strip chart showed the true dose monitor readings for Unit 3 on August 14: 103.1% of baseline at the first control point, 103.2% at the second, 103.0% at the third.

Three readings above the 2.0% action threshold.

The chart that MOSAIQ no longer held.

The chart that the vendor cloud had confirmed.

Dennis had slid it under the lead insert during a Joint Commission mock survey — he assumed physics staff never moves isotope boxes between surveys.

He did not account for the courier no-show SOP.

He did not account for the logistics protocol I wrote after a pandemic broke the supply chain.

Shielding used to shield truth.

I photographed the strip chart in situ — under the lead insert, beside the yellow caution tape, with the date and tick marks visible.

I called the hot lab tech as witness.

I logged the discovery in the RSO incident report — date, time, box number, contents, chain-of-custody form signed by both of us.

The lead block stack in the cold room should have been eighteen.

It was seventeen.

The missing block was the one Dennis had moved to create the hiding space.

I noted the count in the inventory log.

That night the vault security system triggered a false alarm test — a scheduled event that required after-hours access.

I badged in at 03:38.

The vault door interlock released with a three-millimeter sensor gap — the maintenance tolerance documented in the vendor manual.

I stepped inside.

The linac was in self-test mode.

The clock on the Unit 3 display read 03:45.

The interlock LED glowed green — steady, not blinking.

The machine cycled through its automated routines — gantry rotation check, jaw position verification, dose monitor zero calibration.

I stood alone in the vault.

The oxygen monitor on the wall read twenty-one percent — baseline atmospheric.

The fire extinguisher inspection tag showed the last check: December, hole punched.

I imagined August 14 at 03:47 — the trace uploading to the vendor cloud, the drift numbers flagging above threshold, Dennis’s cursor deleting the local copy before the morning therapists arrived.

A red second ghost overlaying the minute hand.

The machine was innocent.

The human was not.

03:45 was no longer a folklore hour.

It was the seam where the truth uploaded and the lie began.

I left the vault at 03:52.

The interlock closed behind me — five clicks in ascending pitch as the magnetic locks engaged.

At seven-oh-five the next morning I walked to the Chief Medical Officer’s office.

The drop slot beside her door accepted sealed envelopes outside business hours.

I inserted the packet — vendor trace printout, strip chart photographs, MOSAIQ attachment index comparison, HR interview notes with Dennis’s “never override interlocks” margin note, and Dr. Ali-Borges’s independent recalculation memo.

I retained a notarized personal copy.

The envelope’s wax disk cracked at the edge when it bent through the slot.

I then completed the FDA MedWatch device malfunction report — an online form with narrative fields for the event description, device identification, and patient outcome.

I wrote in factual tone.

I described the output drift, the deleted records, the vendor cloud retention, and the strip chart.

I attached the PDFs.

The submission receipt number appeared on my phone screen — a green glow on the glass.

I screenshotted it.

Federal echo.

The hospital boardroom had blinds drawn on every window.

The projector was on — a blue standby screen casting light across the mahogany table.

Six chairs on one side — the CMO at center, the chief of medical staff, the hospital counsel, the compliance officer, a board member who was a retired nurse, and the malpractice carrier’s observer.

The observer’s pen sat on the table, capped.

It stayed capped for the entire session.

Dennis sat on the opposite side with his attorney — a healthcare defense specialist from Richmond whose briefcase was open but whose hands were folded.

Dennis wore a tie I had not seen before — dark blue, no pattern.

His race bib was not pinned to his jacket.

I sat at the presentation station beside the projector.

My exhibit binder was open — eight tabs, each corresponding to a slide in the presentation, each backed by a notarized copy of the original document.

Dr. Ali-Borges was on speaker — her voice slightly compressed by the hospital’s encrypted teleconference line.

The boardroom chair hissed pneumatically when I sat down — a sound that punctuated the silence before the CMO spoke.

My shoes squeaked on the waxed tile as I approached the podium.

The blinds cast a single stripe of sun across the table — it moved slowly toward Dennis’s side as the session progressed.

The CMO opened the session.

“This is a closed board inquiry into TrueBeam Unit 3 output deviation during the week of August 12-16, 2025,” she said.

“Ms. Vause-Adekunle, as the Radiation Safety Officer and reporting party, please present.”

Her cufflinks — city seal embossed — clinked against the table when she set her hands down.

I advanced the projector to the first slide — the MOSAIQ attachment index beside the Varian cloud extract.

“MOSAIQ shows no event trace for Unit 3 on August 14,” I said.

“The Varian remote service cloud shows a complete trace uploaded automatically at 03:47.”

“The file is 4.2 megabytes.”

“The checksum matches the firmware build.”

“The trace was present in the vendor cloud within the 72-hour retention window.”

“It was absent from the local MOSAIQ attachment store.”

“The linac logged heat.”

“MOSAIQ lost the file.”

“Varian kept the receipt.”

I advanced to the second slide — the three consecutive days of output overshoot.

“August 12: 2.4% above baseline.”

“August 13: 2.7%.”

“August 14: 3.1%.”

“The institutional action threshold is 2.0%.”

“All three days exceeded it.”

“The MOSAIQ records for those dates show values within tolerance — 0.4% and 0.6% for the 12th and 13th, and a blank row for the 14th.”

“The local records were truncated before they reached the attachment store.”

“The MOSAIQ downtime maintenance window is scheduled for Sunday at two a.m.”

“The deletion event on the vendor’s log corresponds to a Monday morning session — outside the maintenance window.”

“This was not automated cleanup.”

“This was a manual deletion performed during work hours.”

I advanced to the third slide — the strip chart photograph.

“This strip chart was recovered from isotope transport box four in the hot lab cold room,” I said.

“It was folded under the lead shielding insert.”

“It is dated August 14, in Dr. Horvath’s handwriting, with tick marks beside the output readings.”

“The readings match the vendor cloud data — 103.1%, 103.2%, 103.0%.”

“The chart confirms that Dr. Horvath had access to the true output readings and chose not to report them.”

I advanced to the fourth slide — Dennis’s hiring file.

“This is page six of Dr. Horvath’s interview panel notes from 2011,” I said.

“In the margin, initialed DH, he wrote: ‘Never override vendor safety interlocks — policy and personal commitment.'”

“His own words.”

“Fourteen years ago he made that commitment in front of a hiring panel.”

“The strip chart under the lead insert shows he broke it.”

The CMO leaned forward.

“Ms. Vause-Adekunle, how was the strip chart discovered?”

“Courier no-show protocol,” I said.

“RSO-CNS-SOP-003 Rev.3 — the standard I authored in 2021 requires physical verification of all isotope transport containers within twenty-four hours of a courier failure.”

“I moved box four during the inventory.”

“The strip chart was folded under the lead shielding insert.”

“Dr. Horvath placed it there during a Joint Commission mock survey — he assumed physics staff never moves isotope boxes between surveys.”

“He did not account for the courier no-show SOP that he signed on page nine.”

Dr. Ali-Borges spoke through the speaker.

“I performed an independent recalculation of the Unit 3 output for the Margaret Hsu SBRT plan fractions delivered during August 12-16,” she said.

“The dose-volume histogram shows a delivery overshoot consistent with the vendor trace data — the target boundary DVH curve shows a bulge at the 3.1% level on the August 14 fraction.”

“The clinical outcome — moist desquamation exceeding expected grade — is consistent with hot delivery at this magnitude.”

“This is not variance within clinical expectation.”

“This is a measurable overdose.”

Dennis’s attorney spoke.

“Dr. Horvath exercised clinical judgment regarding acceptable variance during a period of high case volume,” he said.

“The SBRT backlog included time-sensitive liver cases.”

“The output readings, while above the action threshold, were within the range that clinical literature considers manageable.”

“Clinical judgment does not delete vendor traces,” I said.

“The MOSAIQ attachment was removed manually.”

“The strip chart was hidden under lead shielding in a transport box.”

“These are not judgment calls.”

“These are concealment acts.”

Dennis spoke.

“The backlog saved lives too,” he said.

His voice was steady.

“Every day we delayed a liver fraction was a day the tumor grew.”

“I made a call.”

“I have made that call for fourteen years.”

“Margaret Hsu’s skin paid the interest on that call,” I said.

“Three point one percent above baseline on a five-fraction SBRT plan is not a rounding error.”

“It is tissue injury.”

“The burn clinic documented it.”

“The pathology consult confirmed it.”

“You cleared a backlog by hiding the data that would have stopped the machine.”

“The department was under pressure,” Dennis said.

“Every physicist in this room knows what throughput demands look like.”

“I have run Unit 3 for fourteen years without a single reportable incident.”

“You have run Unit 3 for fourteen years without a single reported incident,” I said.

“That is not the same thing.”

“The strip chart suggests this was not the first time output exceeded threshold.”

“It was the first time someone kept the receipt.”

Dennis looked at his attorney.

His attorney looked at the table.

“Dr. Horvath,” the CMO said.

“Your hospital privileges are suspended effective immediately.”

“You are placed on administrative leave pending review by the Virginia Department of Health Office of Licensure and Certification.”

“Your credentials for QA sign-off authority are revoked.”

“The malpractice carrier has been notified and a reserve has been triggered.”

The boardroom was quiet.

The projector hummed.

The malpractice carrier’s observer uncapped his pen for the first time — wrote a single line — and recapped it.

Dennis stood.

He did not look at me.

“I taught you curves,” he said.

He picked up his briefcase.

He left the race bib on the chair — “DoseRightDennis” in block letters, number 4471, fabric frayed at the hem.

I photographed it for the evidence chain continuation log.

The board member — the retired nurse — removed her glasses and cleaned the lens fog with a cloth from her jacket pocket.

She did not speak.

She put the glasses back on and looked at the empty chair.

The blinds stayed closed.

The projector cycled back to the blue standby screen.

A water cooler in the corner bubbled once in the silence after Dennis left the room.

The new chief medical physicist started in November — Dr. Ramirez, recruited from Johns Hopkins, quiet voice, no race bibs.

She reviewed every QA protocol in the department during her first month and signed off on the updated output constancy procedure I had drafted.

She did not mention Dennis by name.

She did not need to.

The procedures said what the conversation could not.

Margaret Hsu came back to the burn clinic for a follow-up in December.

I was not her clinician.

I was not in the room.

I saw her in the hallway — walking slowly, a compression sleeve on her left arm, floral blue pattern.

She was with her daughter.

They were talking about Christmas.

Margaret did not know my name.

She did not know that the RSO who filed the complaint was the same person who had stood outside her treatment room door and counted to ten while looking at a fire code poster.

She walked past me.

The hand sanitizer dispenser beside the elevator smelled like coconut.

I pressed it once and kept walking.

The department metrics were down.

Throughput on Unit 3 had dropped seventeen percent since the board session — partly because the new QA protocol added fifteen minutes to the morning setup, partly because the institutional review had flagged every fraction delivered during the August week for retrospective audit.

The radiation oncology department newsletter — an internal email chain that Dennis used to run — no longer included me in the distribution.

I was not invited to the physicist social thread.

A colleague in dosimetry told me privately that some staff called what I did “hurting the brand.”

I filed that in the same mental category as the courier no-show SOP — things I do because they are correct, not because they are popular.

Kofi came home from a twenty-four-hour shift with mud on his boots.

He left them at the foyer mat.

Zara had drawn a picture at school — crayon, orange, a shape that looked like a machine with a beam coming out of it.

“That’s Mommy’s work thing,” she said.

She meant the linac.

She had never seen it.

She imagined it as orange.

I put the drawing on the refrigerator beside Kofi’s shift calendar — magnets arranged in a row, each one marking a day he would be at the firehouse and I would be at the hospital and Zara would be at school.

Three schedules on one refrigerator door.

The VDH investigation proceeded through the winter.

I gave my deposition in January — two hours in a state office building with fluorescent lighting and a court reporter whose typing speed was faster than my speech.

The investigator asked about the strip chart.

I told her where I found it, how I found it, and why I was in the hot lab that day.

“Courier no-show,” I said.

“SOP requires inventory verification within twenty-four hours.”

She wrote it down.

She asked if I had any personal grievance against Dr. Horvath.

I told her he taught me isodose curves and bought me decaf I didn’t like.

She did not smile.

She wrote that down too.

Three months later — a Tuesday in March — I badged into the vault at 03:40.

The new chief physicist had approved my request to perform a voluntary manual constancy check during the nightly self-test window.

No one asked me to do this.

The schedule did not require it.

I wanted to see 03:45 without imagining a deleted file.

I set up the ion chamber on the treatment couch — a Farmer-type chamber, calibrated, with the electrometer connected and the bias voltage verified.

I positioned it at isocenter — the intersection point where the beam converges — using the room lasers for alignment.

Green dots on white plastic.

The stopwatch was a Casio — digital, gray, the kind Kofi uses for training drills at the firehouse.

I pressed the start button.

At 03:45 the linac entered self-test.

The gantry rotated to zero.

The jaws opened to the reference field size.

The dose monitor calibrated to zero.

The machine delivered a reference beam — ten seconds, six megavolts, the standard energy for the constancy protocol.

The ion chamber reading appeared on the electrometer display.

I compared it to the gold standard baseline.

0.3% difference.

Within tolerance.

Within honesty.

I logged the reading in the new QA binder — the one with the updated cover sheet that listed the protocol revision number and my name as author.

I signed the entry.

I clicked the stopwatch off — three-button factory sequence, start-stop-reset.

The LCD numerals froze at 03:45:00.

The vault was quiet.

The interlock LED glowed green — steady, clean.

The oxygen monitor read twenty-one percent.

The fire extinguisher tag still showed December.

Nothing had changed about the room.

Everything had changed about the hour.

I packed the ion chamber into its foam-lined case.

I closed the vault door behind me.

The interlock engaged — five clicks, ascending pitch.

At three forty-five the vault light turns green on the first try again.